Merge pull request #11554 from github/siaramist/codeql-template

Update intro tutorial to include Codespaces CodeQL template

.github/ISSUE_TEMPLATE/lgtm-com---false-positive.md

@@ -1,24 +0,0 @@
----
-name: LGTM.com - false positive
-about: Tell us about an alert that shouldn't be reported
-title: LGTM.com - false positive
-labels: false-positive
-assignees: ''
-
----
-
-**Description of the false positive**
-
-<!-- Please explain briefly why you think it shouldn't be included. -->
-
-**URL to the alert on the project page on LGTM.com**
-
-<!--
-1. Open the project on LGTM.com.
-For example, https://lgtm.com/projects/g/pallets/click/.
-2. Switch to the `Alerts` tab. For example, https://lgtm.com/projects/g/pallets/click/alerts/.
-3. Scroll to the alert that you would like to report.
-4. Click on the right most icon `View this alert within the complete file`.
-5. A new browser tab opens. Copy and paste the page URL here.
-For example, https://lgtm.com/projects/g/pallets/click/snapshot/719fb7d8322b0767cdd1e5903ba3eb3233ba8dd5/files/click/_winconsole.py#xa08d213ab3289f87:1.
--->

.github/ISSUE_TEMPLATE/ql---general.md

@@ -10,5 +10,5 @@ assignees: ''
 **Description of the issue**
 
 <!-- Please explain briefly what is the problem.
-If it is about an LGTM project, please include its URL.-->
+If it is about a GitHub project, please include its URL. -->
 

.github/ISSUE_TEMPLATE/ql--false-positive.md

@@ -0,0 +1,36 @@
+---
+name: CodeQL False positive
+about: Report CodeQL alerts that you think should not have been detected (not applicable, not exploitable, etc.)
+title: False positive
+labels: false-positive
+assignees: ''
+
+---
+
+**Description of the false positive**
+
+<!-- Please explain briefly why you think it shouldn't be included. -->
+
+**Code samples or links to source code**
+
+<!--
+For open source code: file links with line numbers on GitHub, for example:
+https://github.com/github/codeql/blob/dc440aaee6695deb0d9676b87e06ea984e1b4ae5/javascript/ql/test/query-tests/Security/CWE-078/CommandInjection/exec-sh2.js#L10
+
+For closed source code: (redacted) code samples that illustrate the problem, for example:
+
+```
+function execSh(command, options) {
+    return cp.spawn(getShell(), ["-c", command], options) // <- command line injection
+};
+```
+-->
+
+**URL to the alert on GitHub code scanning (optional)**
+
+<!--
+1. Open the project on GitHub.com.
+2. Switch to the `Security` tab.
+3. Browse to the alert that you would like to report.
+4. Copy and paste the page URL here.
+-->

.github/actions/fetch-codeql/action.yml

@@ -1,14 +1,22 @@
 name: Fetch CodeQL
 description: Fetches the latest version of CodeQL
+
+inputs:
+  channel:
+    description: 'The CodeQL channel to use'
+    required: false
+    default: 'nightly'
+
 runs:
   using: composite
   steps:
     - name: Fetch CodeQL
       shell: bash
-      run: |
-        gh extension install github/gh-codeql
-        gh codeql set-channel nightly
-        gh codeql version
-        gh codeql version --format=json | jq -r .unpackedLocation >> "${GITHUB_PATH}"
       env:
         GITHUB_TOKEN: ${{ github.token }}
+        CHANNEL: ${{ inputs.channel }}
+      run: |
+        gh extension install github/gh-codeql
+        gh codeql set-channel "$CHANNEL"
+        gh codeql version
+        gh codeql version --format=json | jq -r .unpackedLocation >> "${GITHUB_PATH}"

.github/labeler.yml

@@ -51,3 +51,6 @@ documentation:
   - "java/ql/lib/semmle/code/java/dataflow/internal/tainttracking1/TaintTrackingImpl.qll"
   - "java/ql/lib/semmle/code/java/dataflow/internal/DataFlowImplConsistency.qll"
   - "java/ql/lib/semmle/code/java/dataflow/internal/FlowSummaryImpl.qll"
+
+"ATM":
+  - javascript/ql/experimental/adaptivethreatmodeling/**/*

.github/workflows/atm-check-queries-run.yml

@@ -1,56 +0,0 @@
-name: ATM Check Queries Run
-
-env:
-  DB_PATH: test_db
-  ATM_MODEL_PACK: javascript/ql/experimental/adaptivethreatmodeling/src
-  QUERY_SUITE: codeql-suites/javascript-atm-code-scanning.qls
-
-on:
-  pull_request:
-    paths:
-      - ".github/workflows/atm-check-queries-run.yml"
-      - "javascript/ql/experimental/adaptivethreatmodeling/**"
-  workflow_dispatch:
-
-jobs:
-  run-atm-queries:
-    runs-on: ubuntu-latest
-
-    steps:
-      - uses: actions/checkout@v3
-
-      - name: Install CodeQL CLI
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          gh extensions install github/gh-codeql
-          gh codeql download
-
-      - name: Install ATM model pack
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          set -exu
-          
-          # Install ATM model pack
-          gh codeql pack install ${ATM_MODEL_PACK}
-
-          # Retrieve model checksum
-          model_checksum=$(gh codeql resolve extensions ${ATM_MODEL_PACK}/${QUERY_SUITE} | jq -r '.models[0].checksum')
-
-          # Trust the model so that we can use it in the ATM boosted queries
-          mkdir -p "$HOME/.config/codeql"
-          echo "--insecurely-execute-ml-model-checksums ${model_checksum}" >> "$HOME/.config/codeql/config"
-
-      - name: Create test DB
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          gh codeql database create ${RUNNER_TEMP}/${DB_PATH} --source-root config/atm/ --language javascript 
-
-      - name: Run ATM query suite
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          gh codeql database run-queries -vv -- ${RUNNER_TEMP}/${DB_PATH} ${ATM_MODEL_PACK}/${QUERY_SUITE}
-      

.github/workflows/atm-check-query-suite.yml

@@ -0,0 +1,93 @@
+name: "ATM - Check query suite"
+
+env:
+  QUERY_PACK: javascript/ql/experimental/adaptivethreatmodeling/src
+  QUERY_SUITE: codeql-suites/javascript-atm-code-scanning.qls
+
+on:
+  pull_request:
+    paths:
+      - ".github/workflows/atm-check-query-suite.yml"
+      - "javascript/ql/experimental/adaptivethreatmodeling/**"
+  workflow_dispatch:
+
+jobs:
+  atm-check-query-suite:
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v3
+
+      - name: Setup CodeQL
+        uses: ./.github/actions/fetch-codeql
+        with:
+          channel: release
+
+      - name: Install ATM model
+        run: |
+          set -exu
+
+          # Install dependencies of ATM query pack, i.e. the ATM model
+          codeql pack install "${QUERY_PACK}"
+
+          # Retrieve model checksum
+          model_checksum=$(codeql resolve extensions "${QUERY_PACK}/${QUERY_SUITE}" | jq -r '.models[0].checksum')
+
+          # Trust the model so that we can use it in the ATM boosted queries
+          mkdir -p "$HOME/.config/codeql"
+          echo "--insecurely-execute-ml-model-checksums ${model_checksum}" >> "$HOME/.config/codeql/config"
+
+      - name: Create test DB
+        run: |
+          DB_PATH="${RUNNER_TEMP}/db"
+          echo "DB_PATH=${DB_PATH}" >> "${GITHUB_ENV}"
+
+          codeql database create "${DB_PATH}" --source-root config/atm --language javascript 
+
+      - name: Run ATM query suite
+        run: |
+          SARIF_PATH="${RUNNER_TEMP}/sarif.json"
+          echo "SARIF_PATH=${SARIF_PATH}" >> "${GITHUB_ENV}"
+
+          codeql database analyze \
+            --format sarif-latest \
+            --output "${SARIF_PATH}" \
+            --sarif-group-rules-by-pack \
+            -vv \
+            -- \
+            "${DB_PATH}" \
+            "${QUERY_PACK}/${QUERY_SUITE}"
+
+      - name: Upload SARIF
+        uses: actions/upload-artifact@v3
+        with:
+          name: javascript-ml-powered-queries.sarif
+          path: "${{ env.SARIF_PATH }}"
+          retention-days: 5
+
+      - name: Check results
+        run: |
+          # We should run at least the ML-powered queries in `expected_rules`.
+          expected_rules="js/ml-powered/nosql-injection js/ml-powered/path-injection js/ml-powered/sql-injection js/ml-powered/xss"
+
+          for rule in ${expected_rules}; do
+            found_rule=$(jq --arg rule "${rule}" '[.runs[0].tool.extensions[].rules | select(. != null) |
+              flatten | .[].id] | any(. == $rule)' "${SARIF_PATH}")
+            if [[ "${found_rule}" != "true" ]]; then
+              echo "Expected SARIF output to contain rule '${rule}', but found no such rule."
+              exit 1
+            else
+              echo "Found rule '${rule}'."
+            fi
+          done
+
+          # We should have at least one alert from an ML-powered query.
+          num_alerts=$(jq '[.runs[0].results[] |
+            select(.properties.score != null and (.rule.id | startswith("js/ml-powered/")))] | length' \
+            "${SARIF_PATH}")
+          if [[ "${num_alerts}" -eq 0 ]]; then
+            echo "Expected to find at least one alert from an ML-powered query but found ${num_alerts}."
+            exit 1
+          else
+            echo "Found ${num_alerts} alerts from ML-powered queries.";
+          fi

.github/workflows/compile-queries.yml

@@ -0,0 +1,71 @@
+name: "Compile all queries using the latest stable CodeQL CLI"
+
+on:
+  push:
+    branches:  # makes sure the cache gets populated - running on the branches people tend to merge into.
+      - main
+      - "rc/*"
+      - "codeql-cli-*"
+  pull_request:
+
+jobs:
+  compile-queries:
+    runs-on: ubuntu-latest-xl
+
+    steps:
+      - uses: actions/checkout@v3
+      # calculate the merge-base with main, in a way that works both on PRs and pushes to main. 
+      - name: Calculate merge-base
+        if: ${{ github.event_name == 'pull_request' }}
+        env:
+          BASE_BRANCH: ${{ github.base_ref }}
+        run: |
+          MERGE_BASE=$(git cat-file commit $GITHUB_SHA | grep '^parent ' | head -1 | cut -f 2 -d " ")
+          echo "merge-base=$MERGE_BASE" >> $GITHUB_ENV
+      - name: Read CodeQL query compilation - PR
+        if: ${{ github.event_name == 'pull_request' }}
+        uses: actions/cache@v3
+        with:
+          path: '*/ql/src/.cache'
+          key: codeql-compile-pr-${{ github.sha }} # deliberately not using the `compile-compile-main` keys here.
+          restore-keys: |
+            codeql-compile-${{ github.base_ref }}-${{ env.merge-base }}
+            codeql-compile-${{ github.base_ref }}-
+            codeql-compile-main-
+      - name: Fill CodeQL query compilation cache - main
+        if: ${{ github.event_name != 'pull_request' }}
+        uses: actions/cache@v3
+        with:
+          path: '*/ql/src/.cache'
+          key: codeql-compile-${{ github.ref_name }}-${{ github.sha }} # just fill on main
+          restore-keys: | # restore from another random commit, to speed up compilation.
+            codeql-compile-${{ github.ref_name }}- 
+            codeql-compile-main-
+      - name: Setup CodeQL
+        uses: ./.github/actions/fetch-codeql
+        with:
+          channel: 'release'
+      - name: check formatting
+        run: find */ql -type f \( -name "*.qll" -o -name "*.ql" \) -print0 | xargs -0 codeql query format --check-only
+      - name: compile queries - check-only
+        # run with --check-only if running in a PR (github.sha != main)
+        if : ${{ github.event_name == 'pull_request' }}
+        shell: bash
+        run: codeql query compile -j0 */ql/src --keep-going --warnings=error --check-only
+      - name: compile queries - full
+        # do full compile if running on main - this populates the cache
+        if : ${{ github.event_name != 'pull_request' }}
+        shell: bash
+        run: |
+          # Move all the existing cache into another folder, so we only preserve the cache for the current queries.
+          mkdir -p ${COMBINED_CACHE_DIR}
+          rm */ql/src/.cache/{lock,size}
+          # copy the contents of the .cache folders into the combined cache folder.
+          cp -r */ql/src/.cache/* ${COMBINED_CACHE_DIR}/
+          # clean up the .cache folders
+          rm -rf */ql/src/.cache/*
+
+          # compile the queries
+          codeql query compile -j0 */ql/src --keep-going --warnings=error --compilation-cache ${COMBINED_CACHE_DIR}
+        env: 
+          COMBINED_CACHE_DIR: ${{ github.workspace }}/compilation-dir

.github/workflows/js-ml-tests.yml

@@ -23,19 +23,6 @@ defaults:
     working-directory: javascript/ql/experimental/adaptivethreatmodeling
 
 jobs:
-  qlformat:
-    name: Check QL formatting
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v3
-
-      - uses: ./.github/actions/fetch-codeql
-
-      - name: Check QL formatting
-        run: |
-          find . "(" -name "*.ql" -or -name "*.qll" ")" -print0 | \
-            xargs -0 codeql query format --check-only
-
   qlcompile:
     name: Check QL compilation
     runs-on: ubuntu-latest

.github/workflows/ql-for-ql-build.yml

@@ -24,13 +24,13 @@ jobs:
       - uses: actions/checkout@v3
       - name: Find codeql
         id: find-codeql
-        uses: github/codeql-action/init@71a8b35ff4c80fcfcd05bc1cd932fe3c08f943ca
+        uses: github/codeql-action/init@77a8d2d10c0b403a8b4aadbd223dc489ecd22683
         with:
           languages: javascript # does not matter
       - name: Get CodeQL version
         id: get-codeql-version
         run: |
-          echo "::set-output name=version::$("${CODEQL}" --version | head -n 1 | rev | cut -d " " -f 1 | rev)"
+          echo "version=$("${CODEQL}" --version | head -n 1 | rev | cut -d " " -f 1 | rev)" >> $GITHUB_OUTPUT
         shell: bash
         env:
           CODEQL: ${{ steps.find-codeql.outputs.codeql-path }}
@@ -133,7 +133,7 @@ jobs:
         env:
           CONF: ./ql-for-ql-config.yml
       - name: Initialize CodeQL
-        uses: github/codeql-action/init@71a8b35ff4c80fcfcd05bc1cd932fe3c08f943ca
+        uses: github/codeql-action/init@77a8d2d10c0b403a8b4aadbd223dc489ecd22683
         with:
           languages: ql
           db-location: ${{ runner.temp }}/db
@@ -145,7 +145,7 @@ jobs:
           PACK: ${{ runner.temp }}/pack
 
       - name: Perform CodeQL Analysis
-        uses: github/codeql-action/analyze@71a8b35ff4c80fcfcd05bc1cd932fe3c08f943ca
+        uses: github/codeql-action/analyze@77a8d2d10c0b403a8b4aadbd223dc489ecd22683
         with:
           category: "ql-for-ql"
       - name: Copy sarif file to CWD

.github/workflows/ql-for-ql-dataset_measure.yml

@@ -25,7 +25,7 @@ jobs:
 
       - name: Find codeql
         id: find-codeql
-        uses: github/codeql-action/init@71a8b35ff4c80fcfcd05bc1cd932fe3c08f943ca
+        uses: github/codeql-action/init@77a8d2d10c0b403a8b4aadbd223dc489ecd22683
         with:
           languages: javascript # does not matter
       - uses: actions/cache@v3

.github/workflows/ql-for-ql-tests.yml

@@ -22,7 +22,7 @@ jobs:
       - uses: actions/checkout@v3
       - name: Find codeql
         id: find-codeql
-        uses: github/codeql-action/init@71a8b35ff4c80fcfcd05bc1cd932fe3c08f943ca
+        uses: github/codeql-action/init@77a8d2d10c0b403a8b4aadbd223dc489ecd22683
         with:
           languages: javascript # does not matter
       - uses: actions/cache@v3

.github/workflows/ruby-qltest.yml

@@ -28,13 +28,6 @@ defaults:
     working-directory: ruby
 
 jobs:
-  qlformat:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v3
-      - uses: ./.github/actions/fetch-codeql
-      - name: Check QL formatting
-        run: find ql "(" -name "*.ql" -or -name "*.qll" ")" -print0 | xargs -0 codeql query format --check-only
   qlcompile:
     runs-on: ubuntu-latest
     steps:

.github/workflows/swift-autobuilder.yml

@@ -1,27 +0,0 @@
-name: "Swift: Build and test Xcode autobuilder"
-
-on:
-  pull_request:
-    paths:
-      - "swift/xcode-autobuilder/**"
-      - "misc/bazel/**"
-      - "*.bazel*"
-      - .github/workflows/swift-autobuilder.yml
-    branches:
-      - main
-
-jobs:
-  autobuilder:
-    runs-on: macos-latest
-    steps:
-      - uses: actions/checkout@v3
-      - uses: bazelbuild/setup-bazelisk@v2
-      - uses: actions/setup-python@v4
-        with:
-          python-version-file: 'swift/.python-version'
-      - name: Build the Xcode autobuilder
-        run: |
-          bazel build //swift/xcode-autobuilder
-      - name: Test the Xcode autobuilder
-        run: |
-          bazel test //swift/xcode-autobuilder/tests

.github/workflows/swift-codegen.yml

@@ -1,44 +0,0 @@
-name: "Swift: Check code generation"
-
-on:
-  pull_request:
-    paths:
-      - "swift/**"
-      - "misc/bazel/**"
-      - "*.bazel*"
-      - .github/workflows/swift-codegen.yml
-      - .github/actions/fetch-codeql/action.yml
-    branches:
-      - main
-defaults:
-  run:
-    working-directory: swift
-
-jobs:
-  codegen:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v3
-      - uses: ./.github/actions/fetch-codeql
-      - uses: bazelbuild/setup-bazelisk@v2
-      - uses: actions/setup-python@v4
-        with:
-          python-version-file: 'swift/.python-version'
-      - uses: pre-commit/action@v3.0.0
-        name: Check that python code is properly formatted
-        with:
-          extra_args: autopep8 --all-files
-      - name: Run unit tests
-        run: |
-          bazel test //swift/codegen/test --test_output=errors
-      - uses: pre-commit/action@v3.0.0
-        name: Check that QL generated code was checked in
-        with:
-          extra_args: swift-codegen --all-files
-      - name: Generate C++ files
-        run: |
-          bazel run //swift/codegen:codegen -- --generate=trap,cpp --cpp-output=$PWD/swift-generated-cpp-files
-      - uses: actions/upload-artifact@v3
-        with:
-          name: swift-generated-cpp-files
-          path: swift-generated-cpp-files/**

.github/workflows/swift-integration-tests.yml

@@ -1,47 +0,0 @@
-name: "Swift: Run Integration Tests"
-
-on:
-  pull_request:
-    paths:
-      - "swift/**"
-      - "misc/bazel/**"
-      - "*.bazel*"
-      - .github/workflows/swift-integration-tests.yml
-      - .github/actions/fetch-codeql/action.yml
-      - codeql-workspace.yml
-    branches:
-      - main
-defaults:
-  run:
-    working-directory: swift
-
-jobs:
-  integration-tests:
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        os:
-          - ubuntu-20.04
-#          - macos-latest  TODO
-    steps:
-      - uses: actions/checkout@v3
-      - uses: ./.github/actions/fetch-codeql
-      - uses: bazelbuild/setup-bazelisk@v2
-      - uses: actions/setup-python@v4
-        with:
-          python-version-file: 'swift/.python-version'
-      - name: Build Swift extractor
-        run: |
-          bazel run //swift:create-extractor-pack
-      - name: Get Swift version
-        id: get_swift_version
-        run: |
-          VERSION=$(bazel run //swift/extractor -- --version | sed -ne 's/.*version \(\S*\).*/\1/p')
-          echo "::set-output name=version::$VERSION"
-      - uses: swift-actions/setup-swift@v1
-        with:
-          swift-version: "${{steps.get_swift_version.outputs.version}}"
-      - name: Run integration tests
-        run: |
-          python integration-tests/runner.py

.github/workflows/swift-qltest.yml

@@ -1,57 +0,0 @@
-name: "Swift: Run QL Tests"
-
-on:
-  pull_request:
-    paths:
-      - "swift/**"
-      - "misc/bazel/**"
-      - "*.bazel*"
-      - .github/workflows/swift-qltest.yml
-      - .github/actions/fetch-codeql/action.yml
-      - codeql-workspace.yml
-    branches:
-      - main
-defaults:
-  run:
-    working-directory: swift
-
-jobs:
-  qlformat:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v3
-      - uses: ./.github/actions/fetch-codeql
-      - name: Check QL formatting
-        run: find ql "(" -name "*.ql" -or -name "*.qll" ")" -print0 | xargs -0 codeql query format --check-only
-  qltest-test:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v3
-      - uses: bazelbuild/setup-bazelisk@v2
-      - uses: actions/setup-python@v4
-        with:
-          python-version-file: 'swift/.python-version'
-      - name: Test qltest.sh
-        run: |
-          bazel test //swift/tools/test/qltest
-  qltest:
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        os: [ ubuntu-20.04, macos-latest ]
-    steps:
-      - uses: actions/checkout@v3
-      - uses: ./.github/actions/fetch-codeql
-      - uses: bazelbuild/setup-bazelisk@v2
-      - uses: actions/setup-python@v4
-        with:
-          python-version-file: 'swift/.python-version'
-      - name: Build Swift extractor
-        run: |
-          bazel run //swift:create-extractor-pack
-      - name: Run QL tests
-        run: |
-          codeql test run --threads=0 --ram 5000 --search-path "${{ github.workspace }}/swift/extractor-pack" --check-databases --check-unused-labels --check-repeated-labels --check-redefined-labels --check-use-before-definition ql/test
-        env:
-          GITHUB_TOKEN: ${{ github.token }}

.github/workflows/swift.yml

@@ -0,0 +1,120 @@
+name: "Swift"
+
+on:
+  pull_request:
+    paths:
+      - "swift/**"
+      - "misc/bazel/**"
+      - "*.bazel*"
+      - .github/workflows/swift.yml
+      - .github/actions/fetch-codeql/action.yml
+      - codeql-workspace.yml
+      - .pre-commit-config.yaml
+      - "!**/*.md"
+      - "!**/*.qhelp"
+    branches:
+      - main
+
+jobs:
+  changes:
+    runs-on: ubuntu-latest
+    outputs:
+      codegen: ${{ steps.filter.outputs.codegen }}
+      ql: ${{ steps.filter.outputs.ql }}
+    steps:
+      - uses: dorny/paths-filter@4512585405083f25c027a35db413c2b3b9006d50
+        id: filter
+        with:
+          filters: |
+            codegen:
+              - '.github/workflows/swift.yml'
+              - "misc/bazel/**"
+              - "*.bazel*"
+              - 'swift/actions/setup-env/**'
+              - '.pre-commit-config.yaml'
+              - 'swift/codegen/**'
+              - 'swift/schema.py'
+              - 'swift/**/*.dbscheme'
+              - 'swift/ql/lib/codeql/swift/elements.qll'
+              - 'swift/ql/lib/codeql/swift/elements/**'
+              - 'swift/ql/lib/codeql/swift/generated/**'
+              - 'swift/ql/test/extractor-tests/generated/**'
+            ql:
+              - 'github/workflows/swift.yml'
+              - 'swift/**/*.ql'
+              - 'swift/**/*.qll'
+  # not using a matrix as you cannot depend on a specific job in a matrix, and we want to start linux checks
+  # without waiting for the macOS build
+  build-and-test-macos:
+    runs-on: macos-12-xl
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ./swift/actions/create-extractor-pack
+      - uses: ./swift/actions/run-quick-tests
+      - uses: ./swift/actions/print-unextracted
+  build-and-test-linux:
+    runs-on: ubuntu-20.04
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ./swift/actions/create-extractor-pack
+      - uses: ./swift/actions/run-quick-tests
+      - uses: ./swift/actions/print-unextracted
+  qltests-linux:
+    needs: build-and-test-linux
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ./swift/actions/run-ql-tests
+  qltests-macos:
+    needs: build-and-test-macos
+    runs-on: macos-12-xl
+    strategy:
+      fail-fast: false
+      matrix:
+        slice: ["1/2", "2/2"]
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ./swift/actions/run-ql-tests
+        with:
+          flags: --slice ${{ matrix.slice }}
+  integration-tests-linux:
+    needs: build-and-test-linux
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ./swift/actions/run-integration-tests
+  integration-tests-macos:
+    needs: build-and-test-macos
+    runs-on: macos-12-xl
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ./swift/actions/run-integration-tests
+  codegen:
+    runs-on: ubuntu-latest
+    needs: changes
+    if: ${{ needs.changes.outputs.codegen == 'true' }}
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ./swift/actions/setup-env
+      - uses: pre-commit/action@v3.0.0
+        name: Check that python code is properly formatted
+        with:
+          extra_args: autopep8 --all-files
+      - uses: ./.github/actions/fetch-codeql
+      - uses: pre-commit/action@v3.0.0
+        name: Check that QL generated code was checked in
+        with:
+          extra_args: swift-codegen --all-files
+      - name: Generate C++ files
+        run: |
+          bazel run //swift/codegen:codegen -- --generate=trap,cpp --cpp-output=$PWD/generated-cpp-files
+      - uses: actions/upload-artifact@v3
+        with:
+          name: swift-generated-cpp-files
+          path: generated-cpp-files/**
+  database-upgrade-scripts:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ./.github/actions/fetch-codeql
+      - uses: ./swift/actions/database-upgrade-scripts

.pre-commit-config.yaml

@@ -19,7 +19,7 @@ repos:
     rev: v1.6.0
     hooks:
       - id: autopep8
-        files: ^swift/codegen/.*\.py
+        files: ^swift/.*\.py
 
   - repo: local
     hooks:
@@ -31,7 +31,7 @@ repos:
 
       - id: sync-files
         name: Fix files required to be identical
-        files: \.(qll?|qhelp|swift)$
+        files: \.(qll?|qhelp|swift)$|^config/identical-files\.json$
         language: system
         entry: python3 config/sync-files.py --latest
         pass_filenames: false
@@ -44,7 +44,7 @@ repos:
 
       - id: swift-codegen
         name: Run Swift checked in code generation
-        files: ^swift/(codegen/|.*/generated/|ql/lib/(swift\.dbscheme$|codeql/swift/elements))
+        files: ^swift/(schema.py$|codegen/|.*/generated/|ql/lib/(swift\.dbscheme$|codeql/swift/elements))
         language: system
         entry: bazel run //swift/codegen -- --quiet
         pass_filenames: false

CODEOWNERS

@@ -40,8 +40,9 @@ WORKSPACE.bazel @github/codeql-ci-reviewers
 
 # Workflows
 /.github/workflows/ @github/codeql-ci-reviewers
+/.github/workflows/atm-* @github/codeql-ml-powered-queries-reviewers
 /.github/workflows/go-* @github/codeql-go
 /.github/workflows/js-ml-tests.yml @github/codeql-ml-powered-queries-reviewers
 /.github/workflows/ql-for-ql-* @github/codeql-ql-for-ql-reviewers
 /.github/workflows/ruby-* @github/codeql-ruby
-/.github/workflows/swift-* @github/codeql-c
+/.github/workflows/swift.yml @github/codeql-c

codeql-workspace.yml

@@ -4,7 +4,6 @@ provide:
   - "*/ql/test/qlpack.yml"
   - "*/ql/examples/qlpack.yml"
   - "*/ql/consistency-queries/qlpack.yml"
-  - "*/upgrades/qlpack.yml"
   - "shared/*/qlpack.yml"
   - "cpp/ql/test/query-tests/Security/CWE/CWE-190/semmle/tainted/qlpack.yml"
   - "go/ql/config/legacy-support/qlpack.yml"
@@ -18,6 +17,7 @@ provide:
   # - "javascript/ql/experimental/adaptivethreatmodeling/model/qlpack.yml"
   - "javascript/ql/experimental/adaptivethreatmodeling/modelbuilding/qlpack.yml"
   - "javascript/ql/experimental/adaptivethreatmodeling/src/qlpack.yml"
+  - "javascript/ql/experimental/adaptivethreatmodeling/test/qlpack.yml"
   - "csharp/ql/campaigns/Solorigate/lib/qlpack.yml"
   - "csharp/ql/campaigns/Solorigate/src/qlpack.yml"
   - "csharp/ql/campaigns/Solorigate/test/qlpack.yml"
@@ -26,3 +26,9 @@ provide:
   - "ruby/extractor-pack/codeql-extractor.yml"
   - "swift/extractor-pack/codeql-extractor.yml"
   - "ql/extractor-pack/codeql-extractor.ym"
+
+versionPolicies:
+  default:
+    requireChangeNotes: true
+    committedPrereleaseSuffix: dev
+    committedVersion: nextPatchRelease

config/identical-files.json

@@ -94,8 +94,8 @@
     "csharp/ql/lib/semmle/code/csharp/dataflow/internal/rangeanalysis/SsaReadPositionCommon.qll"
   ],
   "Model as Data Generation Java/C# - CaptureModels": [
-    "java/ql/src/utils/model-generator/internal/CaptureModels.qll",
-    "csharp/ql/src/utils/model-generator/internal/CaptureModels.qll"
+    "java/ql/src/utils/modelgenerator/internal/CaptureModels.qll",
+    "csharp/ql/src/utils/modelgenerator/internal/CaptureModels.qll"
   ],
   "Sign Java/C#": [
     "java/ql/lib/semmle/code/java/dataflow/internal/rangeanalysis/Sign.qll",
@@ -486,40 +486,6 @@
     "python/ql/lib/semmle/python/security/internal/SensitiveDataHeuristics.qll",
     "ruby/ql/lib/codeql/ruby/security/internal/SensitiveDataHeuristics.qll"
   ],
-  "ReDoS Util Python/JS/Ruby/Java": [
-    "javascript/ql/lib/semmle/javascript/security/regexp/NfaUtils.qll",
-    "python/ql/lib/semmle/python/security/regexp/NfaUtils.qll",
-    "ruby/ql/lib/codeql/ruby/security/regexp/NfaUtils.qll",
-    "java/ql/lib/semmle/code/java/security/regexp/NfaUtils.qll"
-  ],
-  "ReDoS Exponential Python/JS/Ruby/Java": [
-    "javascript/ql/lib/semmle/javascript/security/regexp/ExponentialBackTracking.qll",
-    "python/ql/lib/semmle/python/security/regexp/ExponentialBackTracking.qll",
-    "ruby/ql/lib/codeql/ruby/security/regexp/ExponentialBackTracking.qll",
-    "java/ql/lib/semmle/code/java/security/regexp/ExponentialBackTracking.qll"
-  ],
-  "ReDoS Polynomial Python/JS/Ruby/Java": [
-    "javascript/ql/lib/semmle/javascript/security/regexp/SuperlinearBackTracking.qll",
-    "python/ql/lib/semmle/python/security/regexp/SuperlinearBackTracking.qll",
-    "ruby/ql/lib/codeql/ruby/security/regexp/SuperlinearBackTracking.qll",
-    "java/ql/lib/semmle/code/java/security/regexp/SuperlinearBackTracking.qll"
-  ],
-  "RegexpMatching Python/JS/Ruby": [
-    "javascript/ql/lib/semmle/javascript/security/regexp/RegexpMatching.qll",
-    "python/ql/lib/semmle/python/security/regexp/RegexpMatching.qll",
-    "ruby/ql/lib/codeql/ruby/security/regexp/RegexpMatching.qll"
-  ],
-  "BadTagFilterQuery Python/JS/Ruby": [
-    "javascript/ql/lib/semmle/javascript/security/BadTagFilterQuery.qll",
-    "python/ql/lib/semmle/python/security/BadTagFilterQuery.qll",
-    "ruby/ql/lib/codeql/ruby/security/BadTagFilterQuery.qll"
-  ],
-  "OverlyLargeRange Python/JS/Ruby/Java": [
-    "javascript/ql/lib/semmle/javascript/security/OverlyLargeRangeQuery.qll",
-    "python/ql/lib/semmle/python/security/OverlyLargeRangeQuery.qll",
-    "ruby/ql/lib/codeql/ruby/security/OverlyLargeRangeQuery.qll",
-    "java/ql/lib/semmle/code/java/security/OverlyLargeRangeQuery.qll"
-  ],
   "CFG": [
     "csharp/ql/lib/semmle/code/csharp/controlflow/internal/ControlFlowGraphImplShared.qll",
     "ruby/ql/lib/codeql/ruby/controlflow/internal/ControlFlowGraphImplShared.qll",
@@ -600,8 +566,12 @@
     "swift/ql/test/extractor-tests/patterns/patterns.swift",
     "swift/ql/test/library-tests/ast/patterns.swift"
   ],
+  "Swift control flow test file": [
+    "swift/ql/test/library-tests/controlflow/graph/cfg.swift",
+    "swift/ql/test/library-tests/ast/cfg.swift"
+  ],
   "IncompleteMultiCharacterSanitization JS/Ruby": [
     "javascript/ql/lib/semmle/javascript/security/IncompleteMultiCharacterSanitizationQuery.qll",
     "ruby/ql/lib/codeql/ruby/security/IncompleteMultiCharacterSanitizationQuery.qll"
   ]
-}
+}

cpp/autobuilder/Semmle.Autobuild.Cpp.Tests/Semmle.Autobuild.Cpp.Tests.csproj

@@ -11,11 +11,12 @@
   <ItemGroup>
     <PackageReference Include="System.IO.FileSystem" Version="4.3.0" />
     <PackageReference Include="System.IO.FileSystem.Primitives" Version="4.3.0" />
-    <PackageReference Include="xunit" Version="2.4.1" />
-    <PackageReference Include="xunit.runner.visualstudio" Version="2.4.1">
+    <PackageReference Include="xunit" Version="2.4.2" />
+    <PackageReference Include="xunit.runner.visualstudio" Version="2.4.5">
       <PrivateAssets>all</PrivateAssets>
       <IncludeAssets>runtime; build; native; contentfiles; analyzers</IncludeAssets>
     </PackageReference>
+    <PackageReference Include="Microsoft.NET.Test.Sdk" Version="17.4.0" />
   </ItemGroup>
 
   <ItemGroup>

cpp/autobuilder/Semmle.Autobuild.Cpp/Semmle.Autobuild.Cpp.csproj

@@ -17,7 +17,7 @@
   </ItemGroup>
 
   <ItemGroup>
-    <PackageReference Include="Microsoft.Build" Version="16.11.0" />
+    <PackageReference Include="Microsoft.Build" Version="17.3.2" />
   </ItemGroup>
 
   <ItemGroup>

cpp/ql/examples/qlpack.yml

@@ -1,4 +1,6 @@
 name: codeql/cpp-examples
-version: 0.0.2
+groups: 
+  - cpp
+  - examples
 dependencies:
-  codeql/cpp-all: "*"
+  codeql/cpp-all: ${workspace}

cpp/ql/lib/BUILD.bazel

@@ -0,0 +1,15 @@
+package(default_visibility = ["//cpp:__pkg__"])
+
+load("@rules_pkg//:mappings.bzl", "pkg_files")
+
+pkg_files(
+    name = "dbscheme",
+    srcs = ["semmlecode.cpp.dbscheme"],
+    prefix = "cpp",
+)
+
+pkg_files(
+    name = "dbscheme-stats",
+    srcs = ["semmlecode.cpp.dbscheme.stats"],
+    prefix = "cpp",
+)

cpp/ql/lib/CHANGELOG.md

@@ -0,0 +1,194 @@
+## 0.4.5
+
+No user-facing changes.
+
+## 0.4.4
+
+No user-facing changes.
+
+## 0.4.3
+
+### Minor Analysis Improvements
+
+* Fixed bugs in the `FormatLiteral` class that were causing `getMaxConvertedLength` and related predicates to return no results when the format literal was `%e`, `%f` or `%g` and an explicit precision was specified.
+
+## 0.4.2
+
+No user-facing changes.
+
+## 0.4.1
+
+No user-facing changes.
+
+## 0.4.0
+
+### Deprecated APIs
+
+* Some classes/modules with upper-case acronyms in their name have been renamed to follow our style-guide. 
+  The old name still exists as a deprecated alias.
+
+### New Features
+
+* Added subclasses of `BuiltInOperations` for `__is_same`, `__is_function`, `__is_layout_compatible`, `__is_pointer_interconvertible_base_of`, `__is_array`, `__array_rank`, `__array_extent`, `__is_arithmetic`, `__is_complete_type`, `__is_compound`, `__is_const`, `__is_floating_point`, `__is_fundamental`, `__is_integral`, `__is_lvalue_reference`, `__is_member_function_pointer`, `__is_member_object_pointer`, `__is_member_pointer`, `__is_object`, `__is_pointer`, `__is_reference`, `__is_rvalue_reference`, `__is_scalar`, `__is_signed`, `__is_unsigned`, `__is_void`, and `__is_volatile`.
+
+### Bug Fixes
+
+* Fixed an issue in the taint tracking analysis where implicit reads were not allowed by default in sinks or additional taint steps that used flow states.
+
+## 0.3.5
+
+## 0.3.4
+
+### Deprecated APIs
+
+* Many classes/predicates/modules with upper-case acronyms in their name have been renamed to follow our style-guide. 
+  The old name still exists as a deprecated alias.
+
+### New Features
+
+* Added support for getting the link targets of global and namespace variables.
+* Added a `BlockAssignExpr` class, which models a `memcpy`-like operation used in compiler generated copy/move constructors and assignment operations.
+
+### Minor Analysis Improvements
+
+* All deprecated predicates/classes/modules that have been deprecated for over a year have been deleted.
+
+## 0.3.3
+
+### New Features
+
+* Added a predicate `getValueConstant` to `AttributeArgument` that yields the argument value as an `Expr` when the value is a constant expression.
+* A new class predicate `MustFlowConfiguration::allowInterproceduralFlow` has been added to the `semmle.code.cpp.ir.dataflow.MustFlow` library. The new predicate can be overridden to disable interprocedural flow.
+* Added subclasses of `BuiltInOperations` for `__builtin_bit_cast`, `__builtin_shuffle`, `__has_unique_object_representations`, `__is_aggregate`, and `__is_assignable`.
+
+### Major Analysis Improvements
+
+* The IR dataflow library now includes flow through global variables. This enables new findings in many scenarios.
+
+## 0.3.2
+
+### Bug Fixes
+
+* Under certain circumstances a variable declaration that is not also a definition could be associated with a `Variable` that did not have the definition as a `VariableDeclarationEntry`. This is now fixed, and a unique `Variable` will exist that has both the declaration and the definition as a `VariableDeclarationEntry`.
+
+## 0.3.1
+
+### Minor Analysis Improvements
+
+* `AnalysedExpr::isNullCheck` and `AnalysedExpr::isValidCheck` have been updated to handle variable accesses on the left-hand side of the C++ logical "and", and variable declarations in conditions.
+
+## 0.3.0
+
+### Deprecated APIs
+
+* The `BarrierGuard` class has been deprecated. Such barriers and sanitizers can now instead be created using the new `BarrierGuard` parameterized module.
+
+### Bug Fixes
+
+* `UserType.getADeclarationEntry()` now yields all forward declarations when the user type is a `class`, `struct`, or `union`.
+
+## 0.2.3
+
+### New Features
+
+* An `isBraced` predicate was added to the `Initializer` class which holds when a C++ braced initializer was used in the initialization.
+
+## 0.2.2
+
+### Deprecated APIs
+
+ * The `AnalysedString` class in the `StringAnalysis` module has been replaced with `AnalyzedString`, to follow our style guide. The old name still exists as a deprecated alias.
+
+### New Features
+
+* A `getInitialization` predicate was added to the `ConstexprIfStmt`, `IfStmt`, and `SwitchStmt` classes that yields the C++17-style initializer of the `if` or `switch` statement when it exists.
+
+## 0.2.1
+
+## 0.2.0
+
+### Breaking Changes
+
+* The signature of `allowImplicitRead` on `DataFlow::Configuration` and `TaintTracking::Configuration` has changed from `allowImplicitRead(DataFlow::Node node, DataFlow::Content c)` to `allowImplicitRead(DataFlow::Node node, DataFlow::ContentSet c)`.
+
+### Minor Analysis Improvements
+
+* More Windows pool allocation functions are now detected as `AllocationFunction`s.
+* The `semmle.code.cpp.commons.Buffer` library has been enhanced to handle array members of classes that do not specify a size.
+
+## 0.1.0
+
+### Breaking Changes
+
+* The recently added flow-state versions of `isBarrierIn`, `isBarrierOut`, `isSanitizerIn`, and `isSanitizerOut` in the data flow and taint tracking libraries have been removed.
+
+### New Features
+
+* A new library `semmle.code.cpp.security.PrivateData` has been added. The new library heuristically detects variables and functions dealing with sensitive private data, such as e-mail addresses and credit card numbers.
+
+### Minor Analysis Improvements
+
+* The `semmle.code.cpp.security.SensitiveExprs` library has been enhanced with some additional rules for detecting credentials.
+
+## 0.0.13
+
+## 0.0.12
+
+### Breaking Changes
+
+* The flow state variants of `isBarrier` and `isAdditionalFlowStep` are no longer exposed in the taint tracking library. The `isSanitizer` and `isAdditionalTaintStep` predicates should be used instead.
+
+### Deprecated APIs
+
+* Many classes/predicates/modules that had upper-case acronyms have been renamed to follow our style-guide. 
+  The old name still exists as a deprecated alias.
+
+### New Features
+
+* The data flow and taint tracking libraries have been extended with versions of `isBarrierIn`, `isBarrierOut`, and `isBarrierGuard`, respectively `isSanitizerIn`, `isSanitizerOut`, and `isSanitizerGuard`, that support flow states.
+
+### Minor Analysis Improvements
+
+* `DefaultOptions::exits` now holds for C11 functions with the `_Noreturn` or `noreturn` specifier.
+* `hasImplicitCopyConstructor` and `hasImplicitCopyAssignmentOperator` now correctly handle implicitly-deleted operators in templates.
+* All deprecated predicates/classes/modules that have been deprecated for over a year have been deleted.
+
+## 0.0.11
+
+### Minor Analysis Improvements
+
+* Many queries now support structured bindings, as structured bindings are now handled in the IR translation.
+
+## 0.0.10
+
+### New Features
+
+* Added a `isStructuredBinding` predicate to the `Variable` class which holds when the variable is declared as part of a structured binding declaration.
+
+## 0.0.9
+
+
+## 0.0.8
+
+### Deprecated APIs
+
+* The `codeql/cpp-upgrades` CodeQL pack has been removed. All upgrades scripts have been merged into the `codeql/cpp-all` CodeQL pack.
+
+### Minor Analysis Improvements
+
+* `FormatLiteral::getMaxConvertedLength` now uses range analysis to provide a
+  more accurate length for integers formatted with `%x`
+
+## 0.0.7
+
+## 0.0.6
+
+## 0.0.5
+
+## 0.0.4
+
+### New Features
+
+* The QL library `semmle.code.cpp.commons.Exclusions` now contains a predicate
+  `isFromSystemMacroDefinition` for identifying code that originates from a
+  macro outside the project being analyzed.

cpp/ql/lib/DefaultOptions.qll

@@ -54,11 +54,13 @@ class Options extends string {
    *
    * By default, this holds for `exit`, `_exit`, `abort`, `__assert_fail`,
    * `longjmp`, `__builtin_unreachable` and any function with a
-   * `noreturn` attribute.
+   * `noreturn` attribute or specifier.
    */
   predicate exits(Function f) {
     f.getAnAttribute().hasName("noreturn")
     or
+    f.getASpecifier().hasName("noreturn")
+    or
     f.hasGlobalOrStdName([
         "exit", "_exit", "abort", "__assert_fail", "longjmp", "__builtin_unreachable"
       ])
@@ -73,7 +75,7 @@ class Options extends string {
    *   __assume(0);
    * ```
    * (note that in this case if the hint is wrong and the expression is reached at
-   * runtime, the program's behaviour is undefined)
+   * runtime, the program's behavior is undefined)
    */
   predicate exprExits(Expr e) {
     e.(AssumeExpr).getChild(0).(CompileTimeConstantInt).getIntValue() = 0 or

cpp/ql/lib/Options.qll

@@ -39,7 +39,7 @@ class CustomOptions extends Options {
    *
    * By default, this holds for `exit`, `_exit`, `abort`, `__assert_fail`,
    * `longjmp`, `error`, `__builtin_unreachable` and any function with a
-   * `noreturn` attribute.
+   * `noreturn` attribute or specifier.
    */
   override predicate exits(Function f) { Options.super.exits(f) }
 
@@ -50,7 +50,7 @@ class CustomOptions extends Options {
    *   __assume(0);
    * ```
    * (note that in this case if the hint is wrong and the expression is reached at
-   * runtime, the program's behaviour is undefined)
+   * runtime, the program's behavior is undefined)
    */
   override predicate exprExits(Expr e) { Options.super.exprExits(e) }
 

cpp/ql/lib/change-notes/released/0.0.10.md

@@ -0,0 +1,5 @@
+## 0.0.10
+
+### New Features
+
+* Added a `isStructuredBinding` predicate to the `Variable` class which holds when the variable is declared as part of a structured binding declaration.

cpp/ql/lib/change-notes/released/0.0.11.md

@@ -0,0 +1,5 @@
+## 0.0.11
+
+### Minor Analysis Improvements
+
+* Many queries now support structured bindings, as structured bindings are now handled in the IR translation.

cpp/ql/lib/change-notes/released/0.0.12.md

@@ -0,0 +1,20 @@
+## 0.0.12
+
+### Breaking Changes
+
+* The flow state variants of `isBarrier` and `isAdditionalFlowStep` are no longer exposed in the taint tracking library. The `isSanitizer` and `isAdditionalTaintStep` predicates should be used instead.
+
+### Deprecated APIs
+
+* Many classes/predicates/modules that had upper-case acronyms have been renamed to follow our style-guide. 
+  The old name still exists as a deprecated alias.
+
+### New Features
+
+* The data flow and taint tracking libraries have been extended with versions of `isBarrierIn`, `isBarrierOut`, and `isBarrierGuard`, respectively `isSanitizerIn`, `isSanitizerOut`, and `isSanitizerGuard`, that support flow states.
+
+### Minor Analysis Improvements
+
+* `DefaultOptions::exits` now holds for C11 functions with the `_Noreturn` or `noreturn` specifier.
+* `hasImplicitCopyConstructor` and `hasImplicitCopyAssignmentOperator` now correctly handle implicitly-deleted operators in templates.
+* All deprecated predicates/classes/modules that have been deprecated for over a year have been deleted.

cpp/ql/lib/change-notes/released/0.0.13.md

@@ -0,0 +1 @@
+## 0.0.13

cpp/ql/lib/change-notes/released/0.0.4.md

@@ -0,0 +1,7 @@
+## 0.0.4
+
+### New Features
+
+* The QL library `semmle.code.cpp.commons.Exclusions` now contains a predicate
+  `isFromSystemMacroDefinition` for identifying code that originates from a
+  macro outside the project being analyzed.

cpp/ql/lib/change-notes/released/0.0.5.md

@@ -0,0 +1 @@
+## 0.0.5

cpp/ql/lib/change-notes/released/0.0.6.md

@@ -0,0 +1 @@
+## 0.0.6

cpp/ql/lib/change-notes/released/0.0.7.md

@@ -0,0 +1 @@
+## 0.0.7

cpp/ql/lib/change-notes/released/0.0.8.md

@@ -0,0 +1,10 @@
+## 0.0.8
+
+### Deprecated APIs
+
+* The `codeql/cpp-upgrades` CodeQL pack has been removed. All upgrades scripts have been merged into the `codeql/cpp-all` CodeQL pack.
+
+### Minor Analysis Improvements
+
+* `FormatLiteral::getMaxConvertedLength` now uses range analysis to provide a
+  more accurate length for integers formatted with `%x`

cpp/ql/lib/change-notes/released/0.0.9.md

@@ -0,0 +1,2 @@
+## 0.0.9
+

cpp/ql/lib/change-notes/released/0.1.0.md

@@ -0,0 +1,13 @@
+## 0.1.0
+
+### Breaking Changes
+
+* The recently added flow-state versions of `isBarrierIn`, `isBarrierOut`, `isSanitizerIn`, and `isSanitizerOut` in the data flow and taint tracking libraries have been removed.
+
+### New Features
+
+* A new library `semmle.code.cpp.security.PrivateData` has been added. The new library heuristically detects variables and functions dealing with sensitive private data, such as e-mail addresses and credit card numbers.
+
+### Minor Analysis Improvements
+
+* The `semmle.code.cpp.security.SensitiveExprs` library has been enhanced with some additional rules for detecting credentials.

cpp/ql/lib/change-notes/released/0.2.0.md

@@ -0,0 +1,10 @@
+## 0.2.0
+
+### Breaking Changes
+
+* The signature of `allowImplicitRead` on `DataFlow::Configuration` and `TaintTracking::Configuration` has changed from `allowImplicitRead(DataFlow::Node node, DataFlow::Content c)` to `allowImplicitRead(DataFlow::Node node, DataFlow::ContentSet c)`.
+
+### Minor Analysis Improvements
+
+* More Windows pool allocation functions are now detected as `AllocationFunction`s.
+* The `semmle.code.cpp.commons.Buffer` library has been enhanced to handle array members of classes that do not specify a size.

cpp/ql/lib/change-notes/released/0.2.1.md

@@ -0,0 +1 @@
+## 0.2.1

cpp/ql/lib/change-notes/released/0.2.2.md

@@ -0,0 +1,9 @@
+## 0.2.2
+
+### Deprecated APIs
+
+ * The `AnalysedString` class in the `StringAnalysis` module has been replaced with `AnalyzedString`, to follow our style guide. The old name still exists as a deprecated alias.
+
+### New Features
+
+* A `getInitialization` predicate was added to the `ConstexprIfStmt`, `IfStmt`, and `SwitchStmt` classes that yields the C++17-style initializer of the `if` or `switch` statement when it exists.

cpp/ql/lib/change-notes/released/0.2.3.md

@@ -0,0 +1,5 @@
+## 0.2.3
+
+### New Features
+
+* An `isBraced` predicate was added to the `Initializer` class which holds when a C++ braced initializer was used in the initialization.

cpp/ql/lib/change-notes/released/0.3.0.md

@@ -0,0 +1,9 @@
+## 0.3.0
+
+### Deprecated APIs
+
+* The `BarrierGuard` class has been deprecated. Such barriers and sanitizers can now instead be created using the new `BarrierGuard` parameterized module.
+
+### Bug Fixes
+
+* `UserType.getADeclarationEntry()` now yields all forward declarations when the user type is a `class`, `struct`, or `union`.

cpp/ql/lib/change-notes/released/0.3.1.md

@@ -0,0 +1,5 @@
+## 0.3.1
+
+### Minor Analysis Improvements
+
+* `AnalysedExpr::isNullCheck` and `AnalysedExpr::isValidCheck` have been updated to handle variable accesses on the left-hand side of the C++ logical "and", and variable declarations in conditions.

cpp/ql/lib/change-notes/released/0.3.2.md

@@ -0,0 +1,5 @@
+## 0.3.2
+
+### Bug Fixes
+
+* Under certain circumstances a variable declaration that is not also a definition could be associated with a `Variable` that did not have the definition as a `VariableDeclarationEntry`. This is now fixed, and a unique `Variable` will exist that has both the declaration and the definition as a `VariableDeclarationEntry`.

cpp/ql/lib/change-notes/released/0.3.3.md

@@ -0,0 +1,11 @@
+## 0.3.3
+
+### New Features
+
+* Added a predicate `getValueConstant` to `AttributeArgument` that yields the argument value as an `Expr` when the value is a constant expression.
+* A new class predicate `MustFlowConfiguration::allowInterproceduralFlow` has been added to the `semmle.code.cpp.ir.dataflow.MustFlow` library. The new predicate can be overridden to disable interprocedural flow.
+* Added subclasses of `BuiltInOperations` for `__builtin_bit_cast`, `__builtin_shuffle`, `__has_unique_object_representations`, `__is_aggregate`, and `__is_assignable`.
+
+### Major Analysis Improvements
+
+* The IR dataflow library now includes flow through global variables. This enables new findings in many scenarios.

cpp/ql/lib/change-notes/released/0.3.4.md

@@ -0,0 +1,15 @@
+## 0.3.4
+
+### Deprecated APIs
+
+* Many classes/predicates/modules with upper-case acronyms in their name have been renamed to follow our style-guide. 
+  The old name still exists as a deprecated alias.
+
+### New Features
+
+* Added support for getting the link targets of global and namespace variables.
+* Added a `BlockAssignExpr` class, which models a `memcpy`-like operation used in compiler generated copy/move constructors and assignment operations.
+
+### Minor Analysis Improvements
+
+* All deprecated predicates/classes/modules that have been deprecated for over a year have been deleted.

cpp/ql/lib/change-notes/released/0.3.5.md

@@ -0,0 +1 @@
+## 0.3.5

cpp/ql/lib/change-notes/released/0.4.0.md

@@ -0,0 +1,14 @@
+## 0.4.0
+
+### Deprecated APIs
+
+* Some classes/modules with upper-case acronyms in their name have been renamed to follow our style-guide. 
+  The old name still exists as a deprecated alias.
+
+### New Features
+
+* Added subclasses of `BuiltInOperations` for `__is_same`, `__is_function`, `__is_layout_compatible`, `__is_pointer_interconvertible_base_of`, `__is_array`, `__array_rank`, `__array_extent`, `__is_arithmetic`, `__is_complete_type`, `__is_compound`, `__is_const`, `__is_floating_point`, `__is_fundamental`, `__is_integral`, `__is_lvalue_reference`, `__is_member_function_pointer`, `__is_member_object_pointer`, `__is_member_pointer`, `__is_object`, `__is_pointer`, `__is_reference`, `__is_rvalue_reference`, `__is_scalar`, `__is_signed`, `__is_unsigned`, `__is_void`, and `__is_volatile`.
+
+### Bug Fixes
+
+* Fixed an issue in the taint tracking analysis where implicit reads were not allowed by default in sinks or additional taint steps that used flow states.

cpp/ql/lib/change-notes/released/0.4.1.md

@@ -0,0 +1,3 @@
+## 0.4.1
+
+No user-facing changes.

cpp/ql/lib/change-notes/released/0.4.2.md

@@ -0,0 +1,3 @@
+## 0.4.2
+
+No user-facing changes.

cpp/ql/lib/change-notes/released/0.4.3.md

@@ -0,0 +1,5 @@
+## 0.4.3
+
+### Minor Analysis Improvements
+
+* Fixed bugs in the `FormatLiteral` class that were causing `getMaxConvertedLength` and related predicates to return no results when the format literal was `%e`, `%f` or `%g` and an explicit precision was specified.

cpp/ql/lib/change-notes/released/0.4.4.md

@@ -0,0 +1,3 @@
+## 0.4.4
+
+No user-facing changes.

cpp/ql/lib/change-notes/released/0.4.5.md

@@ -0,0 +1,3 @@
+## 0.4.5
+
+No user-facing changes.

cpp/ql/lib/codeql-pack.release.yml

@@ -0,0 +1,2 @@
+---
+lastReleaseVersion: 0.4.5

cpp/ql/lib/cpp.qll

@@ -69,6 +69,4 @@ import semmle.code.cpp.Comments
 import semmle.code.cpp.Preprocessor
 import semmle.code.cpp.Iteration
 import semmle.code.cpp.NameQualifiers
-import semmle.code.cpp.ObjectiveC
-import semmle.code.cpp.exprs.ObjectiveC
 import DefaultOptions

cpp/ql/lib/definitions.qll

@@ -0,0 +1,210 @@
+/**
+ * Provides classes and predicates related to jump-to-definition links
+ * in the code viewer.
+ */
+
+import cpp
+import IDEContextual
+
+/**
+ * Any element that might be the source or target of a jump-to-definition
+ * link.
+ *
+ * In some cases it is preferable to modify locations (the
+ * `hasLocationInfo()` predicate) so that they are short, and
+ * non-overlapping with other locations that might be reported as
+ * code scanning alerts on GitHub.
+ *
+ * We need to give locations that may not be in the database, so
+ * we use `hasLocationInfo()` rather than `getLocation()`.
+ */
+class Top extends Element {
+  /**
+   * Holds if this element is at the specified location.
+   * The location spans column `startcolumn` of line `startline` to
+   * column `endcolumn` of line `endline` in file `filepath`.
+   * For more information, see
+   * [Locations](https://codeql.github.com/docs/writing-codeql-queries/providing-locations-in-codeql-queries/).
+   */
+  pragma[noopt]
+  final predicate hasLocationInfo(
+    string filepath, int startline, int startcolumn, int endline, int endcolumn
+  ) {
+    interestingElement(this) and
+    not this instanceof MacroAccess and
+    not this instanceof Include and
+    exists(Location l |
+      l = this.getLocation() and
+      l.hasLocationInfo(filepath, startline, startcolumn, endline, endcolumn)
+    )
+    or
+    // This has a location that covers only the name of the accessed
+    // macro, not its arguments (which are included by `MacroAccess`'s
+    // `getLocation()`).
+    exists(Location l, MacroAccess ma |
+      ma instanceof MacroAccess and
+      ma = this and
+      l = ma.getLocation() and
+      l.hasLocationInfo(filepath, startline, startcolumn, _, _) and
+      endline = startline and
+      exists(string macroName, int nameLength, int nameLengthMinusOne |
+        macroName = ma.getMacroName() and
+        nameLength = macroName.length() and
+        nameLengthMinusOne = nameLength - 1 and
+        endcolumn = startcolumn + nameLengthMinusOne
+      )
+    )
+    or
+    hasLocationInfo_Include(this, filepath, startline, startcolumn, endline, endcolumn)
+  }
+}
+
+/**
+ * An `Include` with a `hasLocationInfo` predicate.
+ *
+ * This has a location that covers only the name of the included
+ * file, not the `#include` text or whitespace before it.
+ */
+predicate hasLocationInfo_Include(Include i, string path, int sl, int sc, int el, int ec) {
+  exists(Location l |
+    l = i.getLocation() and
+    path = l.getFile().getAbsolutePath() and
+    sl = l.getEndLine() and
+    sc = l.getEndColumn() + 1 - i.getIncludeText().length() and
+    el = l.getEndLine() and
+    ec = l.getEndColumn()
+  )
+}
+
+/** Holds if `e` is a source or a target of jump-to-definition. */
+predicate interestingElement(Element e) {
+  exists(definitionOf(e, _))
+  or
+  e = definitionOf(_, _)
+}
+
+/**
+ * Holds if `f`, `line`, `column` indicate the start character
+ * of `cc`.
+ */
+private predicate constructorCallStartLoc(ConstructorCall cc, File f, int line, int column) {
+  exists(Location l |
+    l = cc.getLocation() and
+    l.getFile() = f and
+    l.getStartLine() = line and
+    l.getStartColumn() = column
+  )
+}
+
+/**
+ * Holds if `f`, `line`, `column` indicate the start character
+ * of `tm`, which mentions `t`. Type mentions for instantiations
+ * are filtered out.
+ */
+private predicate typeMentionStartLoc(TypeMention tm, Type t, File f, int line, int column) {
+  exists(Location l |
+    l = tm.getLocation() and
+    l.getFile() = f and
+    l.getStartLine() = line and
+    l.getStartColumn() = column
+  ) and
+  t = tm.getMentionedType() and
+  not t instanceof ClassTemplateInstantiation
+}
+
+/**
+ * Holds if `cc` and `tm` begin at the same character.
+ */
+cached
+private predicate constructorCallTypeMention(ConstructorCall cc, TypeMention tm) {
+  exists(File f, int line, int column |
+    constructorCallStartLoc(cc, f, line, column) and
+    typeMentionStartLoc(tm, _, f, line, column)
+  )
+}
+
+/**
+ * Gets an element, of kind `kind`, that element `e` uses, if any.
+ * Attention: This predicate yields multiple definitions for a single location.
+ *
+ * The `kind` is a string representing what kind of use it is:
+ *  - `"M"` for function and method calls
+ *  - `"T"` for uses of types
+ *  - `"V"` for variable accesses
+ *  - `"X"` for macro accesses
+ *  - `"I"` for import / include directives
+ */
+cached
+Top definitionOf(Top e, string kind) {
+  (
+    // call -> function called
+    kind = "M" and
+    result = e.(Call).getTarget() and
+    not e.(Expr).isCompilerGenerated() and
+    not e instanceof ConstructorCall // handled elsewhere
+    or
+    // access -> function, variable or enum constant accessed
+    kind = "V" and
+    result = e.(Access).getTarget() and
+    not e.(Expr).isCompilerGenerated()
+    or
+    // macro access -> macro accessed
+    kind = "X" and
+    result = e.(MacroAccess).getMacro()
+    or
+    // type mention -> type
+    kind = "T" and
+    e.(TypeMention).getMentionedType() = result and
+    not constructorCallTypeMention(_, e) and // handled elsewhere
+    // Multiple type mentions can be generated when a typedef is used, and
+    // in such cases we want to exclude all but the originating typedef.
+    not exists(Type secondary |
+      exists(TypeMention tm, File f, int startline, int startcol |
+        typeMentionStartLoc(e, result, f, startline, startcol) and
+        typeMentionStartLoc(tm, secondary, f, startline, startcol) and
+        (
+          result = secondary.(TypedefType).getBaseType() or
+          result = secondary.(TypedefType).getBaseType().(SpecifiedType).getBaseType()
+        )
+      )
+    )
+    or
+    // constructor call -> function called
+    //  - but only if there is a corresponding type mention, since
+    //    we don't want links for implicit conversions.
+    //  - using the location of the type mention, since it's
+    //    tighter that the location of the function call.
+    kind = "M" and
+    exists(ConstructorCall cc |
+      constructorCallTypeMention(cc, e) and
+      result = cc.getTarget()
+    )
+    or
+    // include -> included file
+    kind = "I" and
+    result = e.(Include).getIncludedFile() and
+    // exclude `#include` directives containing macros
+    not exists(MacroInvocation mi, Location l1, Location l2 |
+      l1 = e.(Include).getLocation() and
+      l2 = mi.getLocation() and
+      l1.getContainer() = l2.getContainer() and
+      l1.getStartLine() = l2.getStartLine()
+      // (an #include directive must be always on it's own line)
+    )
+  ) and
+  (
+    // exclude things inside macro invocations, as they will overlap
+    // with the macro invocation.
+    not e.(Element).isInMacroExpansion() and
+    // exclude nested macro invocations, as they will overlap with
+    // the top macro invocation.
+    not exists(e.(MacroAccess).getParentInvocation())
+  ) and
+  // Some entities have many locations. This can arise for an external
+  // function that is frequently declared but not defined, or perhaps
+  // for a struct type that is declared in many places. Rather than
+  // letting the result set explode, we just exclude results that are
+  // "too ambiguous" -- we could also arbitrarily pick one location
+  // later on.
+  strictcount(result.getLocation()) < 10
+}

cpp/ql/lib/experimental/semmle/code/cpp/dataflow/ProductFlow.qll

@@ -0,0 +1,301 @@
+import experimental.semmle.code.cpp.ir.dataflow.DataFlow
+import experimental.semmle.code.cpp.ir.dataflow.DataFlow2
+
+module ProductFlow {
+  abstract class Configuration extends string {
+    bindingset[this]
+    Configuration() { any() }
+
+    /**
+     * Holds if `(source1, source2)` is a relevant data flow source.
+     *
+     * `source1` and `source2` must belong to the same callable.
+     */
+    predicate isSourcePair(DataFlow::Node source1, DataFlow::Node source2) { none() }
+
+    /**
+     * Holds if `(source1, source2)` is a relevant data flow source with initial states `state1`
+     * and `state2`, respectively.
+     *
+     * `source1` and `source2` must belong to the same callable.
+     */
+    predicate isSourcePair(
+      DataFlow::Node source1, DataFlow::FlowState state1, DataFlow::Node source2,
+      DataFlow::FlowState state2
+    ) {
+      state1 = "" and
+      state2 = "" and
+      this.isSourcePair(source1, source2)
+    }
+
+    /**
+     * Holds if `(sink1, sink2)` is a relevant data flow sink.
+     *
+     * `sink1` and `sink2` must belong to the same callable.
+     */
+    predicate isSinkPair(DataFlow::Node sink1, DataFlow::Node sink2) { none() }
+
+    /**
+     * Holds if `(sink1, sink2)` is a relevant data flow sink with final states `state1`
+     * and `state2`, respectively.
+     *
+     * `sink1` and `sink2` must belong to the same callable.
+     */
+    predicate isSinkPair(
+      DataFlow::Node sink1, DataFlow::FlowState state1, DataFlow::Node sink2,
+      DataFlow::FlowState state2
+    ) {
+      state1 = "" and
+      state2 = "" and
+      this.isSinkPair(sink1, sink2)
+    }
+
+    /**
+     * Holds if data flow through `node` is prohibited through the first projection of the product
+     * dataflow graph when the flow state is `state`.
+     */
+    predicate isBarrier1(DataFlow::Node node, DataFlow::FlowState state) {
+      this.isBarrier1(node) and state = ""
+    }
+
+    /**
+     * Holds if data flow through `node` is prohibited through the second projection of the product
+     * dataflow graph when the flow state is `state`.
+     */
+    predicate isBarrier2(DataFlow::Node node, DataFlow::FlowState state) {
+      this.isBarrier2(node) and state = ""
+    }
+
+    /**
+     * Holds if data flow through `node` is prohibited through the first projection of the product
+     * dataflow graph.
+     */
+    predicate isBarrier1(DataFlow::Node node) { none() }
+
+    /**
+     * Holds if data flow through `node` is prohibited through the second projection of the product
+     * dataflow graph.
+     */
+    predicate isBarrier2(DataFlow::Node node) { none() }
+
+    /**
+     * Holds if data flow out of `node` is prohibited in the first projection of the product
+     * dataflow graph.
+     */
+    predicate isBarrierOut1(DataFlow::Node node) { none() }
+
+    /**
+     * Holds if data flow out of `node` is prohibited in the second projection of the product
+     * dataflow graph.
+     */
+    predicate isBarrierOut2(DataFlow::Node node) { none() }
+
+    /*
+     * Holds if data may flow from `node1` to `node2` in addition to the normal data-flow steps in
+     * the first projection of the product dataflow graph.
+     */
+
+    predicate isAdditionalFlowStep1(DataFlow::Node node1, DataFlow::Node node2) { none() }
+
+    /**
+     * Holds if data may flow from `node1` to `node2` in addition to the normal data-flow steps in
+     * the first projection of the product dataflow graph.
+     *
+     * This step is only applicable in `state1` and updates the flow state to `state2`.
+     */
+    predicate isAdditionalFlowStep1(
+      DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+      DataFlow::FlowState state2
+    ) {
+      state1 instanceof DataFlow::FlowStateEmpty and
+      state2 instanceof DataFlow::FlowStateEmpty and
+      this.isAdditionalFlowStep1(node1, node2)
+    }
+
+    /**
+     * Holds if data may flow from `node1` to `node2` in addition to the normal data-flow steps in
+     * the second projection of the product dataflow graph.
+     */
+    predicate isAdditionalFlowStep2(DataFlow::Node node1, DataFlow::Node node2) { none() }
+
+    /**
+     * Holds if data may flow from `node1` to `node2` in addition to the normal data-flow steps in
+     * the second projection of the product dataflow graph.
+     *
+     * This step is only applicable in `state1` and updates the flow state to `state2`.
+     */
+    predicate isAdditionalFlowStep2(
+      DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+      DataFlow::FlowState state2
+    ) {
+      state1 instanceof DataFlow::FlowStateEmpty and
+      state2 instanceof DataFlow::FlowStateEmpty and
+      this.isAdditionalFlowStep2(node1, node2)
+    }
+
+    /**
+     * Holds if data flow into `node` is prohibited in the first projection of the product
+     * dataflow graph.
+     */
+    predicate isBarrierIn1(DataFlow::Node node) { none() }
+
+    /**
+     * Holds if data flow into `node` is prohibited in the second projection of the product
+     * dataflow graph.
+     */
+    predicate isBarrierIn2(DataFlow::Node node) { none() }
+
+    predicate hasFlowPath(
+      DataFlow::PathNode source1, DataFlow2::PathNode source2, DataFlow::PathNode sink1,
+      DataFlow2::PathNode sink2
+    ) {
+      reachable(this, source1, source2, sink1, sink2)
+    }
+  }
+
+  private import Internal
+
+  module Internal {
+    class Conf1 extends DataFlow::Configuration {
+      Conf1() { this = "Conf1" }
+
+      override predicate isSource(DataFlow::Node source, DataFlow::FlowState state) {
+        exists(Configuration conf | conf.isSourcePair(source, state, _, _))
+      }
+
+      override predicate isSink(DataFlow::Node sink, DataFlow::FlowState state) {
+        exists(Configuration conf | conf.isSinkPair(sink, state, _, _))
+      }
+
+      override predicate isBarrier(DataFlow::Node node, DataFlow::FlowState state) {
+        exists(Configuration conf | conf.isBarrier1(node, state))
+      }
+
+      override predicate isBarrierOut(DataFlow::Node node) {
+        exists(Configuration conf | conf.isBarrierOut1(node))
+      }
+
+      override predicate isAdditionalFlowStep(
+        DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+        DataFlow::FlowState state2
+      ) {
+        exists(Configuration conf | conf.isAdditionalFlowStep1(node1, state1, node2, state2))
+      }
+
+      override predicate isBarrierIn(DataFlow::Node node) {
+        exists(Configuration conf | conf.isBarrierIn1(node))
+      }
+    }
+
+    class Conf2 extends DataFlow2::Configuration {
+      Conf2() { this = "Conf2" }
+
+      override predicate isSource(DataFlow::Node source, DataFlow::FlowState state) {
+        exists(Configuration conf, DataFlow::PathNode source1 |
+          conf.isSourcePair(source1.getNode(), source1.getState(), source, state) and
+          any(Conf1 c).hasFlowPath(source1, _)
+        )
+      }
+
+      override predicate isSink(DataFlow::Node sink, DataFlow::FlowState state) {
+        exists(Configuration conf, DataFlow::PathNode sink1 |
+          conf.isSinkPair(sink1.getNode(), sink1.getState(), sink, state) and
+          any(Conf1 c).hasFlowPath(_, sink1)
+        )
+      }
+
+      override predicate isBarrier(DataFlow::Node node, DataFlow::FlowState state) {
+        exists(Configuration conf | conf.isBarrier2(node, state))
+      }
+
+      override predicate isBarrierOut(DataFlow::Node node) {
+        exists(Configuration conf | conf.isBarrierOut2(node))
+      }
+
+      override predicate isAdditionalFlowStep(
+        DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+        DataFlow::FlowState state2
+      ) {
+        exists(Configuration conf | conf.isAdditionalFlowStep2(node1, state1, node2, state2))
+      }
+
+      override predicate isBarrierIn(DataFlow::Node node) {
+        exists(Configuration conf | conf.isBarrierIn2(node))
+      }
+    }
+  }
+
+  pragma[nomagic]
+  private predicate reachableInterprocEntry(
+    Configuration conf, DataFlow::PathNode source1, DataFlow2::PathNode source2,
+    DataFlow::PathNode node1, DataFlow2::PathNode node2
+  ) {
+    conf.isSourcePair(node1.getNode(), node1.getState(), node2.getNode(), node2.getState()) and
+    node1 = source1 and
+    node2 = source2
+    or
+    exists(
+      DataFlow::PathNode midEntry1, DataFlow2::PathNode midEntry2, DataFlow::PathNode midExit1,
+      DataFlow2::PathNode midExit2
+    |
+      reachableInterprocEntry(conf, source1, source2, midEntry1, midEntry2) and
+      interprocEdgePair(midExit1, midExit2, node1, node2) and
+      localPathStep1*(midEntry1, midExit1) and
+      localPathStep2*(midEntry2, midExit2)
+    )
+  }
+
+  private predicate localPathStep1(DataFlow::PathNode pred, DataFlow::PathNode succ) {
+    DataFlow::PathGraph::edges(pred, succ) and
+    pragma[only_bind_out](pred.getNode().getEnclosingCallable()) =
+      pragma[only_bind_out](succ.getNode().getEnclosingCallable())
+  }
+
+  private predicate localPathStep2(DataFlow2::PathNode pred, DataFlow2::PathNode succ) {
+    DataFlow2::PathGraph::edges(pred, succ) and
+    pragma[only_bind_out](pred.getNode().getEnclosingCallable()) =
+      pragma[only_bind_out](succ.getNode().getEnclosingCallable())
+  }
+
+  pragma[nomagic]
+  private predicate interprocEdge1(
+    Declaration predDecl, Declaration succDecl, DataFlow::PathNode pred1, DataFlow::PathNode succ1
+  ) {
+    DataFlow::PathGraph::edges(pred1, succ1) and
+    predDecl != succDecl and
+    pred1.getNode().getEnclosingCallable() = predDecl and
+    succ1.getNode().getEnclosingCallable() = succDecl
+  }
+
+  pragma[nomagic]
+  private predicate interprocEdge2(
+    Declaration predDecl, Declaration succDecl, DataFlow2::PathNode pred2, DataFlow2::PathNode succ2
+  ) {
+    DataFlow2::PathGraph::edges(pred2, succ2) and
+    predDecl != succDecl and
+    pred2.getNode().getEnclosingCallable() = predDecl and
+    succ2.getNode().getEnclosingCallable() = succDecl
+  }
+
+  private predicate interprocEdgePair(
+    DataFlow::PathNode pred1, DataFlow2::PathNode pred2, DataFlow::PathNode succ1,
+    DataFlow2::PathNode succ2
+  ) {
+    exists(Declaration predDecl, Declaration succDecl |
+      interprocEdge1(predDecl, succDecl, pred1, succ1) and
+      interprocEdge2(predDecl, succDecl, pred2, succ2)
+    )
+  }
+
+  private predicate reachable(
+    Configuration conf, DataFlow::PathNode source1, DataFlow2::PathNode source2,
+    DataFlow::PathNode sink1, DataFlow2::PathNode sink2
+  ) {
+    exists(DataFlow::PathNode mid1, DataFlow2::PathNode mid2 |
+      reachableInterprocEntry(conf, source1, source2, mid1, mid2) and
+      conf.isSinkPair(sink1.getNode(), sink1.getState(), sink2.getNode(), sink2.getState()) and
+      localPathStep1*(mid1, sink1) and
+      localPathStep2*(mid2, sink2)
+    )
+  }
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/DataFlow.qll

@@ -0,0 +1,26 @@
+/**
+ * Provides a library for local (intra-procedural) and global (inter-procedural)
+ * data flow analysis: deciding whether data can flow from a _source_ to a
+ * _sink_. This library differs from the one in `semmle.code.cpp.dataflow` in that
+ * this library uses the IR (Intermediate Representation) library, which provides
+ * a more precise semantic representation of the program, whereas the other dataflow
+ * library uses the more syntax-oriented ASTs. This library should provide more accurate
+ * results than the AST-based library in most scenarios.
+ *
+ * Unless configured otherwise, _flow_ means that the exact value of
+ * the source may reach the sink. We do not track flow across pointer
+ * dereferences or array indexing.
+ *
+ * To use global (interprocedural) data flow, extend the class
+ * `DataFlow::Configuration` as documented on that class. To use local
+ * (intraprocedural) data flow between expressions, call
+ * `DataFlow::localExprFlow`. For more general cases of local data flow, call
+ * `DataFlow::localFlow` or `DataFlow::localFlowStep` with arguments of type
+ * `DataFlow::Node`.
+ */
+
+import cpp
+
+module DataFlow {
+  import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowImpl
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/DataFlow2.qll

@@ -0,0 +1,16 @@
+/**
+ * Provides a `DataFlow2` module, which is a copy of the `DataFlow` module. Use
+ * this class when data-flow configurations must depend on each other. Two
+ * classes extending `DataFlow::Configuration` should never depend on each
+ * other, but one of them should instead depend on a
+ * `DataFlow2::Configuration`, a `DataFlow3::Configuration`, or a
+ * `DataFlow4::Configuration`.
+ *
+ * See `semmle.code.cpp.ir.dataflow.DataFlow` for the full documentation.
+ */
+
+import cpp
+
+module DataFlow2 {
+  import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowImpl2
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/DataFlow3.qll

@@ -0,0 +1,16 @@
+/**
+ * Provides a `DataFlow3` module, which is a copy of the `DataFlow` module. Use
+ * this class when data-flow configurations must depend on each other. Two
+ * classes extending `DataFlow::Configuration` should never depend on each
+ * other, but one of them should instead depend on a
+ * `DataFlow2::Configuration`, a `DataFlow3::Configuration`, or a
+ * `DataFlow4::Configuration`.
+ *
+ * See `semmle.code.cpp.ir.dataflow.DataFlow` for the full documentation.
+ */
+
+import cpp
+
+module DataFlow3 {
+  import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowImpl3
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/DataFlow4.qll

@@ -0,0 +1,16 @@
+/**
+ * Provides a `DataFlow4` module, which is a copy of the `DataFlow` module. Use
+ * this class when data-flow configurations must depend on each other. Two
+ * classes extending `DataFlow::Configuration` should never depend on each
+ * other, but one of them should instead depend on a
+ * `DataFlow2::Configuration`, a `DataFlow3::Configuration`, or a
+ * `DataFlow4::Configuration`.
+ *
+ * See `semmle.code.cpp.ir.dataflow.DataFlow` for the full documentation.
+ */
+
+import cpp
+
+module DataFlow4 {
+  import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowImpl4
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/ResolveCall.qll

@@ -0,0 +1,23 @@
+/**
+ * Provides a predicate for non-contextual virtual dispatch and function
+ * pointer resolution.
+ */
+
+import cpp
+private import semmle.code.cpp.ir.ValueNumbering
+private import internal.DataFlowDispatch
+private import semmle.code.cpp.ir.IR
+
+/**
+ * Resolve potential target function(s) for `call`.
+ *
+ * If `call` is a call through a function pointer (`ExprCall`) or its target is
+ * a virtual member function, simple data flow analysis is performed in order
+ * to identify the possible target(s).
+ */
+Function resolveCall(Call call) {
+  exists(CallInstruction callInstruction |
+    callInstruction.getAst() = call and
+    result = viableCallable(callInstruction)
+  )
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/TaintTracking.qll

@@ -0,0 +1,23 @@
+/**
+ * Provides classes for performing local (intra-procedural) and
+ * global (inter-procedural) taint-tracking analyses.
+ *
+ * We define _taint propagation_ informally to mean that a substantial part of
+ * the information from the source is preserved at the sink. For example, taint
+ * propagates from `x` to `x + 100`, but it does not propagate from `x` to `x >
+ * 100` since we consider a single bit of information to be too little.
+ *
+ * To use global (interprocedural) taint tracking, extend the class
+ * `TaintTracking::Configuration` as documented on that class. To use local
+ * (intraprocedural) taint tracking between expressions, call
+ * `TaintTracking::localExprTaint`. For more general cases of local taint
+ * tracking, call `TaintTracking::localTaint` or
+ * `TaintTracking::localTaintStep` with arguments of type `DataFlow::Node`.
+ */
+
+import semmle.code.cpp.ir.dataflow.DataFlow
+import semmle.code.cpp.ir.dataflow.DataFlow2
+
+module TaintTracking {
+  import experimental.semmle.code.cpp.ir.dataflow.internal.tainttracking1.TaintTrackingImpl
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/TaintTracking2.qll

@@ -0,0 +1,15 @@
+/**
+ * Provides a `TaintTracking2` module, which is a copy of the `TaintTracking`
+ * module. Use this class when data-flow configurations or taint-tracking
+ * configurations must depend on each other. Two classes extending
+ * `DataFlow::Configuration` should never depend on each other, but one of them
+ * should instead depend on a `DataFlow2::Configuration`, a
+ * `DataFlow3::Configuration`, or a `DataFlow4::Configuration`. The
+ * `TaintTracking::Configuration` class extends `DataFlow::Configuration`, and
+ * `TaintTracking2::Configuration` extends `DataFlow2::Configuration`.
+ *
+ * See `semmle.code.cpp.ir.dataflow.TaintTracking` for the full documentation.
+ */
+module TaintTracking2 {
+  import experimental.semmle.code.cpp.ir.dataflow.internal.tainttracking2.TaintTrackingImpl
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/TaintTracking3.qll

@@ -0,0 +1,15 @@
+/**
+ * Provides a `TaintTracking3` module, which is a copy of the `TaintTracking`
+ * module. Use this class when data-flow configurations or taint-tracking
+ * configurations must depend on each other. Two classes extending
+ * `DataFlow::Configuration` should never depend on each other, but one of them
+ * should instead depend on a `DataFlow2::Configuration`, a
+ * `DataFlow3::Configuration`, or a `DataFlow4::Configuration`. The
+ * `TaintTracking::Configuration` class extends `DataFlow::Configuration`, and
+ * `TaintTracking2::Configuration` extends `DataFlow2::Configuration`.
+ *
+ * See `semmle.code.cpp.ir.dataflow.TaintTracking` for the full documentation.
+ */
+module TaintTracking3 {
+  import experimental.semmle.code.cpp.ir.dataflow.internal.tainttracking3.TaintTrackingImpl
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/DataFlowDispatch.qll

@@ -0,0 +1,273 @@
+private import cpp
+private import semmle.code.cpp.ir.IR
+private import experimental.semmle.code.cpp.ir.dataflow.DataFlow
+private import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowPrivate
+private import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowUtil
+private import DataFlowImplCommon as DataFlowImplCommon
+
+/**
+ * Gets a function that might be called by `call`.
+ */
+cached
+Function viableCallable(CallInstruction call) {
+  DataFlowImplCommon::forceCachingInSameStage() and
+  result = call.getStaticCallTarget()
+  or
+  // If the target of the call does not have a body in the snapshot, it might
+  // be because the target is just a header declaration, and the real target
+  // will be determined at run time when the caller and callee are linked
+  // together by the operating system's dynamic linker. In case a _unique_
+  // function with the right signature is present in the database, we return
+  // that as a potential callee.
+  exists(string qualifiedName, int nparams |
+    callSignatureWithoutBody(qualifiedName, nparams, call) and
+    functionSignatureWithBody(qualifiedName, nparams, result) and
+    strictcount(Function other | functionSignatureWithBody(qualifiedName, nparams, other)) = 1
+  )
+  or
+  // Virtual dispatch
+  result = call.(VirtualDispatch::DataSensitiveCall).resolve()
+}
+
+/**
+ * Provides virtual dispatch support compatible with the original
+ * implementation of `semmle.code.cpp.security.TaintTracking`.
+ */
+private module VirtualDispatch {
+  /** A call that may dispatch differently depending on the qualifier value. */
+  abstract class DataSensitiveCall extends DataFlowCall {
+    /**
+     * Gets the node whose value determines the target of this call. This node
+     * could be the qualifier of a virtual dispatch or the function-pointer
+     * expression in a call to a function pointer. What they have in common is
+     * that we need to find out which data flows there, and then it's up to the
+     * `resolve` predicate to stitch that information together and resolve the
+     * call.
+     */
+    abstract DataFlow::Node getDispatchValue();
+
+    /** Gets a candidate target for this call. */
+    abstract Function resolve();
+
+    /**
+     * Whether `src` can flow to this call.
+     *
+     * Searches backwards from `getDispatchValue()` to `src`. The `allowFromArg`
+     * parameter is true when the search is allowed to continue backwards into
+     * a parameter; non-recursive callers should pass `_` for `allowFromArg`.
+     */
+    predicate flowsFrom(DataFlow::Node src, boolean allowFromArg) {
+      src = this.getDispatchValue() and allowFromArg = true
+      or
+      exists(DataFlow::Node other, boolean allowOtherFromArg |
+        this.flowsFrom(other, allowOtherFromArg)
+      |
+        // Call argument
+        exists(DataFlowCall call, Position i |
+          other
+              .(DataFlow::ParameterNode)
+              .isParameterOf(pragma[only_bind_into](call).getStaticCallTarget(), i) and
+          src.(ArgumentNode).argumentOf(call, pragma[only_bind_into](pragma[only_bind_out](i)))
+        ) and
+        allowOtherFromArg = true and
+        allowFromArg = true
+        or
+        // Call return
+        exists(DataFlowCall call, ReturnKind returnKind |
+          other = getAnOutNode(call, returnKind) and
+          returnNodeWithKindAndEnclosingCallable(src, returnKind, call.getStaticCallTarget())
+        ) and
+        allowFromArg = false
+        or
+        // Local flow
+        DataFlow::localFlowStep(src, other) and
+        allowFromArg = allowOtherFromArg
+        or
+        // Flow from global variable to load.
+        exists(LoadInstruction load, GlobalOrNamespaceVariable var |
+          var = src.asVariable() and
+          other.asInstruction() = load and
+          addressOfGlobal(load.getSourceAddress(), var) and
+          // The `allowFromArg` concept doesn't play a role when `src` is a
+          // global variable, so we just set it to a single arbitrary value for
+          // performance.
+          allowFromArg = true
+        )
+        or
+        // Flow from store to global variable.
+        exists(StoreInstruction store, GlobalOrNamespaceVariable var |
+          var = other.asVariable() and
+          store = src.asInstruction() and
+          storeIntoGlobal(store, var) and
+          // Setting `allowFromArg` to `true` like in the base case means we
+          // treat a store to a global variable like the dispatch itself: flow
+          // may come from anywhere.
+          allowFromArg = true
+        )
+      )
+    }
+  }
+
+  pragma[noinline]
+  private predicate storeIntoGlobal(StoreInstruction store, GlobalOrNamespaceVariable var) {
+    addressOfGlobal(store.getDestinationAddress(), var)
+  }
+
+  /** Holds if `addressInstr` is an instruction that produces the address of `var`. */
+  private predicate addressOfGlobal(Instruction addressInstr, GlobalOrNamespaceVariable var) {
+    // Access directly to the global variable
+    addressInstr.(VariableAddressInstruction).getAstVariable() = var
+    or
+    // Access to a field on a global union
+    exists(FieldAddressInstruction fa |
+      fa = addressInstr and
+      fa.getObjectAddress().(VariableAddressInstruction).getAstVariable() = var and
+      fa.getField().getDeclaringType() instanceof Union
+    )
+  }
+
+  /**
+   * A ReturnNode with its ReturnKind and its enclosing callable.
+   *
+   * Used to fix a join ordering issue in flowsFrom.
+   */
+  pragma[noinline]
+  private predicate returnNodeWithKindAndEnclosingCallable(
+    ReturnNode node, ReturnKind kind, DataFlowCallable callable
+  ) {
+    node.getKind() = kind and
+    node.getEnclosingCallable() = callable
+  }
+
+  /** Call through a function pointer. */
+  private class DataSensitiveExprCall extends DataSensitiveCall {
+    DataSensitiveExprCall() { not exists(this.getStaticCallTarget()) }
+
+    override DataFlow::Node getDispatchValue() { result.asInstruction() = this.getCallTarget() }
+
+    override Function resolve() {
+      exists(FunctionInstruction fi |
+        this.flowsFrom(DataFlow::instructionNode(fi), _) and
+        result = fi.getFunctionSymbol()
+      ) and
+      (
+        this.getNumberOfArguments() <= result.getEffectiveNumberOfParameters() and
+        this.getNumberOfArguments() >= result.getEffectiveNumberOfParameters()
+        or
+        result.isVarargs()
+      )
+    }
+  }
+
+  /** Call to a virtual function. */
+  private class DataSensitiveOverriddenFunctionCall extends DataSensitiveCall {
+    DataSensitiveOverriddenFunctionCall() {
+      exists(this.getStaticCallTarget().(VirtualFunction).getAnOverridingFunction())
+    }
+
+    override DataFlow::Node getDispatchValue() { result.asInstruction() = this.getThisArgument() }
+
+    override MemberFunction resolve() {
+      exists(Class overridingClass |
+        this.overrideMayAffectCall(overridingClass, result) and
+        this.hasFlowFromCastFrom(overridingClass)
+      )
+    }
+
+    /**
+     * Holds if `this` is a virtual function call whose static target is
+     * overridden by `overridingFunction` in `overridingClass`.
+     */
+    pragma[noinline]
+    private predicate overrideMayAffectCall(Class overridingClass, MemberFunction overridingFunction) {
+      overridingFunction.getAnOverriddenFunction+() = this.getStaticCallTarget().(VirtualFunction) and
+      overridingFunction.getDeclaringType() = overridingClass
+    }
+
+    /**
+     * Holds if the qualifier of `this` has flow from an upcast from
+     * `derivedClass`.
+     */
+    pragma[noinline]
+    private predicate hasFlowFromCastFrom(Class derivedClass) {
+      exists(ConvertToBaseInstruction toBase |
+        this.flowsFrom(DataFlow::instructionNode(toBase), _) and
+        derivedClass = toBase.getDerivedClass()
+      )
+    }
+  }
+}
+
+/**
+ * Holds if `f` is a function with a body that has name `qualifiedName` and
+ * `nparams` parameter count. See `functionSignature`.
+ */
+private predicate functionSignatureWithBody(string qualifiedName, int nparams, Function f) {
+  functionSignature(f, qualifiedName, nparams) and
+  exists(f.getBlock())
+}
+
+/**
+ * Holds if the target of `call` is a function _with no definition_ that has
+ * name `qualifiedName` and `nparams` parameter count. See `functionSignature`.
+ */
+pragma[noinline]
+private predicate callSignatureWithoutBody(string qualifiedName, int nparams, CallInstruction call) {
+  exists(Function target |
+    target = call.getStaticCallTarget() and
+    not exists(target.getBlock()) and
+    functionSignature(target, qualifiedName, nparams)
+  )
+}
+
+/**
+ * Holds if `f` has name `qualifiedName` and `nparams` parameter count. This is
+ * an approximation of its signature for the purpose of matching functions that
+ * might be the same across link targets.
+ */
+private predicate functionSignature(Function f, string qualifiedName, int nparams) {
+  qualifiedName = f.getQualifiedName() and
+  nparams = f.getNumberOfParameters() and
+  not f.isStatic()
+}
+
+/**
+ * Holds if the set of viable implementations that can be called by `call`
+ * might be improved by knowing the call context.
+ */
+predicate mayBenefitFromCallContext(CallInstruction call, Function f) {
+  mayBenefitFromCallContext(call, f, _)
+}
+
+/**
+ * Holds if `call` is a call through a function pointer, and the pointer
+ * value is given as the `arg`'th argument to `f`.
+ */
+private predicate mayBenefitFromCallContext(
+  VirtualDispatch::DataSensitiveCall call, Function f, int arg
+) {
+  f = pragma[only_bind_out](call).getEnclosingCallable() and
+  exists(InitializeParameterInstruction init |
+    not exists(call.getStaticCallTarget()) and
+    init.getEnclosingFunction() = f and
+    call.flowsFrom(DataFlow::instructionNode(init), _) and
+    init.getParameter().getIndex() = arg
+  )
+}
+
+/**
+ * Gets a viable dispatch target of `call` in the context `ctx`. This is
+ * restricted to those `call`s for which a context might make a difference.
+ */
+Function viableImplInCallContext(CallInstruction call, CallInstruction ctx) {
+  result = viableCallable(call) and
+  exists(int i, Function f |
+    mayBenefitFromCallContext(pragma[only_bind_into](call), f, i) and
+    f = ctx.getStaticCallTarget() and
+    result = ctx.getArgument(i).getUnconvertedResultExpression().(FunctionAccess).getTarget()
+  )
+}
+
+/** Holds if arguments at position `apos` match parameters at position `ppos`. */
+pragma[inline]
+predicate parameterMatch(ParameterPosition ppos, ArgumentPosition apos) { ppos = apos }

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/DataFlowImplConsistency.qll

@@ -0,0 +1,247 @@
+/**
+ * Provides consistency queries for checking invariants in the language-specific
+ * data-flow classes and predicates.
+ */
+
+private import DataFlowImplSpecific::Private
+private import DataFlowImplSpecific::Public
+private import tainttracking1.TaintTrackingParameter::Private
+private import tainttracking1.TaintTrackingParameter::Public
+
+module Consistency {
+  private newtype TConsistencyConfiguration = MkConsistencyConfiguration()
+
+  /** A class for configuring the consistency queries. */
+  class ConsistencyConfiguration extends TConsistencyConfiguration {
+    string toString() { none() }
+
+    /** Holds if `n` should be excluded from the consistency test `uniqueEnclosingCallable`. */
+    predicate uniqueEnclosingCallableExclude(Node n) { none() }
+
+    /** Holds if `n` should be excluded from the consistency test `uniqueNodeLocation`. */
+    predicate uniqueNodeLocationExclude(Node n) { none() }
+
+    /** Holds if `n` should be excluded from the consistency test `missingLocation`. */
+    predicate missingLocationExclude(Node n) { none() }
+
+    /** Holds if `n` should be excluded from the consistency test `postWithInFlow`. */
+    predicate postWithInFlowExclude(Node n) { none() }
+
+    /** Holds if `n` should be excluded from the consistency test `argHasPostUpdate`. */
+    predicate argHasPostUpdateExclude(ArgumentNode n) { none() }
+
+    /** Holds if `n` should be excluded from the consistency test `reverseRead`. */
+    predicate reverseReadExclude(Node n) { none() }
+
+    /** Holds if `n` should be excluded from the consistency test `postHasUniquePre`. */
+    predicate postHasUniquePreExclude(PostUpdateNode n) { none() }
+
+    /** Holds if `n` should be excluded from the consistency test `uniquePostUpdate`. */
+    predicate uniquePostUpdateExclude(Node n) { none() }
+
+    /** Holds if `(call, ctx)` should be excluded from the consistency test `viableImplInCallContextTooLargeExclude`. */
+    predicate viableImplInCallContextTooLargeExclude(
+      DataFlowCall call, DataFlowCall ctx, DataFlowCallable callable
+    ) {
+      none()
+    }
+  }
+
+  private class RelevantNode extends Node {
+    RelevantNode() {
+      this instanceof ArgumentNode or
+      this instanceof ParameterNode or
+      this instanceof ReturnNode or
+      this = getAnOutNode(_, _) or
+      simpleLocalFlowStep(this, _) or
+      simpleLocalFlowStep(_, this) or
+      jumpStep(this, _) or
+      jumpStep(_, this) or
+      storeStep(this, _, _) or
+      storeStep(_, _, this) or
+      readStep(this, _, _) or
+      readStep(_, _, this) or
+      defaultAdditionalTaintStep(this, _) or
+      defaultAdditionalTaintStep(_, this)
+    }
+  }
+
+  query predicate uniqueEnclosingCallable(Node n, string msg) {
+    exists(int c |
+      n instanceof RelevantNode and
+      c = count(nodeGetEnclosingCallable(n)) and
+      c != 1 and
+      not any(ConsistencyConfiguration conf).uniqueEnclosingCallableExclude(n) and
+      msg = "Node should have one enclosing callable but has " + c + "."
+    )
+  }
+
+  query predicate uniqueType(Node n, string msg) {
+    exists(int c |
+      n instanceof RelevantNode and
+      c = count(getNodeType(n)) and
+      c != 1 and
+      msg = "Node should have one type but has " + c + "."
+    )
+  }
+
+  query predicate uniqueNodeLocation(Node n, string msg) {
+    exists(int c |
+      c =
+        count(string filepath, int startline, int startcolumn, int endline, int endcolumn |
+          n.hasLocationInfo(filepath, startline, startcolumn, endline, endcolumn)
+        ) and
+      c != 1 and
+      not any(ConsistencyConfiguration conf).uniqueNodeLocationExclude(n) and
+      msg = "Node should have one location but has " + c + "."
+    )
+  }
+
+  query predicate missingLocation(string msg) {
+    exists(int c |
+      c =
+        strictcount(Node n |
+          not exists(string filepath, int startline, int startcolumn, int endline, int endcolumn |
+            n.hasLocationInfo(filepath, startline, startcolumn, endline, endcolumn)
+          ) and
+          not any(ConsistencyConfiguration conf).missingLocationExclude(n)
+        ) and
+      msg = "Nodes without location: " + c
+    )
+  }
+
+  query predicate uniqueNodeToString(Node n, string msg) {
+    exists(int c |
+      c = count(n.toString()) and
+      c != 1 and
+      msg = "Node should have one toString but has " + c + "."
+    )
+  }
+
+  query predicate missingToString(string msg) {
+    exists(int c |
+      c = strictcount(Node n | not exists(n.toString())) and
+      msg = "Nodes without toString: " + c
+    )
+  }
+
+  query predicate parameterCallable(ParameterNode p, string msg) {
+    exists(DataFlowCallable c | isParameterNode(p, c, _) and c != nodeGetEnclosingCallable(p)) and
+    msg = "Callable mismatch for parameter."
+  }
+
+  query predicate localFlowIsLocal(Node n1, Node n2, string msg) {
+    simpleLocalFlowStep(n1, n2) and
+    nodeGetEnclosingCallable(n1) != nodeGetEnclosingCallable(n2) and
+    msg = "Local flow step does not preserve enclosing callable."
+  }
+
+  query predicate readStepIsLocal(Node n1, Node n2, string msg) {
+    readStep(n1, _, n2) and
+    nodeGetEnclosingCallable(n1) != nodeGetEnclosingCallable(n2) and
+    msg = "Read step does not preserve enclosing callable."
+  }
+
+  query predicate storeStepIsLocal(Node n1, Node n2, string msg) {
+    storeStep(n1, _, n2) and
+    nodeGetEnclosingCallable(n1) != nodeGetEnclosingCallable(n2) and
+    msg = "Store step does not preserve enclosing callable."
+  }
+
+  private DataFlowType typeRepr() { result = getNodeType(_) }
+
+  query predicate compatibleTypesReflexive(DataFlowType t, string msg) {
+    t = typeRepr() and
+    not compatibleTypes(t, t) and
+    msg = "Type compatibility predicate is not reflexive."
+  }
+
+  query predicate unreachableNodeCCtx(Node n, DataFlowCall call, string msg) {
+    isUnreachableInCall(n, call) and
+    exists(DataFlowCallable c |
+      c = nodeGetEnclosingCallable(n) and
+      not viableCallable(call) = c
+    ) and
+    msg = "Call context for isUnreachableInCall is inconsistent with call graph."
+  }
+
+  query predicate localCallNodes(DataFlowCall call, Node n, string msg) {
+    (
+      n = getAnOutNode(call, _) and
+      msg = "OutNode and call does not share enclosing callable."
+      or
+      n.(ArgumentNode).argumentOf(call, _) and
+      msg = "ArgumentNode and call does not share enclosing callable."
+    ) and
+    nodeGetEnclosingCallable(n) != call.getEnclosingCallable()
+  }
+
+  // This predicate helps the compiler forget that in some languages
+  // it is impossible for a result of `getPreUpdateNode` to be an
+  // instance of `PostUpdateNode`.
+  private Node getPre(PostUpdateNode n) {
+    result = n.getPreUpdateNode()
+    or
+    none()
+  }
+
+  query predicate postIsNotPre(PostUpdateNode n, string msg) {
+    getPre(n) = n and
+    msg = "PostUpdateNode should not equal its pre-update node."
+  }
+
+  query predicate postHasUniquePre(PostUpdateNode n, string msg) {
+    not any(ConsistencyConfiguration conf).postHasUniquePreExclude(n) and
+    exists(int c |
+      c = count(n.getPreUpdateNode()) and
+      c != 1 and
+      msg = "PostUpdateNode should have one pre-update node but has " + c + "."
+    )
+  }
+
+  query predicate uniquePostUpdate(Node n, string msg) {
+    not any(ConsistencyConfiguration conf).uniquePostUpdateExclude(n) and
+    1 < strictcount(PostUpdateNode post | post.getPreUpdateNode() = n) and
+    msg = "Node has multiple PostUpdateNodes."
+  }
+
+  query predicate postIsInSameCallable(PostUpdateNode n, string msg) {
+    nodeGetEnclosingCallable(n) != nodeGetEnclosingCallable(n.getPreUpdateNode()) and
+    msg = "PostUpdateNode does not share callable with its pre-update node."
+  }
+
+  private predicate hasPost(Node n) { exists(PostUpdateNode post | post.getPreUpdateNode() = n) }
+
+  query predicate reverseRead(Node n, string msg) {
+    exists(Node n2 | readStep(n, _, n2) and hasPost(n2) and not hasPost(n)) and
+    not any(ConsistencyConfiguration conf).reverseReadExclude(n) and
+    msg = "Origin of readStep is missing a PostUpdateNode."
+  }
+
+  query predicate argHasPostUpdate(ArgumentNode n, string msg) {
+    not hasPost(n) and
+    not any(ConsistencyConfiguration c).argHasPostUpdateExclude(n) and
+    msg = "ArgumentNode is missing PostUpdateNode."
+  }
+
+  // This predicate helps the compiler forget that in some languages
+  // it is impossible for a `PostUpdateNode` to be the target of
+  // `simpleLocalFlowStep`.
+  private predicate isPostUpdateNode(Node n) { n instanceof PostUpdateNode or none() }
+
+  query predicate postWithInFlow(Node n, string msg) {
+    isPostUpdateNode(n) and
+    not clearsContent(n, _) and
+    simpleLocalFlowStep(_, n) and
+    not any(ConsistencyConfiguration c).postWithInFlowExclude(n) and
+    msg = "PostUpdateNode should not be the target of local flow."
+  }
+
+  query predicate viableImplInCallContextTooLarge(
+    DataFlowCall call, DataFlowCall ctx, DataFlowCallable callable
+  ) {
+    callable = viableImplInCallContext(call, ctx) and
+    not callable = viableCallable(call) and
+    not any(ConsistencyConfiguration c).viableImplInCallContextTooLargeExclude(call, ctx, callable)
+  }
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/DataFlowImplSpecific.qll

@@ -0,0 +1,11 @@
+/**
+ * Provides IR-specific definitions for use in the data flow library.
+ */
+module Private {
+  import DataFlowPrivate
+  import DataFlowDispatch
+}
+
+module Public {
+  import DataFlowUtil
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/DataFlowPrivate.qll

@@ -0,0 +1,560 @@
+private import cpp as Cpp
+private import DataFlowUtil
+private import semmle.code.cpp.ir.IR
+private import DataFlowDispatch
+private import DataFlowImplConsistency
+private import semmle.code.cpp.ir.internal.IRCppLanguage
+private import SsaInternals as Ssa
+
+/** Gets the callable in which this node occurs. */
+DataFlowCallable nodeGetEnclosingCallable(Node n) { result = n.getEnclosingCallable() }
+
+/** Holds if `p` is a `ParameterNode` of `c` with position `pos`. */
+predicate isParameterNode(ParameterNode p, DataFlowCallable c, ParameterPosition pos) {
+  p.isParameterOf(c, pos)
+}
+
+/** Holds if `arg` is an `ArgumentNode` of `c` with position `pos`. */
+predicate isArgumentNode(ArgumentNode arg, DataFlowCall c, ArgumentPosition pos) {
+  arg.argumentOf(c, pos)
+}
+
+/**
+ * A data flow node that occurs as the argument of a call and is passed as-is
+ * to the callable. Instance arguments (`this` pointer) and read side effects
+ * on parameters are also included.
+ */
+abstract class ArgumentNode extends Node {
+  /**
+   * Holds if this argument occurs at the given position in the given call.
+   * The instance argument is considered to have index `-1`.
+   */
+  abstract predicate argumentOf(DataFlowCall call, ArgumentPosition pos);
+
+  /** Gets the call in which this node is an argument. */
+  DataFlowCall getCall() { this.argumentOf(result, _) }
+}
+
+/**
+ * A data flow node that occurs as the argument to a call, or an
+ * implicit `this` pointer argument.
+ */
+private class PrimaryArgumentNode extends ArgumentNode, OperandNode {
+  override ArgumentOperand op;
+
+  PrimaryArgumentNode() { exists(CallInstruction call | op = call.getAnArgumentOperand()) }
+
+  override predicate argumentOf(DataFlowCall call, ArgumentPosition pos) {
+    op = call.getArgumentOperand(pos.(DirectPosition).getIndex())
+  }
+
+  override string toStringImpl() { result = argumentOperandToString(op) }
+}
+
+private string argumentOperandToString(ArgumentOperand op) {
+  exists(Expr unconverted |
+    unconverted = op.getDef().getUnconvertedResultExpression() and
+    result = unconverted.toString()
+  )
+  or
+  // Certain instructions don't map to an unconverted result expression. For these cases
+  // we fall back to a simpler naming scheme. This can happen in IR-generated constructors.
+  not exists(op.getDef().getUnconvertedResultExpression()) and
+  (
+    result = "Argument " + op.(PositionalArgumentOperand).getIndex()
+    or
+    op instanceof ThisArgumentOperand and result = "Argument this"
+  )
+}
+
+private class SideEffectArgumentNode extends ArgumentNode, SideEffectOperandNode {
+  override predicate argumentOf(DataFlowCall dfCall, ArgumentPosition pos) {
+    this.getCallInstruction() = dfCall and
+    pos.(IndirectionPosition).getArgumentIndex() = this.getArgumentIndex() and
+    pos.(IndirectionPosition).getIndirectionIndex() = super.getIndirectionIndex()
+  }
+
+  override string toStringImpl() {
+    result = argumentOperandToString(this.getAddressOperand()) + " indirection"
+  }
+}
+
+/** A parameter position represented by an integer. */
+class ParameterPosition = Position;
+
+/** An argument position represented by an integer. */
+class ArgumentPosition = Position;
+
+class Position extends TPosition {
+  abstract string toString();
+}
+
+class DirectPosition extends Position, TDirectPosition {
+  int index;
+
+  DirectPosition() { this = TDirectPosition(index) }
+
+  override string toString() { if index = -1 then result = "this" else result = index.toString() }
+
+  int getIndex() { result = index }
+}
+
+class IndirectionPosition extends Position, TIndirectionPosition {
+  int argumentIndex;
+  int indirectionIndex;
+
+  IndirectionPosition() { this = TIndirectionPosition(argumentIndex, indirectionIndex) }
+
+  override string toString() {
+    if argumentIndex = -1
+    then if indirectionIndex > 0 then result = "this indirection" else result = "this"
+    else
+      if indirectionIndex > 0
+      then result = argumentIndex.toString() + " indirection"
+      else result = argumentIndex.toString()
+  }
+
+  int getArgumentIndex() { result = argumentIndex }
+
+  int getIndirectionIndex() { result = indirectionIndex }
+}
+
+newtype TPosition =
+  TDirectPosition(int index) { exists(any(CallInstruction c).getArgument(index)) } or
+  TIndirectionPosition(int argumentIndex, int indirectionIndex) {
+    hasOperandAndIndex(_, any(CallInstruction call).getArgumentOperand(argumentIndex),
+      indirectionIndex)
+  }
+
+private newtype TReturnKind =
+  TNormalReturnKind(int index) {
+    exists(IndirectReturnNode return |
+      return.getAddressOperand() = any(ReturnValueInstruction r).getReturnAddressOperand() and
+      index = return.getIndirectionIndex() - 1 // We subtract one because the return loads the value.
+    )
+  } or
+  TIndirectReturnKind(int argumentIndex, int indirectionIndex) {
+    exists(IndirectReturnNode return, ReturnIndirectionInstruction returnInd |
+      returnInd.hasIndex(argumentIndex) and
+      return.getAddressOperand() = returnInd.getSourceAddressOperand() and
+      indirectionIndex = return.getIndirectionIndex()
+    )
+  }
+
+/**
+ * A return kind. A return kind describes how a value can be returned
+ * from a callable. For C++, this is simply a function return.
+ */
+class ReturnKind extends TReturnKind {
+  /** Gets a textual representation of this return kind. */
+  abstract string toString();
+}
+
+private class NormalReturnKind extends ReturnKind, TNormalReturnKind {
+  int index;
+
+  NormalReturnKind() { this = TNormalReturnKind(index) }
+
+  override string toString() { result = "indirect return" }
+}
+
+private class IndirectReturnKind extends ReturnKind, TIndirectReturnKind {
+  int argumentIndex;
+  int indirectionIndex;
+
+  IndirectReturnKind() { this = TIndirectReturnKind(argumentIndex, indirectionIndex) }
+
+  override string toString() { result = "indirect outparam[" + argumentIndex.toString() + "]" }
+}
+
+/** A data flow node that occurs as the result of a `ReturnStmt`. */
+class ReturnNode extends Node instanceof IndirectReturnNode {
+  /** Gets the kind of this returned value. */
+  abstract ReturnKind getKind();
+}
+
+/**
+ * This predicate represents an annoying hack that we have to do. We use the
+ * `ReturnIndirectionInstruction` to determine which variables need flow back
+ * out of a function. However, the IR will unconditionally create those for a
+ * variable passed to a function even though the variable was never updated by
+ * the function. And if a function has too many `ReturnNode`s the dataflow
+ * library lowers its precision for that function by disabling field flow.
+ *
+ * So we those eliminate `ReturnNode`s that would have otherwise been created
+ * by this unconditional `ReturnIndirectionInstruction` by requiring that there
+ * must exist an SSA definition of the IR variable in the function.
+ */
+private predicate hasNonInitializeParameterDef(IRVariable v) {
+  exists(Ssa::Def def |
+    not def.getDefiningInstruction() instanceof InitializeParameterInstruction and
+    v = def.getSourceVariable().getBaseVariable().(Ssa::BaseIRVariable).getIRVariable()
+  )
+}
+
+class ReturnIndirectionNode extends IndirectReturnNode, ReturnNode {
+  override ReturnKind getKind() {
+    exists(int argumentIndex, ReturnIndirectionInstruction returnInd |
+      returnInd.hasIndex(argumentIndex) and
+      this.getAddressOperand() = returnInd.getSourceAddressOperand() and
+      result = TIndirectReturnKind(argumentIndex, this.getIndirectionIndex()) and
+      hasNonInitializeParameterDef(returnInd.getIRVariable())
+    )
+    or
+    this.getAddressOperand() = any(ReturnValueInstruction r).getReturnAddressOperand() and
+    result = TNormalReturnKind(this.getIndirectionIndex() - 1)
+  }
+}
+
+private Operand fullyConvertedCallStep(Operand op) {
+  not exists(getANonConversionUse(op)) and
+  exists(Instruction instr |
+    conversionFlow(op, instr, _) and
+    result = getAUse(instr)
+  )
+}
+
+/**
+ * Gets the instruction that uses this operand, if the instruction is not
+ * ignored for dataflow purposes.
+ */
+private Instruction getUse(Operand op) {
+  result = op.getUse() and
+  not Ssa::ignoreOperand(op)
+}
+
+/** Gets a use of the instruction `instr` that is not ignored for dataflow purposes. */
+Operand getAUse(Instruction instr) {
+  result = instr.getAUse() and
+  not Ssa::ignoreOperand(result)
+}
+
+/**
+ * Gets a use of `operand` that is:
+ * - not ignored for dataflow purposes, and
+ * - not a conversion-like instruction.
+ */
+private Instruction getANonConversionUse(Operand operand) {
+  result = getUse(operand) and
+  not conversionFlow(_, result, _)
+}
+
+/**
+ * Gets the operand that represents the first use of the value of `call` following
+ * a sequence of conversion-like instructions.
+ */
+predicate operandForfullyConvertedCall(Operand operand, CallInstruction call) {
+  exists(getANonConversionUse(operand)) and
+  (
+    operand = getAUse(call)
+    or
+    operand = fullyConvertedCallStep*(getAUse(call))
+  )
+}
+
+/**
+ * Gets the instruction that represents the first use of the value of `call` following
+ * a sequence of conversion-like instructions.
+ *
+ * This predicate only holds if there is no suitable operand (i.e., no operand of a non-
+ * conversion instruction) to use to represent the value of `call` after conversions.
+ */
+predicate instructionForfullyConvertedCall(Instruction instr, CallInstruction call) {
+  not operandForfullyConvertedCall(_, call) and
+  (
+    // If there is no use of the call then we pick the call instruction
+    not exists(getAUse(call)) and
+    instr = call
+    or
+    // Otherwise, flow to the first non-conversion use.
+    exists(Operand operand | operand = fullyConvertedCallStep*(getAUse(call)) |
+      instr = getANonConversionUse(operand)
+    )
+  )
+}
+
+/** Holds if `node` represents the output node for `call`. */
+private predicate simpleOutNode(Node node, CallInstruction call) {
+  operandForfullyConvertedCall(node.asOperand(), call)
+  or
+  instructionForfullyConvertedCall(node.asInstruction(), call)
+}
+
+/** A data flow node that represents the output of a call. */
+class OutNode extends Node {
+  OutNode() {
+    // Return values not hidden behind indirections
+    simpleOutNode(this, _)
+    or
+    // Return values hidden behind indirections
+    this instanceof IndirectReturnOutNode
+    or
+    // Modified arguments hidden behind indirections
+    this instanceof IndirectArgumentOutNode
+  }
+
+  /** Gets the underlying call. */
+  abstract DataFlowCall getCall();
+
+  abstract ReturnKind getReturnKind();
+}
+
+private class DirectCallOutNode extends OutNode {
+  CallInstruction call;
+
+  DirectCallOutNode() { simpleOutNode(this, call) }
+
+  override DataFlowCall getCall() { result = call }
+
+  override ReturnKind getReturnKind() { result = TNormalReturnKind(0) }
+}
+
+private class IndirectCallOutNode extends OutNode, IndirectReturnOutNode {
+  override DataFlowCall getCall() { result = this.getCallInstruction() }
+
+  override ReturnKind getReturnKind() { result = TNormalReturnKind(this.getIndirectionIndex()) }
+}
+
+private class SideEffectOutNode extends OutNode, IndirectArgumentOutNode {
+  override DataFlowCall getCall() { result = this.getCallInstruction() }
+
+  override ReturnKind getReturnKind() {
+    result = TIndirectReturnKind(this.getArgumentIndex(), this.getIndirectionIndex())
+  }
+}
+
+/**
+ * Gets a node that can read the value returned from `call` with return kind
+ * `kind`.
+ */
+OutNode getAnOutNode(DataFlowCall call, ReturnKind kind) {
+  result.getCall() = call and
+  result.getReturnKind() = kind
+}
+
+/**
+ * Holds if data can flow from `node1` to `node2` in a way that loses the
+ * calling context. For example, this would happen with flow through a
+ * global or static variable.
+ */
+predicate jumpStep(Node n1, Node n2) {
+  exists(Cpp::GlobalOrNamespaceVariable v |
+    v =
+      n1.asInstruction()
+          .(StoreInstruction)
+          .getResultAddress()
+          .(VariableAddressInstruction)
+          .getAstVariable() and
+    v = n2.asVariable()
+    or
+    v =
+      n2.asInstruction()
+          .(LoadInstruction)
+          .getSourceAddress()
+          .(VariableAddressInstruction)
+          .getAstVariable() and
+    v = n1.asVariable()
+  )
+}
+
+/**
+ * Holds if data can flow from `node1` to `node2` via an assignment to `f`.
+ * Thus, `node2` references an object with a field `f` that contains the
+ * value of `node1`.
+ */
+predicate storeStep(Node node1, Content c, PostFieldUpdateNode node2) {
+  exists(int indirectionIndex1, int numberOfLoads, StoreInstruction store |
+    nodeHasInstruction(node1, store, pragma[only_bind_into](indirectionIndex1)) and
+    node2.getIndirectionIndex() = 1 and
+    numberOfLoadsFromOperand(node2.getFieldAddress(), store.getDestinationAddressOperand(),
+      numberOfLoads)
+  |
+    exists(FieldContent fc | fc = c |
+      fc.getField() = node2.getUpdatedField() and
+      fc.getIndirectionIndex() = 1 + indirectionIndex1 + numberOfLoads
+    )
+    or
+    exists(UnionContent uc | uc = c |
+      uc.getAField() = node2.getUpdatedField() and
+      uc.getIndirectionIndex() = 1 + indirectionIndex1 + numberOfLoads
+    )
+  )
+}
+
+/**
+ * Holds if `operandFrom` flows to `operandTo` using a sequence of conversion-like
+ * operations and exactly `n` `LoadInstruction` operations.
+ */
+private predicate numberOfLoadsFromOperandRec(Operand operandFrom, Operand operandTo, int ind) {
+  exists(LoadInstruction load | load.getSourceAddressOperand() = operandFrom |
+    operandTo = operandFrom and ind = 0
+    or
+    numberOfLoadsFromOperand(load.getAUse(), operandTo, ind - 1)
+  )
+  or
+  exists(Operand op, Instruction instr |
+    instr = op.getDef() and
+    conversionFlow(operandFrom, instr, _) and
+    numberOfLoadsFromOperand(op, operandTo, ind)
+  )
+}
+
+/**
+ * Holds if `operandFrom` flows to `operandTo` using a sequence of conversion-like
+ * operations and exactly `n` `LoadInstruction` operations.
+ */
+private predicate numberOfLoadsFromOperand(Operand operandFrom, Operand operandTo, int n) {
+  numberOfLoadsFromOperandRec(operandFrom, operandTo, n)
+  or
+  not any(LoadInstruction load).getSourceAddressOperand() = operandFrom and
+  not conversionFlow(operandFrom, _, _) and
+  operandFrom = operandTo and
+  n = 0
+}
+
+// Needed to join on both an operand and an index at the same time.
+pragma[noinline]
+predicate nodeHasOperand(Node node, Operand operand, int indirectionIndex) {
+  node.asOperand() = operand and indirectionIndex = 0
+  or
+  hasOperandAndIndex(node, operand, indirectionIndex)
+}
+
+// Needed to join on both an instruction and an index at the same time.
+pragma[noinline]
+predicate nodeHasInstruction(Node node, Instruction instr, int indirectionIndex) {
+  node.asInstruction() = instr and indirectionIndex = 0
+  or
+  hasInstructionAndIndex(node, instr, indirectionIndex)
+}
+
+/**
+ * Holds if data can flow from `node1` to `node2` via a read of `f`.
+ * Thus, `node1` references an object with a field `f` whose value ends up in
+ * `node2`.
+ */
+predicate readStep(Node node1, Content c, Node node2) {
+  exists(FieldAddress fa1, Operand operand, int numberOfLoads, int indirectionIndex2 |
+    nodeHasOperand(node2, operand, indirectionIndex2) and
+    nodeHasOperand(node1, fa1.getObjectAddressOperand(), _) and
+    numberOfLoadsFromOperand(fa1, operand, numberOfLoads)
+  |
+    exists(FieldContent fc | fc = c |
+      fc.getField() = fa1.getField() and
+      fc.getIndirectionIndex() = indirectionIndex2 + numberOfLoads
+    )
+    or
+    exists(UnionContent uc | uc = c |
+      uc.getAField() = fa1.getField() and
+      uc.getIndirectionIndex() = indirectionIndex2 + numberOfLoads
+    )
+  )
+}
+
+/**
+ * Holds if values stored inside content `c` are cleared at node `n`.
+ */
+predicate clearsContent(Node n, Content c) {
+  none() // stub implementation
+}
+
+/**
+ * Holds if the value that is being tracked is expected to be stored inside content `c`
+ * at node `n`.
+ */
+predicate expectsContent(Node n, ContentSet c) { none() }
+
+/** Gets the type of `n` used for type pruning. */
+DataFlowType getNodeType(Node n) {
+  suppressUnusedNode(n) and
+  result instanceof VoidType // stub implementation
+}
+
+/** Gets a string representation of a type returned by `getNodeType`. */
+string ppReprType(DataFlowType t) { none() } // stub implementation
+
+/**
+ * Holds if `t1` and `t2` are compatible, that is, whether data can flow from
+ * a node of type `t1` to a node of type `t2`.
+ */
+pragma[inline]
+predicate compatibleTypes(DataFlowType t1, DataFlowType t2) {
+  any() // stub implementation
+}
+
+private predicate suppressUnusedNode(Node n) { any() }
+
+//////////////////////////////////////////////////////////////////////////////
+// Java QL library compatibility wrappers
+//////////////////////////////////////////////////////////////////////////////
+/** A node that performs a type cast. */
+class CastNode extends Node {
+  CastNode() { none() } // stub implementation
+}
+
+/**
+ * A function that may contain code or a variable that may contain itself. When
+ * flow crosses from one _enclosing callable_ to another, the interprocedural
+ * data-flow library discards call contexts and inserts a node in the big-step
+ * relation used for human-readable path explanations.
+ */
+class DataFlowCallable = Cpp::Declaration;
+
+class DataFlowExpr = Expr;
+
+class DataFlowType = Type;
+
+/** A function call relevant for data flow. */
+class DataFlowCall extends CallInstruction {
+  Function getEnclosingCallable() { result = this.getEnclosingFunction() }
+}
+
+predicate isUnreachableInCall(Node n, DataFlowCall call) { none() } // stub implementation
+
+int accessPathLimit() { result = 5 }
+
+/**
+ * Holds if access paths with `c` at their head always should be tracked at high
+ * precision. This disables adaptive access path precision for such access paths.
+ */
+predicate forceHighPrecision(Content c) { none() }
+
+/** The unit type. */
+private newtype TUnit = TMkUnit()
+
+/** The trivial type with a single element. */
+class Unit extends TUnit {
+  /** Gets a textual representation of this element. */
+  string toString() { result = "unit" }
+}
+
+/** Holds if `n` should be hidden from path explanations. */
+predicate nodeIsHidden(Node n) { n instanceof OperandNode and not n instanceof ArgumentNode }
+
+class LambdaCallKind = Unit;
+
+/** Holds if `creation` is an expression that creates a lambda of kind `kind` for `c`. */
+predicate lambdaCreation(Node creation, LambdaCallKind kind, DataFlowCallable c) { none() }
+
+/** Holds if `call` is a lambda call of kind `kind` where `receiver` is the lambda expression. */
+predicate lambdaCall(DataFlowCall call, LambdaCallKind kind, Node receiver) { none() }
+
+/** Extra data-flow steps needed for lambda flow analysis. */
+predicate additionalLambdaFlowStep(Node nodeFrom, Node nodeTo, boolean preservesValue) { none() }
+
+/**
+ * Holds if flow is allowed to pass from parameter `p` and back to itself as a
+ * side-effect, resulting in a summary from `p` to itself.
+ *
+ * One example would be to allow flow like `p.foo = p.bar;`, which is disallowed
+ * by default as a heuristic.
+ */
+predicate allowParameterReturnInSelf(ParameterNode p) { none() }
+
+private class MyConsistencyConfiguration extends Consistency::ConsistencyConfiguration {
+  override predicate argHasPostUpdateExclude(ArgumentNode n) {
+    // The rules for whether an IR argument gets a post-update node are too
+    // complex to model here.
+    any()
+  }
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/ModelUtil.qll

@@ -0,0 +1,93 @@
+/**
+ * Provides predicates for mapping the `FunctionInput` and `FunctionOutput`
+ * classes used in function models to the corresponding instructions.
+ */
+
+private import semmle.code.cpp.ir.IR
+private import experimental.semmle.code.cpp.ir.dataflow.DataFlow
+private import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowUtil
+private import SsaInternals as Ssa
+
+/**
+ * Gets the instruction that goes into `input` for `call`.
+ */
+DataFlow::Node callInput(CallInstruction call, FunctionInput input) {
+  // An argument or qualifier
+  exists(int index |
+    result.asOperand() = call.getArgumentOperand(index) and
+    input.isParameterOrQualifierAddress(index)
+  )
+  or
+  // A value pointed to by an argument or qualifier
+  exists(int index, int indirectionIndex |
+    hasOperandAndIndex(result, call.getArgumentOperand(index), indirectionIndex) and
+    input.isParameterDerefOrQualifierObject(index, indirectionIndex)
+  )
+  or
+  exists(int ind |
+    result = getIndirectReturnOutNode(call, ind) and
+    input.isReturnValueDeref(ind)
+  )
+}
+
+/**
+ * Gets the instruction that holds the `output` for `call`.
+ */
+Node callOutput(CallInstruction call, FunctionOutput output) {
+  // The return value
+  result.asInstruction() = call and
+  output.isReturnValue()
+  or
+  // The side effect of a call on the value pointed to by an argument or qualifier
+  exists(int index, int indirectionIndex |
+    result.(IndirectArgumentOutNode).getArgumentIndex() = index and
+    result.(IndirectArgumentOutNode).getIndirectionIndex() = indirectionIndex and
+    result.(IndirectArgumentOutNode).getCallInstruction() = call and
+    output.isParameterDerefOrQualifierObject(index, indirectionIndex)
+  )
+  or
+  exists(int ind |
+    result = getIndirectReturnOutNode(call, ind) and
+    output.isReturnValueDeref(ind)
+  )
+}
+
+DataFlow::Node callInput(CallInstruction call, FunctionInput input, int d) {
+  exists(DataFlow::Node n | n = callInput(call, input) and d > 0 |
+    // An argument or qualifier
+    hasOperandAndIndex(result, n.asOperand(), d)
+    or
+    exists(Operand operand, int indirectionIndex |
+      // A value pointed to by an argument or qualifier
+      hasOperandAndIndex(n, operand, indirectionIndex) and
+      hasOperandAndIndex(result, operand, indirectionIndex + d)
+    )
+  )
+}
+
+private IndirectReturnOutNode getIndirectReturnOutNode(CallInstruction call, int d) {
+  result.getCallInstruction() = call and
+  result.getIndirectionIndex() = d
+}
+
+/**
+ * Gets the instruction that holds the `output` for `call`.
+ */
+bindingset[d]
+Node callOutput(CallInstruction call, FunctionOutput output, int d) {
+  exists(DataFlow::Node n | n = callOutput(call, output) and d > 0 |
+    // The return value
+    result = getIndirectReturnOutNode(n.asInstruction(), d)
+    or
+    // If there isn't an indirect out node for the call with indirection `d` then
+    // we conflate this with the underlying `CallInstruction`.
+    not exists(getIndirectReturnOutNode(call, d)) and
+    n.asInstruction() = result.asInstruction()
+    or
+    // The side effect of a call on the value pointed to by an argument or qualifier
+    exists(Operand operand, int indirectionIndex |
+      Ssa::outNodeHasAddressAndIndex(n, operand, indirectionIndex) and
+      Ssa::outNodeHasAddressAndIndex(result, operand, indirectionIndex + d)
+    )
+  )
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/PrintIRLocalFlow.qll

@@ -0,0 +1,136 @@
+private import cpp
+// The `ValueNumbering` library has to be imported right after `cpp` to ensure
+// that the cached IR gets the same checksum here as it does in queries that use
+// `ValueNumbering` without `DataFlow`.
+private import semmle.code.cpp.ir.ValueNumbering
+private import semmle.code.cpp.ir.IR
+private import semmle.code.cpp.ir.dataflow.DataFlow
+private import semmle.code.cpp.ir.dataflow.internal.DataFlowUtil
+private import PrintIRUtilities
+
+/**
+ * Gets the local dataflow from other nodes in the same function to this node.
+ */
+private string getFromFlow(DataFlow::Node useNode, int order1, int order2) {
+  exists(DataFlow::Node defNode, string prefix |
+    (
+      simpleLocalFlowStep(defNode, useNode) and prefix = ""
+      or
+      any(DataFlow::Configuration cfg).isAdditionalFlowStep(defNode, useNode) and
+      defNode.getEnclosingCallable() = useNode.getEnclosingCallable() and
+      prefix = "+"
+    ) and
+    if defNode.asInstruction() = useNode.asOperand().getAnyDef()
+    then
+      // Shorthand for flow from the def of this operand.
+      result = prefix + "def" and
+      order1 = -1 and
+      order2 = 0
+    else
+      if defNode.asOperand().getUse() = useNode.asInstruction()
+      then
+        // Shorthand for flow from an operand of this instruction
+        result = prefix + defNode.asOperand().getDumpId() and
+        order1 = -1 and
+        order2 = defNode.asOperand().getDumpSortOrder()
+      else result = prefix + nodeId(defNode, order1, order2)
+  )
+}
+
+/**
+ * Gets the local dataflow from this node to other nodes in the same function.
+ */
+private string getToFlow(DataFlow::Node defNode, int order1, int order2) {
+  exists(DataFlow::Node useNode, string prefix |
+    (
+      simpleLocalFlowStep(defNode, useNode) and prefix = ""
+      or
+      any(DataFlow::Configuration cfg).isAdditionalFlowStep(defNode, useNode) and
+      defNode.getEnclosingCallable() = useNode.getEnclosingCallable() and
+      prefix = "+"
+    ) and
+    if useNode.asInstruction() = defNode.asOperand().getUse()
+    then
+      // Shorthand for flow to this operand's instruction.
+      result = prefix + "result" and
+      order1 = -1 and
+      order2 = 0
+    else result = prefix + nodeId(useNode, order1, order2)
+  )
+}
+
+/**
+ * Gets the properties of the dataflow node `node`.
+ */
+private string getNodeProperty(DataFlow::Node node, string key) {
+  // List dataflow into and out of this node. Flow into this node is printed as `src->@`, and flow
+  // out of this node is printed as `@->dest`.
+  key = "flow" and
+  result =
+    strictconcat(string flow, boolean to, int order1, int order2 |
+      flow = getFromFlow(node, order1, order2) + "->@" and to = false
+      or
+      flow = "@->" + getToFlow(node, order1, order2) and to = true
+    |
+      flow, ", " order by to, order1, order2, flow
+    )
+  or
+  // Is this node a dataflow sink?
+  key = "sink" and
+  any(DataFlow::Configuration cfg).isSink(node) and
+  result = "true"
+  or
+  // Is this node a dataflow source?
+  key = "source" and
+  any(DataFlow::Configuration cfg).isSource(node) and
+  result = "true"
+  or
+  // Is this node a dataflow barrier, and if so, what kind?
+  key = "barrier" and
+  result =
+    strictconcat(string kind |
+      any(DataFlow::Configuration cfg).isBarrier(node) and kind = "full"
+      or
+      any(DataFlow::Configuration cfg).isBarrierIn(node) and kind = "in"
+      or
+      any(DataFlow::Configuration cfg).isBarrierOut(node) and kind = "out"
+    |
+      kind, ", "
+    )
+  or
+  // Is there partial flow from a source to this node?
+  // This property will only be emitted if partial flow is enabled by overriding
+  // `DataFlow::Configuration::explorationLimit()`.
+  key = "pflow" and
+  result =
+    strictconcat(DataFlow::PartialPathNode sourceNode, DataFlow::PartialPathNode destNode, int dist,
+      int order1, int order2 |
+      any(DataFlow::Configuration cfg).hasPartialFlow(sourceNode, destNode, dist) and
+      destNode.getNode() = node and
+      // Only print flow from a source in the same function.
+      sourceNode.getNode().getEnclosingCallable() = node.getEnclosingCallable()
+    |
+      nodeId(sourceNode.getNode(), order1, order2) + "+" + dist.toString(), ", "
+      order by
+        order1, order2, dist desc
+    )
+}
+
+/**
+ * Property provider for local IR dataflow.
+ */
+class LocalFlowPropertyProvider extends IRPropertyProvider {
+  override string getOperandProperty(Operand operand, string key) {
+    exists(DataFlow::Node node |
+      operand = node.asOperand() and
+      result = getNodeProperty(node, key)
+    )
+  }
+
+  override string getInstructionProperty(Instruction instruction, string key) {
+    exists(DataFlow::Node node |
+      instruction = node.asInstruction() and
+      result = getNodeProperty(node, key)
+    )
+  }
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/PrintIRStoreSteps.qll

@@ -0,0 +1,33 @@
+/**
+ * Print the dataflow local store steps in IR dumps.
+ */
+
+private import cpp
+// The `ValueNumbering` library has to be imported right after `cpp` to ensure
+// that the cached IR gets the same checksum here as it does in queries that use
+// `ValueNumbering` without `DataFlow`.
+private import semmle.code.cpp.ir.ValueNumbering
+private import semmle.code.cpp.ir.IR
+private import semmle.code.cpp.ir.dataflow.DataFlow
+private import semmle.code.cpp.ir.dataflow.internal.DataFlowUtil
+private import semmle.code.cpp.ir.dataflow.internal.DataFlowPrivate
+private import PrintIRUtilities
+
+/**
+ * Property provider for local IR dataflow store steps.
+ */
+class LocalFlowPropertyProvider extends IRPropertyProvider {
+  override string getInstructionProperty(Instruction instruction, string key) {
+    exists(DataFlow::Node objectNode, Content content |
+      key = "content[" + content.toString() + "]" and
+      instruction = objectNode.asInstruction() and
+      result =
+        strictconcat(string element, DataFlow::Node fieldNode |
+          storeStep(fieldNode, content, objectNode) and
+          element = nodeId(fieldNode, _, _)
+        |
+          element, ", "
+        )
+    )
+  }
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/PrintIRUtilities.qll

@@ -0,0 +1,39 @@
+/**
+ * Shared utilities used when printing dataflow annotations in IR dumps.
+ */
+
+private import cpp
+// The `ValueNumbering` library has to be imported right after `cpp` to ensure
+// that the cached IR gets the same checksum here as it does in queries that use
+// `ValueNumbering` without `DataFlow`.
+private import semmle.code.cpp.ir.ValueNumbering
+private import semmle.code.cpp.ir.IR
+private import semmle.code.cpp.ir.dataflow.DataFlow
+
+/**
+ * Gets a short ID for an IR dataflow node.
+ * - For `Instruction`s, this is just the result ID of the instruction (e.g. `m128`).
+ * - For `Operand`s, this is the label of the operand, prefixed with the result ID of the
+ *   instruction and a dot (e.g. `m128.left`).
+ * - For `Variable`s, this is the qualified name of the variable.
+ */
+string nodeId(DataFlow::Node node, int order1, int order2) {
+  exists(Instruction instruction | instruction = node.asInstruction() |
+    result = instruction.getResultId() and
+    order1 = instruction.getBlock().getDisplayIndex() and
+    order2 = instruction.getDisplayIndexInBlock()
+  )
+  or
+  exists(Operand operand, Instruction instruction |
+    operand = node.asOperand() and
+    instruction = operand.getUse()
+  |
+    result = instruction.getResultId() + "." + operand.getDumpId() and
+    order1 = instruction.getBlock().getDisplayIndex() and
+    order2 = instruction.getDisplayIndexInBlock()
+  )
+  or
+  result = "var(" + node.asVariable().getQualifiedName() + ")" and
+  order1 = 1000000 and
+  order2 = 0
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/SsaInternals.qll

@@ -0,0 +1,552 @@
+private import codeql.ssa.Ssa as SsaImplCommon
+private import semmle.code.cpp.ir.IR
+private import DataFlowUtil
+private import DataFlowImplCommon as DataFlowImplCommon
+private import semmle.code.cpp.models.interfaces.Allocation as Alloc
+private import semmle.code.cpp.models.interfaces.DataFlow as DataFlow
+private import semmle.code.cpp.ir.internal.IRCppLanguage
+private import DataFlowPrivate
+private import ssa0.SsaInternals as SsaInternals0
+import SsaInternalsCommon
+
+private module SourceVariables {
+  int getMaxIndirectionForIRVariable(IRVariable var) {
+    exists(Type type, boolean isGLValue |
+      var.getLanguageType().hasType(type, isGLValue) and
+      if isGLValue = true
+      then result = 1 + getMaxIndirectionsForType(type)
+      else result = getMaxIndirectionsForType(type)
+    )
+  }
+
+  class BaseSourceVariable = SsaInternals0::BaseSourceVariable;
+
+  class BaseIRVariable = SsaInternals0::BaseIRVariable;
+
+  class BaseCallVariable = SsaInternals0::BaseCallVariable;
+
+  cached
+  private newtype TSourceVariable =
+    TSourceIRVariable(BaseIRVariable baseVar, int ind) {
+      ind = [0 .. getMaxIndirectionForIRVariable(baseVar.getIRVariable())]
+    } or
+    TCallVariable(AllocationInstruction call, int ind) {
+      ind = [0 .. countIndirectionsForCppType(getResultLanguageType(call))]
+    }
+
+  abstract class SourceVariable extends TSourceVariable {
+    int ind;
+
+    bindingset[ind]
+    SourceVariable() { any() }
+
+    abstract string toString();
+
+    int getIndirection() { result = ind }
+
+    abstract BaseSourceVariable getBaseVariable();
+  }
+
+  class SourceIRVariable extends SourceVariable, TSourceIRVariable {
+    BaseIRVariable var;
+
+    SourceIRVariable() { this = TSourceIRVariable(var, ind) }
+
+    IRVariable getIRVariable() { result = var.getIRVariable() }
+
+    override BaseIRVariable getBaseVariable() { result.getIRVariable() = this.getIRVariable() }
+
+    override string toString() {
+      ind = 0 and
+      result = this.getIRVariable().toString()
+      or
+      ind > 0 and
+      result = this.getIRVariable().toString() + " indirection"
+    }
+  }
+
+  class CallVariable extends SourceVariable, TCallVariable {
+    AllocationInstruction call;
+
+    CallVariable() { this = TCallVariable(call, ind) }
+
+    AllocationInstruction getCall() { result = call }
+
+    override BaseCallVariable getBaseVariable() { result.getCallInstruction() = call }
+
+    override string toString() {
+      ind = 0 and
+      result = "Call"
+      or
+      ind > 0 and
+      result = "Call indirection"
+    }
+  }
+}
+
+import SourceVariables
+
+predicate hasIndirectOperand(Operand op, int indirectionIndex) {
+  exists(CppType type, int m |
+    not ignoreOperand(op) and
+    type = getLanguageType(op) and
+    m = countIndirectionsForCppType(type) and
+    indirectionIndex = [1 .. m]
+  )
+}
+
+predicate hasIndirectInstruction(Instruction instr, int indirectionIndex) {
+  exists(CppType type, int m |
+    not ignoreInstruction(instr) and
+    type = getResultLanguageType(instr) and
+    m = countIndirectionsForCppType(type) and
+    indirectionIndex = [1 .. m]
+  )
+}
+
+cached
+private newtype TDefOrUseImpl =
+  TDefImpl(Operand address, int indirectionIndex) {
+    isDef(_, _, address, _, _, indirectionIndex) and
+    // We only include the definition if the SSA pruning stage
+    // concluded that the definition is live after the write.
+    any(SsaInternals0::Def def).getAddressOperand() = address
+  } or
+  TUseImpl(Operand operand, int indirectionIndex) {
+    isUse(_, operand, _, _, indirectionIndex) and
+    not isDef(_, _, operand, _, _, _)
+  }
+
+abstract private class DefOrUseImpl extends TDefOrUseImpl {
+  /** Gets a textual representation of this element. */
+  abstract string toString();
+
+  /** Gets the block of this definition or use. */
+  abstract IRBlock getBlock();
+
+  /** Holds if this definition or use has index `index` in block `block`. */
+  abstract predicate hasIndexInBlock(IRBlock block, int index);
+
+  final predicate hasIndexInBlock(IRBlock block, int index, SourceVariable sv) {
+    this.hasIndexInBlock(block, index) and
+    sv = this.getSourceVariable()
+  }
+
+  /** Gets the location of this element. */
+  abstract Cpp::Location getLocation();
+
+  /**
+   * Gets the index (i.e., the number of loads required) of this
+   * definition or use.
+   *
+   * Note that this is _not_ the definition's (or use's) index in
+   * the enclosing basic block. To obtain this index, use
+   * `DefOrUseImpl::hasIndexInBlock/2` or `DefOrUseImpl::hasIndexInBlock/3`.
+   */
+  abstract int getIndirectionIndex();
+
+  /**
+   * Gets the instruction that computes the base of this definition or use.
+   * This is always a `VariableAddressInstruction` or an `AllocationInstruction`.
+   */
+  abstract Instruction getBase();
+
+  final BaseSourceVariable getBaseSourceVariable() {
+    exists(IRVariable var |
+      result.(BaseIRVariable).getIRVariable() = var and
+      instructionHasIRVariable(this.getBase(), var)
+    )
+    or
+    result.(BaseCallVariable).getCallInstruction() = this.getBase()
+  }
+
+  /** Gets the variable that is defined or used. */
+  final SourceVariable getSourceVariable() {
+    exists(BaseSourceVariable v, int ind |
+      sourceVariableHasBaseAndIndex(result, v, ind) and
+      defOrUseHasSourceVariable(this, v, ind)
+    )
+  }
+}
+
+pragma[noinline]
+private predicate instructionHasIRVariable(VariableAddressInstruction vai, IRVariable var) {
+  vai.getIRVariable() = var
+}
+
+private predicate defOrUseHasSourceVariable(DefOrUseImpl defOrUse, BaseSourceVariable bv, int ind) {
+  defHasSourceVariable(defOrUse, bv, ind)
+  or
+  useHasSourceVariable(defOrUse, bv, ind)
+}
+
+pragma[noinline]
+private predicate defHasSourceVariable(DefImpl def, BaseSourceVariable bv, int ind) {
+  bv = def.getBaseSourceVariable() and
+  ind = def.getIndirection()
+}
+
+pragma[noinline]
+private predicate useHasSourceVariable(UseImpl use, BaseSourceVariable bv, int ind) {
+  bv = use.getBaseSourceVariable() and
+  ind = use.getIndirection()
+}
+
+pragma[noinline]
+private predicate sourceVariableHasBaseAndIndex(SourceVariable v, BaseSourceVariable bv, int ind) {
+  v.getBaseVariable() = bv and
+  v.getIndirection() = ind
+}
+
+class DefImpl extends DefOrUseImpl, TDefImpl {
+  Operand address;
+  int ind;
+
+  DefImpl() { this = TDefImpl(address, ind) }
+
+  override Instruction getBase() { isDef(_, _, address, result, _, _) }
+
+  Operand getAddressOperand() { result = address }
+
+  int getIndirection() { isDef(_, _, address, _, result, ind) }
+
+  override int getIndirectionIndex() { result = ind }
+
+  Instruction getDefiningInstruction() { isDef(_, result, address, _, _, _) }
+
+  override string toString() { result = "DefImpl" }
+
+  override IRBlock getBlock() { result = this.getDefiningInstruction().getBlock() }
+
+  override Cpp::Location getLocation() { result = this.getDefiningInstruction().getLocation() }
+
+  final override predicate hasIndexInBlock(IRBlock block, int index) {
+    this.getDefiningInstruction() = block.getInstruction(index)
+  }
+
+  predicate isCertain() { isDef(true, _, address, _, _, ind) }
+}
+
+class UseImpl extends DefOrUseImpl, TUseImpl {
+  Operand operand;
+  int ind;
+
+  UseImpl() { this = TUseImpl(operand, ind) }
+
+  Operand getOperand() { result = operand }
+
+  override string toString() { result = "UseImpl" }
+
+  final override predicate hasIndexInBlock(IRBlock block, int index) {
+    operand.getUse() = block.getInstruction(index)
+  }
+
+  final override IRBlock getBlock() { result = operand.getUse().getBlock() }
+
+  final override Cpp::Location getLocation() { result = operand.getLocation() }
+
+  final int getIndirection() { isUse(_, operand, _, result, ind) }
+
+  override int getIndirectionIndex() { result = ind }
+
+  override Instruction getBase() { isUse(_, operand, result, _, ind) }
+
+  predicate isCertain() { isUse(true, operand, _, _, ind) }
+}
+
+/**
+ * Holds if `defOrUse1` is a definition which is first read by `use`,
+ * or if `defOrUse1` is a use and `use` is a next subsequent use.
+ *
+ * In both cases, `use` can either be an explicit use written in the
+ * source file, or it can be a phi node as computed by the SSA library.
+ */
+predicate adjacentDefRead(DefOrUse defOrUse1, UseOrPhi use) {
+  exists(IRBlock bb1, int i1, SourceVariable v |
+    defOrUse1.asDefOrUse().hasIndexInBlock(bb1, i1, v)
+  |
+    exists(IRBlock bb2, int i2 |
+      adjacentDefRead(_, pragma[only_bind_into](bb1), pragma[only_bind_into](i1),
+        pragma[only_bind_into](bb2), pragma[only_bind_into](i2))
+    |
+      use.asDefOrUse().(UseImpl).hasIndexInBlock(bb2, i2, v)
+    )
+    or
+    exists(PhiNode phi |
+      lastRefRedef(_, bb1, i1, phi) and
+      use.asPhi() = phi and
+      phi.getSourceVariable() = pragma[only_bind_into](v)
+    )
+  )
+}
+
+private predicate useToNode(UseOrPhi use, Node nodeTo) {
+  exists(UseImpl useImpl |
+    useImpl = use.asDefOrUse() and
+    nodeHasOperand(nodeTo, useImpl.getOperand(), useImpl.getIndirectionIndex())
+  )
+  or
+  nodeTo.(SsaPhiNode).getPhiNode() = use.asPhi()
+}
+
+pragma[noinline]
+predicate outNodeHasAddressAndIndex(
+  IndirectArgumentOutNode out, Operand address, int indirectionIndex
+) {
+  out.getAddressOperand() = address and
+  out.getIndirectionIndex() = indirectionIndex
+}
+
+private predicate defToNode(Node nodeFrom, Def def) {
+  nodeHasInstruction(nodeFrom, def.getDefiningInstruction(), def.getIndirectionIndex())
+}
+
+/**
+ * INTERNAL: Do not use.
+ *
+ * Holds if `nodeFrom` is the node that correspond to the definition or use `defOrUse`.
+ */
+predicate nodeToDefOrUse(Node nodeFrom, SsaDefOrUse defOrUse) {
+  // Node -> Def
+  defToNode(nodeFrom, defOrUse)
+  or
+  // Node -> Use
+  useToNode(defOrUse, nodeFrom)
+}
+
+/**
+ * Perform a single conversion-like step from `nFrom` to `nTo`. This relation
+ * only holds when there is no use-use relation out of `nTo`.
+ */
+private predicate indirectConversionFlowStep(Node nFrom, Node nTo) {
+  not exists(UseOrPhi defOrUse |
+    nodeToDefOrUse(nTo, defOrUse) and
+    adjacentDefRead(defOrUse, _)
+  ) and
+  exists(Operand op1, Operand op2, int indirectionIndex, Instruction instr |
+    hasOperandAndIndex(nFrom, op1, pragma[only_bind_into](indirectionIndex)) and
+    hasOperandAndIndex(nTo, op2, pragma[only_bind_into](indirectionIndex)) and
+    instr = op2.getDef() and
+    conversionFlow(op1, instr, _)
+  )
+}
+
+/**
+ * The reason for this predicate is a bit annoying:
+ * We cannot mark a `PointerArithmeticInstruction` that computes an offset based on some SSA
+ * variable `x` as a use of `x` since this creates taint-flow in the following example:
+ * ```c
+ * int x = array[source]
+ * sink(*array)
+ * ```
+ * This is because `source` would flow from the operand of `PointerArithmeticInstruction` to the
+ * result of the instruction, and into the `IndirectOperand` that represents the value of `*array`.
+ * Then, via use-use flow, flow will arrive at `*array` in `sink(*array)`.
+ *
+ * So this predicate recurses back along conversions and `PointerArithmeticInstruction`s to find the
+ * first use that has provides use-use flow, and uses that target as the target of the `nodeFrom`.
+ */
+private predicate adjustForPointerArith(Node nodeFrom, UseOrPhi use) {
+  nodeFrom = any(PostUpdateNode pun).getPreUpdateNode() and
+  exists(DefOrUse defOrUse, Node adjusted |
+    indirectConversionFlowStep*(adjusted, nodeFrom) and
+    nodeToDefOrUse(adjusted, defOrUse) and
+    adjacentDefRead(defOrUse, use)
+  )
+}
+
+/** Holds if there is def-use or use-use flow from `nodeFrom` to `nodeTo`. */
+predicate ssaFlow(Node nodeFrom, Node nodeTo) {
+  // `nodeFrom = any(PostUpdateNode pun).getPreUpdateNode()` is implied by adjustedForPointerArith.
+  exists(UseOrPhi use |
+    adjustForPointerArith(nodeFrom, use) and
+    useToNode(use, nodeTo)
+  )
+  or
+  not nodeFrom = any(PostUpdateNode pun).getPreUpdateNode() and
+  exists(DefOrUse defOrUse1, UseOrPhi use |
+    nodeToDefOrUse(nodeFrom, defOrUse1) and
+    adjacentDefRead(defOrUse1, use) and
+    useToNode(use, nodeTo)
+  )
+}
+
+/** Holds if `nodeTo` receives flow from the phi node `nodeFrom`. */
+predicate fromPhiNode(SsaPhiNode nodeFrom, Node nodeTo) {
+  exists(PhiNode phi, SourceVariable sv, IRBlock bb1, int i1, UseOrPhi use |
+    phi = nodeFrom.getPhiNode() and
+    phi.definesAt(sv, bb1, i1) and
+    useToNode(use, nodeTo)
+  |
+    exists(IRBlock bb2, int i2 |
+      use.asDefOrUse().hasIndexInBlock(bb2, i2, sv) and
+      adjacentDefRead(phi, bb1, i1, bb2, i2)
+    )
+    or
+    exists(PhiNode phiTo |
+      lastRefRedef(phi, _, _, phiTo) and
+      nodeTo.(SsaPhiNode).getPhiNode() = phiTo
+    )
+  )
+}
+
+private SsaInternals0::SourceVariable getOldSourceVariable(SourceVariable v) {
+  v.getBaseVariable().(BaseIRVariable).getIRVariable() =
+    result.getBaseVariable().(SsaInternals0::BaseIRVariable).getIRVariable()
+  or
+  v.getBaseVariable().(BaseCallVariable).getCallInstruction() =
+    result.getBaseVariable().(SsaInternals0::BaseCallVariable).getCallInstruction()
+}
+
+/**
+ * Holds if there is a write at index `i` in basic block `bb` to variable `v` that's
+ * subsequently read (as determined by the SSA pruning stage).
+ */
+private predicate variableWriteCand(IRBlock bb, int i, SourceVariable v) {
+  exists(SsaInternals0::Def def, SsaInternals0::SourceVariable v0 |
+    def.asDefOrUse().hasIndexInBlock(bb, i, v0) and
+    v0 = getOldSourceVariable(v)
+  )
+}
+
+private module SsaInput implements SsaImplCommon::InputSig {
+  import InputSigCommon
+  import SourceVariables
+
+  /**
+   * Holds if the `i`'th write in block `bb` writes to the variable `v`.
+   * `certain` is `true` if the write is guaranteed to overwrite the entire variable.
+   */
+  predicate variableWrite(IRBlock bb, int i, SourceVariable v, boolean certain) {
+    DataFlowImplCommon::forceCachingInSameStage() and
+    variableWriteCand(bb, i, v) and
+    exists(DefImpl def | def.hasIndexInBlock(bb, i, v) |
+      if def.isCertain() then certain = true else certain = false
+    )
+  }
+
+  /**
+   * Holds if the `i`'th read in block `bb` reads to the variable `v`.
+   * `certain` is `true` if the read is guaranteed. For C++, this is always the case.
+   */
+  predicate variableRead(IRBlock bb, int i, SourceVariable v, boolean certain) {
+    exists(UseImpl use | use.hasIndexInBlock(bb, i, v) |
+      if use.isCertain() then certain = true else certain = false
+    )
+  }
+}
+
+/**
+ * The final SSA predicates used for dataflow purposes.
+ */
+cached
+module SsaCached {
+  /**
+   * Holds if `def` is accessed at index `i1` in basic block `bb1` (either a read
+   * or a write), `def` is read at index `i2` in basic block `bb2`, and there is a
+   * path between them without any read of `def`.
+   */
+  cached
+  predicate adjacentDefRead(Definition def, IRBlock bb1, int i1, IRBlock bb2, int i2) {
+    SsaImpl::adjacentDefRead(def, bb1, i1, bb2, i2)
+  }
+
+  /**
+   * Holds if the node at index `i` in `bb` is a last reference to SSA definition
+   * `def`. The reference is last because it can reach another write `next`,
+   * without passing through another read or write.
+   */
+  cached
+  predicate lastRefRedef(Definition def, IRBlock bb, int i, Definition next) {
+    SsaImpl::lastRefRedef(def, bb, i, next)
+  }
+}
+
+cached
+private newtype TSsaDefOrUse =
+  TDefOrUse(DefOrUseImpl defOrUse) {
+    defOrUse instanceof UseImpl
+    or
+    // Like in the pruning stage, we only include definition that's live after the
+    // write as the final definitions computed by SSA.
+    exists(Definition def, SourceVariable sv, IRBlock bb, int i |
+      def.definesAt(sv, bb, i) and
+      defOrUse.(DefImpl).hasIndexInBlock(bb, i, sv)
+    )
+  } or
+  TPhi(PhiNode phi)
+
+abstract private class SsaDefOrUse extends TSsaDefOrUse {
+  string toString() { none() }
+
+  DefOrUseImpl asDefOrUse() { none() }
+
+  PhiNode asPhi() { none() }
+
+  abstract Location getLocation();
+}
+
+class DefOrUse extends TDefOrUse, SsaDefOrUse {
+  DefOrUseImpl defOrUse;
+
+  DefOrUse() { this = TDefOrUse(defOrUse) }
+
+  final override DefOrUseImpl asDefOrUse() { result = defOrUse }
+
+  final override Location getLocation() { result = defOrUse.getLocation() }
+
+  final SourceVariable getSourceVariable() { result = defOrUse.getSourceVariable() }
+
+  override string toString() { result = defOrUse.toString() }
+}
+
+class Phi extends TPhi, SsaDefOrUse {
+  PhiNode phi;
+
+  Phi() { this = TPhi(phi) }
+
+  final override PhiNode asPhi() { result = phi }
+
+  final override Location getLocation() { result = phi.getBasicBlock().getLocation() }
+
+  override string toString() { result = "Phi" }
+}
+
+class UseOrPhi extends SsaDefOrUse {
+  UseOrPhi() {
+    this.asDefOrUse() instanceof UseImpl
+    or
+    this instanceof Phi
+  }
+
+  final override Location getLocation() {
+    result = this.asDefOrUse().getLocation() or result = this.(Phi).getLocation()
+  }
+}
+
+class Def extends DefOrUse {
+  override DefImpl defOrUse;
+
+  Operand getAddressOperand() { result = defOrUse.getAddressOperand() }
+
+  Instruction getAddress() { result = this.getAddressOperand().getDef() }
+
+  /**
+   * This predicate ensures that joins go from `defOrUse` to the result
+   * instead of the other way around.
+   */
+  pragma[inline]
+  int getIndirectionIndex() {
+    pragma[only_bind_into](result) = pragma[only_bind_out](defOrUse).getIndirectionIndex()
+  }
+
+  Instruction getDefiningInstruction() { result = defOrUse.getDefiningInstruction() }
+}
+
+private module SsaImpl = SsaImplCommon::Make<SsaInput>;
+
+class PhiNode = SsaImpl::PhiNode;
+
+class Definition = SsaImpl::Definition;
+
+import SsaCached

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/SsaInternalsCommon.qll

@@ -0,0 +1,270 @@
+import cpp as Cpp
+import semmle.code.cpp.ir.IR
+import semmle.code.cpp.ir.internal.IRCppLanguage
+private import semmle.code.cpp.ir.implementation.raw.internal.SideEffects as SideEffects
+private import DataFlowImplCommon as DataFlowImplCommon
+private import DataFlowUtil
+
+/**
+ * Holds if `operand` is an operand that is not used by the dataflow library.
+ * Ignored operands are not recognizd as uses by SSA, and they don't have a
+ * corresponding `(Indirect)OperandNode`.
+ */
+predicate ignoreOperand(Operand operand) {
+  operand = any(Instruction instr | ignoreInstruction(instr)).getAnOperand() or
+  operand = any(Instruction instr | ignoreInstruction(instr)).getAUse() or
+  operand instanceof MemoryOperand
+}
+
+/**
+ * Holds if `instr` is an instruction that is not used by the dataflow library.
+ * Ignored instructions are not recognized as reads/writes by SSA, and they
+ * don't have a corresponding `(Indirect)InstructionNode`.
+ */
+predicate ignoreInstruction(Instruction instr) {
+  DataFlowImplCommon::forceCachingInSameStage() and
+  (
+    instr instanceof WriteSideEffectInstruction or
+    instr instanceof PhiInstruction or
+    instr instanceof ReadSideEffectInstruction or
+    instr instanceof ChiInstruction or
+    instr instanceof InitializeIndirectionInstruction
+  )
+}
+
+/**
+ * Gets the C++ type of `this` in the member function `f`.
+ * The result is a glvalue if `isGLValue` is true, and
+ * a prvalue if `isGLValue` is false.
+ */
+bindingset[isGLValue]
+private CppType getThisType(Cpp::MemberFunction f, boolean isGLValue) {
+  result.hasType(f.getTypeOfThis(), isGLValue)
+}
+
+/**
+ * Gets the C++ type of the instruction `i`.
+ *
+ * This is equivalent to `i.getResultLanguageType()` with the exception
+ * of instructions that directly references a `this` IRVariable. In this
+ * case, `i.getResultLanguageType()` gives an unknown type, whereas the
+ * predicate gives the expected type (i.e., a potentially cv-qualified
+ * type `A*` where `A` is the declaring type of the member function that
+ * contains `i`).
+ */
+cached
+CppType getResultLanguageType(Instruction i) {
+  if i.(VariableAddressInstruction).getIRVariable() instanceof IRThisVariable
+  then
+    if i.isGLValue()
+    then result = getThisType(i.getEnclosingFunction(), true)
+    else result = getThisType(i.getEnclosingFunction(), false)
+  else result = i.getResultLanguageType()
+}
+
+/**
+ * Gets the C++ type of the operand `operand`.
+ * This is equivalent to the type of the operand's defining instruction.
+ *
+ * See `getResultLanguageType` for a description of this behavior.
+ */
+CppType getLanguageType(Operand operand) { result = getResultLanguageType(operand.getDef()) }
+
+/**
+ * Gets the maximum number of indirections a glvalue of type `type` can have.
+ * For example:
+ * - If `type = int`, the result is 1
+ * - If `type = MyStruct`, the result is 1
+ * - If `type = char*`, the result is 2
+ */
+int getMaxIndirectionsForType(Type type) {
+  result = countIndirectionsForCppType(getTypeForGLValue(type))
+}
+
+/**
+ * Gets the maximum number of indirections a value of type `type` can have.
+ *
+ * Note that this predicate is intended to be called on unspecified types
+ * (i.e., `countIndirections(e.getUnspecifiedType())`).
+ */
+private int countIndirections(Type t) {
+  result =
+    1 +
+      countIndirections([t.(Cpp::PointerType).getBaseType(), t.(Cpp::ReferenceType).getBaseType()])
+  or
+  not t instanceof Cpp::PointerType and
+  not t instanceof Cpp::ReferenceType and
+  result = 0
+}
+
+/**
+ * Gets the maximum number of indirections a value of C++
+ * type `langType` can have.
+ */
+int countIndirectionsForCppType(LanguageType langType) {
+  exists(Type type | langType.hasType(type, true) |
+    result = 1 + countIndirections(type.getUnspecifiedType())
+  )
+  or
+  exists(Type type | langType.hasType(type, false) |
+    result = countIndirections(type.getUnspecifiedType())
+  )
+}
+
+/**
+ * A `CallInstruction` that calls an allocation function such
+ * as `malloc` or `operator new`.
+ */
+class AllocationInstruction extends CallInstruction {
+  AllocationInstruction() { this.getStaticCallTarget() instanceof Cpp::AllocationFunction }
+}
+
+/**
+ * Holds if `i` is a base instruction that starts a sequence of uses
+ * of some variable that SSA can handle.
+ *
+ * This is either when `i` is a `VariableAddressInstruction` or when
+ * `i` is a fresh allocation produced by an `AllocationInstruction`.
+ */
+private predicate isSourceVariableBase(Instruction i) {
+  i instanceof VariableAddressInstruction or i instanceof AllocationInstruction
+}
+
+/**
+ * Holds if the value pointed to by `operand` can potentially be
+ * modified be the caller.
+ */
+predicate isModifiableByCall(ArgumentOperand operand) {
+  exists(CallInstruction call, int index, CppType type |
+    type = getLanguageType(operand) and
+    call.getArgumentOperand(index) = operand and
+    if index = -1
+    then not call.getStaticCallTarget() instanceof Cpp::ConstMemberFunction
+    else not SideEffects::isConstPointerLike(any(Type t | type.hasType(t, _)))
+  )
+}
+
+cached
+private module Cached {
+  /**
+   * Holds if `op` is a use of an SSA variable rooted at `base` with `ind` number
+   * of indirections.
+   *
+   * `certain` is `true` if the operand is guaranteed to read the variable, and
+   * `indirectionIndex` specifies the number of loads required to read the variable.
+   */
+  cached
+  predicate isUse(boolean certain, Operand op, Instruction base, int ind, int indirectionIndex) {
+    not ignoreOperand(op) and
+    certain = true and
+    exists(LanguageType type, int m, int ind0 |
+      type = getLanguageType(op) and
+      m = countIndirectionsForCppType(type) and
+      isUseImpl(op, base, ind0) and
+      ind = ind0 + [0 .. m] and
+      indirectionIndex = ind - ind0
+    )
+  }
+
+  /**
+   * Holds if `operand` is a use of an SSA variable rooted at `base`, and the
+   * path from `base` to `operand` passes through `ind` load-like instructions.
+   */
+  private predicate isUseImpl(Operand operand, Instruction base, int ind) {
+    DataFlowImplCommon::forceCachingInSameStage() and
+    ind = 0 and
+    operand.getDef() = base and
+    isSourceVariableBase(base)
+    or
+    exists(Operand mid, Instruction instr |
+      isUseImpl(mid, base, ind) and
+      instr = operand.getDef() and
+      conversionFlow(mid, instr, false)
+    )
+    or
+    exists(int ind0 |
+      isUseImpl(operand.getDef().(LoadInstruction).getSourceAddressOperand(), base, ind0)
+      or
+      isUseImpl(operand.getDef().(InitializeParameterInstruction).getAnOperand(), base, ind0)
+    |
+      ind0 = ind - 1
+    )
+  }
+
+  /**
+   * Holds if `address` is an address of an SSA variable rooted at `base`,
+   * and `instr` is a definition of the SSA variable with `ind` number of indirections.
+   *
+   * `certain` is `true` if `instr` is guaranteed to write to the variable, and
+   * `indirectionIndex` specifies the number of loads required to read the variable
+   * after the write operation.
+   */
+  cached
+  predicate isDef(
+    boolean certain, Instruction instr, Operand address, Instruction base, int ind,
+    int indirectionIndex
+  ) {
+    certain = true and
+    exists(int ind0, CppType type, int m |
+      address =
+        [
+          instr.(StoreInstruction).getDestinationAddressOperand(),
+          instr.(InitializeParameterInstruction).getAnOperand(),
+          instr.(InitializeDynamicAllocationInstruction).getAllocationAddressOperand(),
+          instr.(UninitializedInstruction).getAnOperand()
+        ]
+    |
+      isDefImpl(address, base, ind0) and
+      type = getLanguageType(address) and
+      m = countIndirectionsForCppType(type) and
+      ind = ind0 + [1 .. m] and
+      indirectionIndex = ind - (ind0 + 1)
+    )
+  }
+
+  /**
+   * Holds if `address` is a use of an SSA variable rooted at `base`, and the
+   * path from `base` to `address` passes through `ind` load-like instructions.
+   *
+   * Note: Unlike `isUseImpl`, this predicate recurses through pointer-arithmetic
+   * instructions.
+   */
+  private predicate isDefImpl(Operand address, Instruction base, int ind) {
+    DataFlowImplCommon::forceCachingInSameStage() and
+    ind = 0 and
+    address.getDef() = base and
+    isSourceVariableBase(base)
+    or
+    exists(Operand mid, Instruction instr |
+      isDefImpl(mid, base, ind) and
+      instr = address.getDef() and
+      conversionFlow(mid, instr, _)
+    )
+    or
+    exists(int ind0 |
+      isDefImpl(address.getDef().(LoadInstruction).getSourceAddressOperand(), base, ind0)
+      or
+      isDefImpl(address.getDef().(InitializeParameterInstruction).getAnOperand(), base, ind0)
+    |
+      ind0 = ind - 1
+    )
+  }
+}
+
+import Cached
+
+/**
+ * Inputs to the shared SSA library's parameterized module that is shared
+ * between the SSA pruning stage, and the final SSA stage.
+ */
+module InputSigCommon {
+  class BasicBlock = IRBlock;
+
+  BasicBlock getImmediateBasicBlockDominator(BasicBlock bb) { result.immediatelyDominates(bb) }
+
+  BasicBlock getABasicBlockSuccessor(BasicBlock bb) { result = bb.getASuccessor() }
+
+  class ExitBasicBlock extends IRBlock {
+    ExitBasicBlock() { this.getLastInstruction() instanceof ExitFunctionInstruction }
+  }
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/TaintTrackingUtil.qll

@@ -0,0 +1,208 @@
+private import semmle.code.cpp.ir.IR
+private import experimental.semmle.code.cpp.ir.dataflow.DataFlow
+private import ModelUtil
+private import semmle.code.cpp.models.interfaces.DataFlow
+private import semmle.code.cpp.models.interfaces.SideEffect
+private import DataFlowUtil
+private import DataFlowPrivate
+private import semmle.code.cpp.models.Models
+
+/**
+ * Holds if taint propagates from `nodeFrom` to `nodeTo` in exactly one local
+ * (intra-procedural) step.
+ */
+predicate localTaintStep(DataFlow::Node nodeFrom, DataFlow::Node nodeTo) {
+  DataFlow::localFlowStep(nodeFrom, nodeTo)
+  or
+  localAdditionalTaintStep(nodeFrom, nodeTo)
+}
+
+/**
+ * Holds if taint can flow in one local step from `nodeFrom` to `nodeTo` excluding
+ * local data flow steps. That is, `nodeFrom` and `nodeTo` are likely to represent
+ * different objects.
+ */
+cached
+predicate localAdditionalTaintStep(DataFlow::Node nodeFrom, DataFlow::Node nodeTo) {
+  operandToInstructionTaintStep(nodeFrom.asOperand(), nodeTo.asInstruction())
+  or
+  modeledTaintStep(nodeFrom, nodeTo)
+  or
+  // Flow from `op` to `*op`.
+  exists(Operand operand, int indirectionIndex |
+    nodeHasOperand(nodeFrom, operand, indirectionIndex) and
+    nodeHasOperand(nodeTo, operand, indirectionIndex - 1)
+  )
+  or
+  // Flow from `instr` to `*instr`.
+  exists(Instruction instr, int indirectionIndex |
+    nodeHasInstruction(nodeFrom, instr, indirectionIndex) and
+    nodeHasInstruction(nodeTo, instr, indirectionIndex - 1)
+  )
+  or
+  // Flow from (the indirection of) an operand of a pointer arithmetic instruction to the
+  // indirection of the pointer arithmetic instruction. This provides flow from `source`
+  // in `x[source]` to the result of the associated load instruction.
+  exists(PointerArithmeticInstruction pai, int indirectionIndex |
+    nodeHasOperand(nodeFrom, pai.getAnOperand(), pragma[only_bind_into](indirectionIndex)) and
+    hasInstructionAndIndex(nodeTo, pai, indirectionIndex + 1)
+  )
+}
+
+/**
+ * Holds if taint propagates from `nodeFrom` to `nodeTo` in exactly one local
+ * (intra-procedural) step.
+ */
+private predicate operandToInstructionTaintStep(Operand opFrom, Instruction instrTo) {
+  // Taint can flow through expressions that alter the value but preserve
+  // more than one bit of it _or_ expressions that follow data through
+  // pointer indirections.
+  instrTo.getAnOperand() = opFrom and
+  (
+    instrTo instanceof ArithmeticInstruction
+    or
+    instrTo instanceof BitwiseInstruction
+    or
+    instrTo instanceof PointerArithmeticInstruction
+  )
+  or
+  // The `CopyInstruction` case is also present in non-taint data flow, but
+  // that uses `getDef` rather than `getAnyDef`. For taint, we want flow
+  // from a definition of `myStruct` to a `myStruct.myField` expression.
+  instrTo.(LoadInstruction).getSourceAddressOperand() = opFrom
+  or
+  // Unary instructions tend to preserve enough information in practice that we
+  // want taint to flow through.
+  // The exception is `FieldAddressInstruction`. Together with the rules below for
+  // `LoadInstruction`s and `ChiInstruction`s, flow through `FieldAddressInstruction`
+  // could cause flow into one field to come out an unrelated field.
+  // This would happen across function boundaries, where the IR would not be able to
+  // match loads to stores.
+  instrTo.(UnaryInstruction).getUnaryOperand() = opFrom and
+  (
+    not instrTo instanceof FieldAddressInstruction
+    or
+    instrTo.(FieldAddressInstruction).getField().getDeclaringType() instanceof Union
+  )
+}
+
+/**
+ * Holds if taint may propagate from `source` to `sink` in zero or more local
+ * (intra-procedural) steps.
+ */
+pragma[inline]
+predicate localTaint(DataFlow::Node source, DataFlow::Node sink) { localTaintStep*(source, sink) }
+
+/**
+ * Holds if taint can flow from `i1` to `i2` in zero or more
+ * local (intra-procedural) steps.
+ */
+pragma[inline]
+predicate localInstructionTaint(Instruction i1, Instruction i2) {
+  localTaint(DataFlow::instructionNode(i1), DataFlow::instructionNode(i2))
+}
+
+/**
+ * Holds if taint can flow from `e1` to `e2` in zero or more
+ * local (intra-procedural) steps.
+ */
+pragma[inline]
+predicate localExprTaint(Expr e1, Expr e2) {
+  localTaint(DataFlow::exprNode(e1), DataFlow::exprNode(e2))
+}
+
+/**
+ * Holds if the additional step from `src` to `sink` should be included in all
+ * global taint flow configurations.
+ */
+predicate defaultAdditionalTaintStep(DataFlow::Node src, DataFlow::Node sink) {
+  localAdditionalTaintStep(src, sink)
+}
+
+/**
+ * Holds if default `TaintTracking::Configuration`s should allow implicit reads
+ * of `c` at sinks and inputs to additional taint steps.
+ */
+bindingset[node]
+predicate defaultImplicitTaintRead(DataFlow::Node node, DataFlow::Content c) { none() }
+
+/**
+ * Holds if `node` should be a sanitizer in all global taint flow configurations
+ * but not in local taint.
+ */
+predicate defaultTaintSanitizer(DataFlow::Node node) { none() }
+
+/**
+ * Holds if taint can flow from `instrIn` to `instrOut` through a call to a
+ * modeled function.
+ */
+predicate modeledTaintStep(DataFlow::Node nodeIn, DataFlow::Node nodeOut) {
+  // Normal taint steps
+  exists(CallInstruction call, TaintFunction func, FunctionInput modelIn, FunctionOutput modelOut |
+    call.getStaticCallTarget() = func and
+    func.hasTaintFlow(modelIn, modelOut)
+  |
+    (
+      nodeIn = callInput(call, modelIn)
+      or
+      exists(int n |
+        modelIn.isParameterDerefOrQualifierObject(n) and
+        if n = -1
+        then nodeIn = callInput(call, any(InQualifierAddress inQualifier))
+        else nodeIn = callInput(call, any(InParameter inParam | inParam.getIndex() = n))
+      )
+    ) and
+    nodeOut = callOutput(call, modelOut)
+    or
+    exists(int d |
+      nodeIn = callInput(call, modelIn, d)
+      or
+      exists(int n |
+        d = 1 and
+        modelIn.isParameterDerefOrQualifierObject(n) and
+        if n = -1
+        then nodeIn = callInput(call, any(InQualifierAddress inQualifier))
+        else nodeIn = callInput(call, any(InParameter inParam | inParam.getIndex() = n))
+      )
+    |
+      call.getStaticCallTarget() = func and
+      func.hasTaintFlow(modelIn, modelOut) and
+      nodeOut = callOutput(call, modelOut, d)
+    )
+  )
+  or
+  // Taint flow from one argument to another and data flow from an argument to a
+  // return value. This happens in functions like `strcat` and `memcpy`. We
+  // could model this flow in two separate steps, but that would add reverse
+  // flow from the write side-effect to the call instruction, which may not be
+  // desirable.
+  exists(
+    CallInstruction call, Function func, FunctionInput modelIn, OutParameterDeref modelMidOut,
+    int indexMid, InParameter modelMidIn, OutReturnValue modelOut
+  |
+    nodeIn = callInput(call, modelIn) and
+    nodeOut = callOutput(call, modelOut) and
+    call.getStaticCallTarget() = func and
+    func.(TaintFunction).hasTaintFlow(modelIn, modelMidOut) and
+    func.(DataFlowFunction).hasDataFlow(modelMidIn, modelOut) and
+    modelMidOut.isParameterDeref(indexMid) and
+    modelMidIn.isParameter(indexMid)
+  )
+  or
+  // Taint flow from a pointer argument to an output, when the model specifies flow from the deref
+  // to that output, but the deref is not modeled in the IR for the caller.
+  exists(
+    CallInstruction call, DataFlow::SideEffectOperandNode indirectArgument, Function func,
+    FunctionInput modelIn, FunctionOutput modelOut
+  |
+    indirectArgument = callInput(call, modelIn) and
+    indirectArgument.getAddressOperand() = nodeIn.asOperand() and
+    call.getStaticCallTarget() = func and
+    (
+      func.(DataFlowFunction).hasDataFlow(modelIn, modelOut)
+      or
+      func.(TaintFunction).hasTaintFlow(modelIn, modelOut)
+    ) and
+    nodeOut = callOutput(call, modelOut)
+  )
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/ssa0/SsaInternals.qll

@@ -0,0 +1,314 @@
+/**
+ * This module defines an initial SSA pruning stage that doesn't take
+ * indirections into account.
+ */
+
+private import codeql.ssa.Ssa as SsaImplCommon
+private import semmle.code.cpp.ir.IR
+private import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowImplCommon as DataFlowImplCommon
+private import semmle.code.cpp.models.interfaces.Allocation as Alloc
+private import semmle.code.cpp.models.interfaces.DataFlow as DataFlow
+private import semmle.code.cpp.ir.implementation.raw.internal.SideEffects as SideEffects
+private import semmle.code.cpp.ir.internal.IRCppLanguage
+private import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowPrivate
+private import experimental.semmle.code.cpp.ir.dataflow.internal.DataFlowUtil
+private import experimental.semmle.code.cpp.ir.dataflow.internal.SsaInternalsCommon
+
+private module SourceVariables {
+  newtype TBaseSourceVariable =
+    // Each IR variable gets its own source variable
+    TBaseIRVariable(IRVariable var) or
+    // Each allocation gets its own source variable
+    TBaseCallVariable(AllocationInstruction call)
+
+  abstract class BaseSourceVariable extends TBaseSourceVariable {
+    abstract string toString();
+
+    abstract DataFlowType getType();
+  }
+
+  class BaseIRVariable extends BaseSourceVariable, TBaseIRVariable {
+    IRVariable var;
+
+    IRVariable getIRVariable() { result = var }
+
+    BaseIRVariable() { this = TBaseIRVariable(var) }
+
+    override string toString() { result = var.toString() }
+
+    override DataFlowType getType() { result = var.getType() }
+  }
+
+  class BaseCallVariable extends BaseSourceVariable, TBaseCallVariable {
+    AllocationInstruction call;
+
+    BaseCallVariable() { this = TBaseCallVariable(call) }
+
+    AllocationInstruction getCallInstruction() { result = call }
+
+    override string toString() { result = call.toString() }
+
+    override DataFlowType getType() { result = call.getResultType() }
+  }
+
+  private newtype TSourceVariable =
+    TSourceIRVariable(BaseIRVariable baseVar) or
+    TCallVariable(AllocationInstruction call)
+
+  abstract class SourceVariable extends TSourceVariable {
+    abstract string toString();
+
+    abstract BaseSourceVariable getBaseVariable();
+  }
+
+  class SourceIRVariable extends SourceVariable, TSourceIRVariable {
+    BaseIRVariable var;
+
+    SourceIRVariable() { this = TSourceIRVariable(var) }
+
+    IRVariable getIRVariable() { result = var.getIRVariable() }
+
+    override BaseIRVariable getBaseVariable() { result.getIRVariable() = this.getIRVariable() }
+
+    override string toString() { result = this.getIRVariable().toString() }
+  }
+
+  class CallVariable extends SourceVariable, TCallVariable {
+    AllocationInstruction call;
+
+    CallVariable() { this = TCallVariable(call) }
+
+    AllocationInstruction getCall() { result = call }
+
+    override BaseCallVariable getBaseVariable() { result.getCallInstruction() = call }
+
+    override string toString() { result = "Call" }
+  }
+}
+
+import SourceVariables
+
+private newtype TDefOrUseImpl =
+  TDefImpl(Operand address) { isDef(_, _, address, _, _, _) } or
+  TUseImpl(Operand operand) {
+    isUse(_, operand, _, _, _) and
+    not isDef(_, _, operand, _, _, _)
+  }
+
+abstract private class DefOrUseImpl extends TDefOrUseImpl {
+  /** Gets a textual representation of this element. */
+  abstract string toString();
+
+  /** Gets the block of this definition or use. */
+  abstract IRBlock getBlock();
+
+  /** Holds if this definition or use has index `index` in block `block`. */
+  abstract predicate hasIndexInBlock(IRBlock block, int index);
+
+  final predicate hasIndexInBlock(IRBlock block, int index, SourceVariable sv) {
+    this.hasIndexInBlock(block, index) and
+    sv = this.getSourceVariable()
+  }
+
+  /** Gets the location of this element. */
+  abstract Cpp::Location getLocation();
+
+  abstract Instruction getBase();
+
+  final BaseSourceVariable getBaseSourceVariable() {
+    exists(IRVariable var |
+      result.(BaseIRVariable).getIRVariable() = var and
+      instructionHasIRVariable(this.getBase(), var)
+    )
+    or
+    result.(BaseCallVariable).getCallInstruction() = this.getBase()
+  }
+
+  /** Gets the variable that is defined or used. */
+  final SourceVariable getSourceVariable() {
+    exists(BaseSourceVariable v |
+      sourceVariableHasBaseAndIndex(result, v) and
+      defOrUseHasSourceVariable(this, v)
+    )
+  }
+}
+
+pragma[noinline]
+private predicate instructionHasIRVariable(VariableAddressInstruction vai, IRVariable var) {
+  vai.getIRVariable() = var
+}
+
+private predicate defOrUseHasSourceVariable(DefOrUseImpl defOrUse, BaseSourceVariable bv) {
+  defHasSourceVariable(defOrUse, bv)
+  or
+  useHasSourceVariable(defOrUse, bv)
+}
+
+pragma[noinline]
+private predicate defHasSourceVariable(DefImpl def, BaseSourceVariable bv) {
+  bv = def.getBaseSourceVariable()
+}
+
+pragma[noinline]
+private predicate useHasSourceVariable(UseImpl use, BaseSourceVariable bv) {
+  bv = use.getBaseSourceVariable()
+}
+
+pragma[noinline]
+private predicate sourceVariableHasBaseAndIndex(SourceVariable v, BaseSourceVariable bv) {
+  v.getBaseVariable() = bv
+}
+
+class DefImpl extends DefOrUseImpl, TDefImpl {
+  Operand address;
+
+  DefImpl() { this = TDefImpl(address) }
+
+  override Instruction getBase() { isDef(_, _, address, result, _, _) }
+
+  Operand getAddressOperand() { result = address }
+
+  Instruction getDefiningInstruction() { isDef(_, result, address, _, _, _) }
+
+  override string toString() { result = address.toString() }
+
+  override IRBlock getBlock() { result = this.getDefiningInstruction().getBlock() }
+
+  override Cpp::Location getLocation() { result = this.getDefiningInstruction().getLocation() }
+
+  final override predicate hasIndexInBlock(IRBlock block, int index) {
+    this.getDefiningInstruction() = block.getInstruction(index)
+  }
+
+  predicate isCertain() { isDef(true, _, address, _, _, _) }
+}
+
+class UseImpl extends DefOrUseImpl, TUseImpl {
+  Operand operand;
+
+  UseImpl() { this = TUseImpl(operand) }
+
+  Operand getOperand() { result = operand }
+
+  override string toString() { result = operand.toString() }
+
+  final override predicate hasIndexInBlock(IRBlock block, int index) {
+    operand.getUse() = block.getInstruction(index)
+  }
+
+  final override IRBlock getBlock() { result = operand.getUse().getBlock() }
+
+  final override Cpp::Location getLocation() { result = operand.getLocation() }
+
+  override Instruction getBase() { isUse(_, operand, result, _, _) }
+
+  predicate isCertain() { isUse(true, operand, _, _, _) }
+}
+
+private module SsaInput implements SsaImplCommon::InputSig {
+  import InputSigCommon
+  import SourceVariables
+
+  /**
+   * Holds if the `i`'th write in block `bb` writes to the variable `v`.
+   * `certain` is `true` if the write is guaranteed to overwrite the entire variable.
+   */
+  predicate variableWrite(IRBlock bb, int i, SourceVariable v, boolean certain) {
+    DataFlowImplCommon::forceCachingInSameStage() and
+    exists(DefImpl def | def.hasIndexInBlock(bb, i, v) |
+      if def.isCertain() then certain = true else certain = false
+    )
+  }
+
+  /**
+   * Holds if the `i`'th read in block `bb` reads to the variable `v`.
+   * `certain` is `true` if the read is guaranteed.
+   */
+  predicate variableRead(IRBlock bb, int i, SourceVariable v, boolean certain) {
+    exists(UseImpl use | use.hasIndexInBlock(bb, i, v) |
+      if use.isCertain() then certain = true else certain = false
+    )
+  }
+}
+
+private newtype TSsaDefOrUse =
+  TDefOrUse(DefOrUseImpl defOrUse) {
+    defOrUse instanceof UseImpl
+    or
+    // If `defOrUse` is a definition we only include it if the
+    // SSA library concludes that it's live after the write.
+    exists(Definition def, SourceVariable sv, IRBlock bb, int i |
+      def.definesAt(sv, bb, i) and
+      defOrUse.(DefImpl).hasIndexInBlock(bb, i, sv)
+    )
+  } or
+  TPhi(PhiNode phi)
+
+abstract private class SsaDefOrUse extends TSsaDefOrUse {
+  string toString() { result = "SsaDefOrUse" }
+
+  DefOrUseImpl asDefOrUse() { none() }
+
+  PhiNode asPhi() { none() }
+
+  abstract Location getLocation();
+}
+
+class DefOrUse extends TDefOrUse, SsaDefOrUse {
+  DefOrUseImpl defOrUse;
+
+  DefOrUse() { this = TDefOrUse(defOrUse) }
+
+  final override DefOrUseImpl asDefOrUse() { result = defOrUse }
+
+  final override Location getLocation() { result = defOrUse.getLocation() }
+
+  final SourceVariable getSourceVariable() { result = defOrUse.getSourceVariable() }
+}
+
+class Phi extends TPhi, SsaDefOrUse {
+  PhiNode phi;
+
+  Phi() { this = TPhi(phi) }
+
+  final override PhiNode asPhi() { result = phi }
+
+  final override Location getLocation() { result = phi.getBasicBlock().getLocation() }
+}
+
+class UseOrPhi extends SsaDefOrUse {
+  UseOrPhi() {
+    this.asDefOrUse() instanceof UseImpl
+    or
+    this instanceof Phi
+  }
+
+  final override Location getLocation() {
+    result = this.asDefOrUse().getLocation() or result = this.(Phi).getLocation()
+  }
+
+  override string toString() {
+    result = this.asDefOrUse().toString()
+    or
+    this instanceof Phi and
+    result = "Phi"
+  }
+}
+
+class Def extends DefOrUse {
+  override DefImpl defOrUse;
+
+  Operand getAddressOperand() { result = defOrUse.getAddressOperand() }
+
+  Instruction getAddress() { result = this.getAddressOperand().getDef() }
+
+  Instruction getDefiningInstruction() { result = defOrUse.getDefiningInstruction() }
+
+  override string toString() { result = this.asDefOrUse().toString() + " (def)" }
+}
+
+private module SsaImpl = SsaImplCommon::Make<SsaInput>;
+
+class PhiNode = SsaImpl::PhiNode;
+
+class Definition = SsaImpl::Definition;

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/tainttracking1/TaintTrackingImpl.qll

@@ -0,0 +1,191 @@
+/**
+ * Provides an implementation of global (interprocedural) taint tracking.
+ * This file re-exports the local (intraprocedural) taint-tracking analysis
+ * from `TaintTrackingParameter::Public` and adds a global analysis, mainly
+ * exposed through the `Configuration` class. For some languages, this file
+ * exists in several identical copies, allowing queries to use multiple
+ * `Configuration` classes that depend on each other without introducing
+ * mutual recursion among those configurations.
+ */
+
+import TaintTrackingParameter::Public
+private import TaintTrackingParameter::Private
+
+/**
+ * A configuration of interprocedural taint tracking analysis. This defines
+ * sources, sinks, and any other configurable aspect of the analysis. Each
+ * use of the taint tracking library must define its own unique extension of
+ * this abstract class.
+ *
+ * A taint-tracking configuration is a special data flow configuration
+ * (`DataFlow::Configuration`) that allows for flow through nodes that do not
+ * necessarily preserve values but are still relevant from a taint tracking
+ * perspective. (For example, string concatenation, where one of the operands
+ * is tainted.)
+ *
+ * To create a configuration, extend this class with a subclass whose
+ * characteristic predicate is a unique singleton string. For example, write
+ *
+ * ```ql
+ * class MyAnalysisConfiguration extends TaintTracking::Configuration {
+ *   MyAnalysisConfiguration() { this = "MyAnalysisConfiguration" }
+ *   // Override `isSource` and `isSink`.
+ *   // Optionally override `isSanitizer`.
+ *   // Optionally override `isSanitizerIn`.
+ *   // Optionally override `isSanitizerOut`.
+ *   // Optionally override `isSanitizerGuard`.
+ *   // Optionally override `isAdditionalTaintStep`.
+ * }
+ * ```
+ *
+ * Then, to query whether there is flow between some `source` and `sink`,
+ * write
+ *
+ * ```ql
+ * exists(MyAnalysisConfiguration cfg | cfg.hasFlow(source, sink))
+ * ```
+ *
+ * Multiple configurations can coexist, but it is unsupported to depend on
+ * another `TaintTracking::Configuration` or a `DataFlow::Configuration` in the
+ * overridden predicates that define sources, sinks, or additional steps.
+ * Instead, the dependency should go to a `TaintTracking2::Configuration` or a
+ * `DataFlow2::Configuration`, `DataFlow3::Configuration`, etc.
+ */
+abstract class Configuration extends DataFlow::Configuration {
+  bindingset[this]
+  Configuration() { any() }
+
+  /**
+   * Holds if `source` is a relevant taint source.
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSource(DataFlow::Node source) { none() }
+
+  /**
+   * Holds if `source` is a relevant taint source with the given initial
+   * `state`.
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSource(DataFlow::Node source, DataFlow::FlowState state) { none() }
+
+  /**
+   * Holds if `sink` is a relevant taint sink
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSink(DataFlow::Node sink) { none() }
+
+  /**
+   * Holds if `sink` is a relevant taint sink accepting `state`.
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSink(DataFlow::Node sink, DataFlow::FlowState state) { none() }
+
+  /** Holds if the node `node` is a taint sanitizer. */
+  predicate isSanitizer(DataFlow::Node node) { none() }
+
+  final override predicate isBarrier(DataFlow::Node node) {
+    this.isSanitizer(node) or
+    defaultTaintSanitizer(node)
+  }
+
+  /**
+   * Holds if the node `node` is a taint sanitizer when the flow state is
+   * `state`.
+   */
+  predicate isSanitizer(DataFlow::Node node, DataFlow::FlowState state) { none() }
+
+  final override predicate isBarrier(DataFlow::Node node, DataFlow::FlowState state) {
+    this.isSanitizer(node, state)
+  }
+
+  /** Holds if taint propagation into `node` is prohibited. */
+  predicate isSanitizerIn(DataFlow::Node node) { none() }
+
+  final override predicate isBarrierIn(DataFlow::Node node) { this.isSanitizerIn(node) }
+
+  /** Holds if taint propagation out of `node` is prohibited. */
+  predicate isSanitizerOut(DataFlow::Node node) { none() }
+
+  final override predicate isBarrierOut(DataFlow::Node node) { this.isSanitizerOut(node) }
+
+  /**
+   * DEPRECATED: Use `isSanitizer` and `BarrierGuard` module instead.
+   *
+   * Holds if taint propagation through nodes guarded by `guard` is prohibited.
+   */
+  deprecated predicate isSanitizerGuard(DataFlow::BarrierGuard guard) { none() }
+
+  deprecated final override predicate isBarrierGuard(DataFlow::BarrierGuard guard) {
+    this.isSanitizerGuard(guard)
+  }
+
+  /**
+   * DEPRECATED: Use `isSanitizer` and `BarrierGuard` module instead.
+   *
+   * Holds if taint propagation through nodes guarded by `guard` is prohibited
+   * when the flow state is `state`.
+   */
+  deprecated predicate isSanitizerGuard(DataFlow::BarrierGuard guard, DataFlow::FlowState state) {
+    none()
+  }
+
+  deprecated final override predicate isBarrierGuard(
+    DataFlow::BarrierGuard guard, DataFlow::FlowState state
+  ) {
+    this.isSanitizerGuard(guard, state)
+  }
+
+  /**
+   * Holds if taint may propagate from `node1` to `node2` in addition to the normal data-flow and taint steps.
+   */
+  predicate isAdditionalTaintStep(DataFlow::Node node1, DataFlow::Node node2) { none() }
+
+  final override predicate isAdditionalFlowStep(DataFlow::Node node1, DataFlow::Node node2) {
+    this.isAdditionalTaintStep(node1, node2) or
+    defaultAdditionalTaintStep(node1, node2)
+  }
+
+  /**
+   * Holds if taint may propagate from `node1` to `node2` in addition to the normal data-flow and taint steps.
+   * This step is only applicable in `state1` and updates the flow state to `state2`.
+   */
+  predicate isAdditionalTaintStep(
+    DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+    DataFlow::FlowState state2
+  ) {
+    none()
+  }
+
+  final override predicate isAdditionalFlowStep(
+    DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+    DataFlow::FlowState state2
+  ) {
+    this.isAdditionalTaintStep(node1, state1, node2, state2)
+  }
+
+  override predicate allowImplicitRead(DataFlow::Node node, DataFlow::ContentSet c) {
+    (
+      this.isSink(node) or
+      this.isSink(node, _) or
+      this.isAdditionalTaintStep(node, _) or
+      this.isAdditionalTaintStep(node, _, _, _)
+    ) and
+    defaultImplicitTaintRead(node, c)
+  }
+
+  /**
+   * Holds if taint may flow from `source` to `sink` for this configuration.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate hasFlow(DataFlow::Node source, DataFlow::Node sink) {
+    super.hasFlow(source, sink)
+  }
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/tainttracking1/TaintTrackingParameter.qll

@@ -0,0 +1,5 @@
+import experimental.semmle.code.cpp.ir.dataflow.internal.TaintTrackingUtil as Public
+
+module Private {
+  import experimental.semmle.code.cpp.ir.dataflow.DataFlow::DataFlow as DataFlow
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/tainttracking2/TaintTrackingImpl.qll

@@ -0,0 +1,191 @@
+/**
+ * Provides an implementation of global (interprocedural) taint tracking.
+ * This file re-exports the local (intraprocedural) taint-tracking analysis
+ * from `TaintTrackingParameter::Public` and adds a global analysis, mainly
+ * exposed through the `Configuration` class. For some languages, this file
+ * exists in several identical copies, allowing queries to use multiple
+ * `Configuration` classes that depend on each other without introducing
+ * mutual recursion among those configurations.
+ */
+
+import TaintTrackingParameter::Public
+private import TaintTrackingParameter::Private
+
+/**
+ * A configuration of interprocedural taint tracking analysis. This defines
+ * sources, sinks, and any other configurable aspect of the analysis. Each
+ * use of the taint tracking library must define its own unique extension of
+ * this abstract class.
+ *
+ * A taint-tracking configuration is a special data flow configuration
+ * (`DataFlow::Configuration`) that allows for flow through nodes that do not
+ * necessarily preserve values but are still relevant from a taint tracking
+ * perspective. (For example, string concatenation, where one of the operands
+ * is tainted.)
+ *
+ * To create a configuration, extend this class with a subclass whose
+ * characteristic predicate is a unique singleton string. For example, write
+ *
+ * ```ql
+ * class MyAnalysisConfiguration extends TaintTracking::Configuration {
+ *   MyAnalysisConfiguration() { this = "MyAnalysisConfiguration" }
+ *   // Override `isSource` and `isSink`.
+ *   // Optionally override `isSanitizer`.
+ *   // Optionally override `isSanitizerIn`.
+ *   // Optionally override `isSanitizerOut`.
+ *   // Optionally override `isSanitizerGuard`.
+ *   // Optionally override `isAdditionalTaintStep`.
+ * }
+ * ```
+ *
+ * Then, to query whether there is flow between some `source` and `sink`,
+ * write
+ *
+ * ```ql
+ * exists(MyAnalysisConfiguration cfg | cfg.hasFlow(source, sink))
+ * ```
+ *
+ * Multiple configurations can coexist, but it is unsupported to depend on
+ * another `TaintTracking::Configuration` or a `DataFlow::Configuration` in the
+ * overridden predicates that define sources, sinks, or additional steps.
+ * Instead, the dependency should go to a `TaintTracking2::Configuration` or a
+ * `DataFlow2::Configuration`, `DataFlow3::Configuration`, etc.
+ */
+abstract class Configuration extends DataFlow::Configuration {
+  bindingset[this]
+  Configuration() { any() }
+
+  /**
+   * Holds if `source` is a relevant taint source.
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSource(DataFlow::Node source) { none() }
+
+  /**
+   * Holds if `source` is a relevant taint source with the given initial
+   * `state`.
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSource(DataFlow::Node source, DataFlow::FlowState state) { none() }
+
+  /**
+   * Holds if `sink` is a relevant taint sink
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSink(DataFlow::Node sink) { none() }
+
+  /**
+   * Holds if `sink` is a relevant taint sink accepting `state`.
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSink(DataFlow::Node sink, DataFlow::FlowState state) { none() }
+
+  /** Holds if the node `node` is a taint sanitizer. */
+  predicate isSanitizer(DataFlow::Node node) { none() }
+
+  final override predicate isBarrier(DataFlow::Node node) {
+    this.isSanitizer(node) or
+    defaultTaintSanitizer(node)
+  }
+
+  /**
+   * Holds if the node `node` is a taint sanitizer when the flow state is
+   * `state`.
+   */
+  predicate isSanitizer(DataFlow::Node node, DataFlow::FlowState state) { none() }
+
+  final override predicate isBarrier(DataFlow::Node node, DataFlow::FlowState state) {
+    this.isSanitizer(node, state)
+  }
+
+  /** Holds if taint propagation into `node` is prohibited. */
+  predicate isSanitizerIn(DataFlow::Node node) { none() }
+
+  final override predicate isBarrierIn(DataFlow::Node node) { this.isSanitizerIn(node) }
+
+  /** Holds if taint propagation out of `node` is prohibited. */
+  predicate isSanitizerOut(DataFlow::Node node) { none() }
+
+  final override predicate isBarrierOut(DataFlow::Node node) { this.isSanitizerOut(node) }
+
+  /**
+   * DEPRECATED: Use `isSanitizer` and `BarrierGuard` module instead.
+   *
+   * Holds if taint propagation through nodes guarded by `guard` is prohibited.
+   */
+  deprecated predicate isSanitizerGuard(DataFlow::BarrierGuard guard) { none() }
+
+  deprecated final override predicate isBarrierGuard(DataFlow::BarrierGuard guard) {
+    this.isSanitizerGuard(guard)
+  }
+
+  /**
+   * DEPRECATED: Use `isSanitizer` and `BarrierGuard` module instead.
+   *
+   * Holds if taint propagation through nodes guarded by `guard` is prohibited
+   * when the flow state is `state`.
+   */
+  deprecated predicate isSanitizerGuard(DataFlow::BarrierGuard guard, DataFlow::FlowState state) {
+    none()
+  }
+
+  deprecated final override predicate isBarrierGuard(
+    DataFlow::BarrierGuard guard, DataFlow::FlowState state
+  ) {
+    this.isSanitizerGuard(guard, state)
+  }
+
+  /**
+   * Holds if taint may propagate from `node1` to `node2` in addition to the normal data-flow and taint steps.
+   */
+  predicate isAdditionalTaintStep(DataFlow::Node node1, DataFlow::Node node2) { none() }
+
+  final override predicate isAdditionalFlowStep(DataFlow::Node node1, DataFlow::Node node2) {
+    this.isAdditionalTaintStep(node1, node2) or
+    defaultAdditionalTaintStep(node1, node2)
+  }
+
+  /**
+   * Holds if taint may propagate from `node1` to `node2` in addition to the normal data-flow and taint steps.
+   * This step is only applicable in `state1` and updates the flow state to `state2`.
+   */
+  predicate isAdditionalTaintStep(
+    DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+    DataFlow::FlowState state2
+  ) {
+    none()
+  }
+
+  final override predicate isAdditionalFlowStep(
+    DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+    DataFlow::FlowState state2
+  ) {
+    this.isAdditionalTaintStep(node1, state1, node2, state2)
+  }
+
+  override predicate allowImplicitRead(DataFlow::Node node, DataFlow::ContentSet c) {
+    (
+      this.isSink(node) or
+      this.isSink(node, _) or
+      this.isAdditionalTaintStep(node, _) or
+      this.isAdditionalTaintStep(node, _, _, _)
+    ) and
+    defaultImplicitTaintRead(node, c)
+  }
+
+  /**
+   * Holds if taint may flow from `source` to `sink` for this configuration.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate hasFlow(DataFlow::Node source, DataFlow::Node sink) {
+    super.hasFlow(source, sink)
+  }
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/tainttracking2/TaintTrackingParameter.qll

@@ -0,0 +1,5 @@
+import experimental.semmle.code.cpp.ir.dataflow.internal.TaintTrackingUtil as Public
+
+module Private {
+  import experimental.semmle.code.cpp.ir.dataflow.DataFlow2::DataFlow2 as DataFlow
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/tainttracking3/TaintTrackingImpl.qll

@@ -0,0 +1,191 @@
+/**
+ * Provides an implementation of global (interprocedural) taint tracking.
+ * This file re-exports the local (intraprocedural) taint-tracking analysis
+ * from `TaintTrackingParameter::Public` and adds a global analysis, mainly
+ * exposed through the `Configuration` class. For some languages, this file
+ * exists in several identical copies, allowing queries to use multiple
+ * `Configuration` classes that depend on each other without introducing
+ * mutual recursion among those configurations.
+ */
+
+import TaintTrackingParameter::Public
+private import TaintTrackingParameter::Private
+
+/**
+ * A configuration of interprocedural taint tracking analysis. This defines
+ * sources, sinks, and any other configurable aspect of the analysis. Each
+ * use of the taint tracking library must define its own unique extension of
+ * this abstract class.
+ *
+ * A taint-tracking configuration is a special data flow configuration
+ * (`DataFlow::Configuration`) that allows for flow through nodes that do not
+ * necessarily preserve values but are still relevant from a taint tracking
+ * perspective. (For example, string concatenation, where one of the operands
+ * is tainted.)
+ *
+ * To create a configuration, extend this class with a subclass whose
+ * characteristic predicate is a unique singleton string. For example, write
+ *
+ * ```ql
+ * class MyAnalysisConfiguration extends TaintTracking::Configuration {
+ *   MyAnalysisConfiguration() { this = "MyAnalysisConfiguration" }
+ *   // Override `isSource` and `isSink`.
+ *   // Optionally override `isSanitizer`.
+ *   // Optionally override `isSanitizerIn`.
+ *   // Optionally override `isSanitizerOut`.
+ *   // Optionally override `isSanitizerGuard`.
+ *   // Optionally override `isAdditionalTaintStep`.
+ * }
+ * ```
+ *
+ * Then, to query whether there is flow between some `source` and `sink`,
+ * write
+ *
+ * ```ql
+ * exists(MyAnalysisConfiguration cfg | cfg.hasFlow(source, sink))
+ * ```
+ *
+ * Multiple configurations can coexist, but it is unsupported to depend on
+ * another `TaintTracking::Configuration` or a `DataFlow::Configuration` in the
+ * overridden predicates that define sources, sinks, or additional steps.
+ * Instead, the dependency should go to a `TaintTracking2::Configuration` or a
+ * `DataFlow2::Configuration`, `DataFlow3::Configuration`, etc.
+ */
+abstract class Configuration extends DataFlow::Configuration {
+  bindingset[this]
+  Configuration() { any() }
+
+  /**
+   * Holds if `source` is a relevant taint source.
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSource(DataFlow::Node source) { none() }
+
+  /**
+   * Holds if `source` is a relevant taint source with the given initial
+   * `state`.
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSource(DataFlow::Node source, DataFlow::FlowState state) { none() }
+
+  /**
+   * Holds if `sink` is a relevant taint sink
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSink(DataFlow::Node sink) { none() }
+
+  /**
+   * Holds if `sink` is a relevant taint sink accepting `state`.
+   *
+   * The smaller this predicate is, the faster `hasFlow()` will converge.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate isSink(DataFlow::Node sink, DataFlow::FlowState state) { none() }
+
+  /** Holds if the node `node` is a taint sanitizer. */
+  predicate isSanitizer(DataFlow::Node node) { none() }
+
+  final override predicate isBarrier(DataFlow::Node node) {
+    this.isSanitizer(node) or
+    defaultTaintSanitizer(node)
+  }
+
+  /**
+   * Holds if the node `node` is a taint sanitizer when the flow state is
+   * `state`.
+   */
+  predicate isSanitizer(DataFlow::Node node, DataFlow::FlowState state) { none() }
+
+  final override predicate isBarrier(DataFlow::Node node, DataFlow::FlowState state) {
+    this.isSanitizer(node, state)
+  }
+
+  /** Holds if taint propagation into `node` is prohibited. */
+  predicate isSanitizerIn(DataFlow::Node node) { none() }
+
+  final override predicate isBarrierIn(DataFlow::Node node) { this.isSanitizerIn(node) }
+
+  /** Holds if taint propagation out of `node` is prohibited. */
+  predicate isSanitizerOut(DataFlow::Node node) { none() }
+
+  final override predicate isBarrierOut(DataFlow::Node node) { this.isSanitizerOut(node) }
+
+  /**
+   * DEPRECATED: Use `isSanitizer` and `BarrierGuard` module instead.
+   *
+   * Holds if taint propagation through nodes guarded by `guard` is prohibited.
+   */
+  deprecated predicate isSanitizerGuard(DataFlow::BarrierGuard guard) { none() }
+
+  deprecated final override predicate isBarrierGuard(DataFlow::BarrierGuard guard) {
+    this.isSanitizerGuard(guard)
+  }
+
+  /**
+   * DEPRECATED: Use `isSanitizer` and `BarrierGuard` module instead.
+   *
+   * Holds if taint propagation through nodes guarded by `guard` is prohibited
+   * when the flow state is `state`.
+   */
+  deprecated predicate isSanitizerGuard(DataFlow::BarrierGuard guard, DataFlow::FlowState state) {
+    none()
+  }
+
+  deprecated final override predicate isBarrierGuard(
+    DataFlow::BarrierGuard guard, DataFlow::FlowState state
+  ) {
+    this.isSanitizerGuard(guard, state)
+  }
+
+  /**
+   * Holds if taint may propagate from `node1` to `node2` in addition to the normal data-flow and taint steps.
+   */
+  predicate isAdditionalTaintStep(DataFlow::Node node1, DataFlow::Node node2) { none() }
+
+  final override predicate isAdditionalFlowStep(DataFlow::Node node1, DataFlow::Node node2) {
+    this.isAdditionalTaintStep(node1, node2) or
+    defaultAdditionalTaintStep(node1, node2)
+  }
+
+  /**
+   * Holds if taint may propagate from `node1` to `node2` in addition to the normal data-flow and taint steps.
+   * This step is only applicable in `state1` and updates the flow state to `state2`.
+   */
+  predicate isAdditionalTaintStep(
+    DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+    DataFlow::FlowState state2
+  ) {
+    none()
+  }
+
+  final override predicate isAdditionalFlowStep(
+    DataFlow::Node node1, DataFlow::FlowState state1, DataFlow::Node node2,
+    DataFlow::FlowState state2
+  ) {
+    this.isAdditionalTaintStep(node1, state1, node2, state2)
+  }
+
+  override predicate allowImplicitRead(DataFlow::Node node, DataFlow::ContentSet c) {
+    (
+      this.isSink(node) or
+      this.isSink(node, _) or
+      this.isAdditionalTaintStep(node, _) or
+      this.isAdditionalTaintStep(node, _, _, _)
+    ) and
+    defaultImplicitTaintRead(node, c)
+  }
+
+  /**
+   * Holds if taint may flow from `source` to `sink` for this configuration.
+   */
+  // overridden to provide taint-tracking specific qldoc
+  override predicate hasFlow(DataFlow::Node source, DataFlow::Node sink) {
+    super.hasFlow(source, sink)
+  }
+}

cpp/ql/lib/experimental/semmle/code/cpp/ir/dataflow/internal/tainttracking3/TaintTrackingParameter.qll

@@ -0,0 +1,5 @@
+import experimental.semmle.code.cpp.ir.dataflow.internal.TaintTrackingUtil as Public
+
+module Private {
+  import experimental.semmle.code.cpp.ir.dataflow.DataFlow3::DataFlow3 as DataFlow
+}

cpp/ql/lib/experimental/semmle/code/cpp/models/interfaces/SimpleRangeAnalysisDefinition.qll

@@ -37,7 +37,7 @@ abstract class SimpleRangeAnalysisDefinition extends RangeSsaDefinition {
    * dependencies. Without this information, range analysis might work for
    * simple cases but will go into infinite loops on complex code.
    *
-   * For example, when modelling the definition by reference in a call to an
+   * For example, when modeling the definition by reference in a call to an
    * overloaded `operator=`, written as `v = e`, the definition of `(this, v)`
    * depends on `e`.
    */

cpp/ql/lib/experimental/semmle/code/cpp/rangeanalysis/Bound.qll

@@ -28,6 +28,10 @@ private newtype TBound =
       i.(LoadInstruction).getSourceAddress() instanceof FieldAddressInstruction
       or
       i.getAUse() instanceof ArgumentOperand
+      or
+      i instanceof PointerArithmeticInstruction
+      or
+      i.getAUse() instanceof AddressOperand
     )
   }
 
@@ -73,7 +77,7 @@ class ValueNumberBound extends Bound, TBoundValueNumber {
     this = TBoundValueNumber(valueNumber(result)) and delta = 0
   }
 
-  override string toString() { result = vn.getExampleInstruction().toString() }
+  override string toString() { result = "ValueNumberBound" }
 
   override Location getLocation() { result = vn.getLocation() }
 

cpp/ql/lib/experimental/semmle/code/cpp/rangeanalysis/InBoundsPointerDeref.qll

@@ -5,7 +5,7 @@
  * `Instruction` level), and then using the array length analysis and the range
  * analysis together to prove that some of these pointer dereferences are safe.
  *
- * The analysis is soundy, i.e. it is sound if no undefined behaviour is present
+ * The analysis is soundy, i.e. it is sound if no undefined behavior is present
  * in the program.
  * Furthermore, it crucially depends on the soundiness of the range analysis and
  * the array length analysis.

cpp/ql/lib/experimental/semmle/code/cpp/security/PrivateCleartextWrite.qll

@@ -4,7 +4,7 @@
 
 import cpp
 import semmle.code.cpp.dataflow.TaintTracking
-import experimental.semmle.code.cpp.security.PrivateData
+import semmle.code.cpp.security.PrivateData
 import semmle.code.cpp.security.FileWrite
 import semmle.code.cpp.security.BufferWrite
 

cpp/ql/lib/experimental/semmle/code/cpp/security/PrivateData.qll

@@ -1,52 +0,0 @@
-/**
- * Provides classes and predicates for identifying private data and functions for security.
- *
- * 'Private' data in general is anything that would compromise user privacy if exposed. This
- * library tries to guess where private data may either be stored in a variable or produced by a
- * function.
- *
- * This library is not concerned with credentials. See `SensitiveActions` for expressions related
- * to credentials.
- */
-
-import cpp
-
-/** A string for `match` that identifies strings that look like they represent private data. */
-private string privateNames() {
-  result =
-    [
-      // Inspired by the list on https://cwe.mitre.org/data/definitions/359.html
-      // Government identifiers, such as Social Security Numbers
-      "%social%security%number%",
-      // Contact information, such as home addresses and telephone numbers
-      "%postcode%", "%zipcode%",
-      // result = "%telephone%" or
-      // Geographic location - where the user is (or was)
-      "%latitude%", "%longitude%",
-      // Financial data - such as credit card numbers, salary, bank accounts, and debts
-      "%creditcard%", "%salary%", "%bankaccount%",
-      // Communications - e-mail addresses, private e-mail messages, SMS text messages, chat logs, etc.
-      // result = "%email%" or
-      // result = "%mobile%" or
-      "%employer%",
-      // Health - medical conditions, insurance status, prescription records
-      "%medical%"
-    ]
-}
-
-/** An expression that might contain private data. */
-abstract class PrivateDataExpr extends Expr { }
-
-/** A functiond call that might produce private data. */
-class PrivateFunctionCall extends PrivateDataExpr, FunctionCall {
-  PrivateFunctionCall() {
-    exists(string s | this.getTarget().getName().toLowerCase() = s | s.matches(privateNames()))
-  }
-}
-
-/** An access to a variable that might contain private data. */
-class PrivateVariableAccess extends PrivateDataExpr, VariableAccess {
-  PrivateVariableAccess() {
-    exists(string s | this.getTarget().getName().toLowerCase() = s | s.matches(privateNames()))
-  }
-}

cpp/ql/lib/experimental/semmle/code/cpp/semantic/Semantic.qll

@@ -0,0 +1,7 @@
+import SemanticExpr
+import SemanticBound
+import SemanticSSA
+import SemanticGuard
+import SemanticCFG
+import SemanticType
+import SemanticOpcode

cpp/ql/lib/experimental/semmle/code/cpp/semantic/SemanticBound.qll

@@ -0,0 +1,42 @@
+/**
+ * Semantic wrapper around the language-specific bounds library.
+ */
+
+private import SemanticExpr
+private import SemanticExprSpecific::SemanticExprConfig as Specific
+private import SemanticSSA
+
+/**
+ * A valid base for an expression bound.
+ *
+ * Can be either a variable (`SemSsaBound`) or zero (`SemZeroBound`).
+ */
+class SemBound instanceof Specific::Bound {
+  final string toString() { result = super.toString() }
+
+  final SemExpr getExpr(int delta) { result = Specific::getBoundExpr(this, delta) }
+}
+
+/**
+ * A bound that is a constant zero.
+ */
+class SemZeroBound extends SemBound {
+  SemZeroBound() { Specific::zeroBound(this) }
+}
+
+/**
+ * A bound that is an SSA definition.
+ */
+class SemSsaBound extends SemBound {
+  /**
+   * The variables whose value is used as the bound.
+   *
+   * Can be multi-valued in some implementations. If so, all variables will be equivalent.
+   */
+  SemSsaVariable var;
+
+  SemSsaBound() { Specific::ssaBound(this, var) }
+
+  /** Gets a variable whose value is used as the bound. */
+  final SemSsaVariable getAVariable() { result = var }
+}