Revert "Merge pull request #5829 from MathiasVP/reorder-get-instruction-opcode"

This reverts commit 390ee3a6b8, reversing
changes made to f03c99ab03.

.github/workflows/close-stale.yml

@@ -0,0 +1,30 @@
+name: Mark stale issues
+
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: "30 1 * * *"
+
+jobs:
+  stale:
+    if: github.repository == 'github/codeql'
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/stale@v3
+      with:
+        repo-token: ${{ secrets.GITHUB_TOKEN }}
+        stale-issue-message: 'This issue is stale because it has been open 14 days with no activity. Comment or remove the `Stale` label in order to avoid having this issue closed in 7 days.'
+        close-issue-message: 'This issue was closed because it has been inactive for 7 days.'
+        days-before-stale: 14
+        days-before-close: 7
+        only-labels: awaiting-response
+
+        # do not mark PRs as stale
+        days-before-pr-stale: -1
+        days-before-pr-close: -1
+
+        # Uncomment for dry-run
+        # debug-only: true
+        # operations-per-run: 1000

.github/workflows/codeql-analysis.yml

@@ -19,13 +19,18 @@ jobs:
 
     runs-on: ubuntu-latest
 
+    permissions:
+      contents: read
+      security-events: write
+      pull-requests: read
+
     steps:
     - name: Checkout repository
       uses: actions/checkout@v2
 
     # Initializes the CodeQL tools for scanning.
     - name: Initialize CodeQL
-      uses: github/codeql-action/init@v1
+      uses: github/codeql-action/init@main
       # Override language selection by uncommenting this and choosing your languages
       with:
         languages: csharp
@@ -34,7 +39,7 @@ jobs:
     # Autobuild attempts to build any compiled languages  (C/C++, C#, or Java).
     # If this step fails, then you should remove it and run the build manually (see below)
     - name: Autobuild
-      uses: github/codeql-action/autobuild@v1
+      uses: github/codeql-action/autobuild@main
 
     # ℹ️ Command-line programs to run using the OS shell.
     # 📚 https://git.io/JvXDl
@@ -48,4 +53,4 @@ jobs:
     #   make release
 
     - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@v1
+      uses: github/codeql-action/analyze@main

.github/workflows/docs-review.yml

@@ -1,29 +0,0 @@
-# When a PR is labelled with 'ready-for-docs-review',
-# this workflow comments on the PR to notify the GitHub CodeQL docs team.
-name: Request docs review
-on:
-  # Runs in the context of the base repo.
-  # This gives the workflow write access to comment on PRs.
-  # The workflow should not check out or build the given ref,
-  # or use untrusted data from the event payload in a command line.
-  pull_request_target:
-    types: [labeled]
-
-jobs:
-  request-docs-review:
-    name: Request docs review
-    # Run only on labelled PRs to the main repository.
-    # Do not run on PRs to forks.
-    if:
-      github.event.label.name == 'ready-for-docs-review'
-      && github.event.pull_request.draft == false
-      && github.event.pull_request.base.repo.full_name == 'github/codeql'
-    runs-on: ubuntu-latest
-    steps:
-      - name: Comment to request docs review
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          PR_NUMBER: ${{ github.event.pull_request.number }}
-        run: |
-          gh pr comment "$PR_NUMBER" --repo "github/codeql" \
-            --body "Hello @github/docs-content-codeql - this PR is ready for docs review."

config/identical-files.json

@@ -5,6 +5,7 @@
     "java/ql/src/semmle/code/java/dataflow/internal/DataFlowImpl3.qll",
     "java/ql/src/semmle/code/java/dataflow/internal/DataFlowImpl4.qll",
     "java/ql/src/semmle/code/java/dataflow/internal/DataFlowImpl5.qll",
+    "java/ql/src/semmle/code/java/dataflow/internal/DataFlowImpl6.qll",
     "cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImpl.qll",
     "cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImpl2.qll",
     "cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImpl3.qll",
@@ -36,6 +37,7 @@
     "cpp/ql/src/semmle/code/cpp/dataflow/internal/tainttracking2/TaintTrackingImpl.qll",
     "cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/tainttracking1/TaintTrackingImpl.qll",
     "cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/tainttracking2/TaintTrackingImpl.qll",
+    "cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/tainttracking3/TaintTrackingImpl.qll",
     "csharp/ql/src/semmle/code/csharp/dataflow/internal/tainttracking1/TaintTrackingImpl.qll",
     "csharp/ql/src/semmle/code/csharp/dataflow/internal/tainttracking2/TaintTrackingImpl.qll",
     "csharp/ql/src/semmle/code/csharp/dataflow/internal/tainttracking3/TaintTrackingImpl.qll",
@@ -55,6 +57,10 @@
     "csharp/ql/src/semmle/code/csharp/dataflow/internal/DataFlowImplConsistency.qll",
     "python/ql/src/semmle/python/dataflow/new/internal/DataFlowImplConsistency.qll"
   ],
+  "DataFlow Java/C# Flow Summaries": [
+    "java/ql/src/semmle/code/java/dataflow/internal/FlowSummaryImpl.qll",
+    "csharp/ql/src/semmle/code/csharp/dataflow/internal/FlowSummaryImpl.qll"
+  ],
   "SsaReadPosition Java/C#": [
     "java/ql/src/semmle/code/java/dataflow/internal/rangeanalysis/SsaReadPositionCommon.qll",
     "csharp/ql/src/semmle/code/csharp/dataflow/internal/rangeanalysis/SsaReadPositionCommon.qll"
@@ -376,7 +382,6 @@
   ],
   "DuplicationProblems.inc.qhelp": [
     "cpp/ql/src/Metrics/Files/DuplicationProblems.inc.qhelp",
-    "csharp/ql/src/Metrics/Files/DuplicationProblems.inc.qhelp",
     "javascript/ql/src/Metrics/DuplicationProblems.inc.qhelp",
     "python/ql/src/Metrics/DuplicationProblems.inc.qhelp"
   ],
@@ -429,10 +434,11 @@
   "SSA C#": [
     "csharp/ql/src/semmle/code/csharp/dataflow/internal/SsaImplCommon.qll",
     "csharp/ql/src/semmle/code/csharp/controlflow/internal/pressa/SsaImplCommon.qll",
-    "csharp/ql/src/semmle/code/csharp/dataflow/internal/basessa/SsaImplCommon.qll"
+    "csharp/ql/src/semmle/code/csharp/dataflow/internal/basessa/SsaImplCommon.qll",
+    "csharp/ql/src/semmle/code/cil/internal/SsaImplCommon.qll"
   ],
   "CryptoAlgorithms Python/JS": [
     "javascript/ql/src/semmle/javascript/security/CryptoAlgorithms.qll",
     "python/ql/src/semmle/crypto/Crypto.qll"
   ]
-}
+}

cpp/autobuilder/Semmle.Autobuild.Cpp.Tests/BuildScripts.cs

@@ -5,6 +5,7 @@ using System;
 using System.Linq;
 using Microsoft.Build.Construction;
 using System.Xml;
+using System.IO;
 
 namespace Semmle.Autobuild.Cpp.Tests
 {
@@ -43,6 +44,8 @@ namespace Semmle.Autobuild.Cpp.Tests
         public IDictionary<string, int> RunProcess = new Dictionary<string, int>();
         public IDictionary<string, string> RunProcessOut = new Dictionary<string, string>();
         public IDictionary<string, string> RunProcessWorkingDirectory = new Dictionary<string, string>();
+        public HashSet<string> CreateDirectories { get; } = new HashSet<string>();
+        public HashSet<(string, string)> DownloadFiles { get; } = new HashSet<(string, string)>();
 
         int IBuildActions.RunProcess(string cmd, string args, string? workingDirectory, IDictionary<string, string>? env, out IList<string> stdOut)
         {
@@ -135,6 +138,14 @@ namespace Semmle.Autobuild.Cpp.Tests
 
         string IBuildActions.GetFullPath(string path) => path;
 
+        string? IBuildActions.GetFileName(string? path) => Path.GetFileName(path?.Replace('\\', '/'));
+
+        public string? GetDirectoryName(string? path)
+        {
+            var dir = Path.GetDirectoryName(path?.Replace('\\', '/'));
+            return dir is null ? path : path?.Substring(0, dir.Length);
+        }
+
         void IBuildActions.WriteAllText(string filename, string contents)
         {
         }
@@ -153,6 +164,18 @@ namespace Semmle.Autobuild.Cpp.Tests
                 s = s.Replace($"%{kvp.Key}%", kvp.Value);
             return s;
         }
+
+        public void CreateDirectory(string path)
+        {
+            if (!CreateDirectories.Contains(path))
+                throw new ArgumentException($"Missing CreateDirectory, {path}");
+        }
+
+        public void DownloadFile(string address, string fileName)
+        {
+            if (!DownloadFiles.Contains((address, fileName)))
+                throw new ArgumentException($"Missing DownloadFile, {address}, {fileName}");
+        }
     }
 
     /// <summary>
@@ -213,6 +236,7 @@ namespace Semmle.Autobuild.Cpp.Tests
             Actions.GetEnvironmentVariable[$"CODEQL_EXTRACTOR_{codeqlUpperLanguage}_SOURCE_ARCHIVE_DIR"] = "";
             Actions.GetEnvironmentVariable[$"CODEQL_EXTRACTOR_{codeqlUpperLanguage}_ROOT"] = $@"C:\codeql\{codeqlUpperLanguage.ToLowerInvariant()}";
             Actions.GetEnvironmentVariable["CODEQL_JAVA_HOME"] = @"C:\codeql\tools\java";
+            Actions.GetEnvironmentVariable["CODEQL_PLATFORM"] = "win64";
             Actions.GetEnvironmentVariable["SEMMLE_DIST"] = @"C:\odasa";
             Actions.GetEnvironmentVariable["SEMMLE_JAVA_HOME"] = @"C:\odasa\tools\java";
             Actions.GetEnvironmentVariable["SEMMLE_PLATFORM_TOOLS"] = @"C:\odasa\tools";
@@ -273,7 +297,8 @@ namespace Semmle.Autobuild.Cpp.Tests
         [Fact]
         public void TestCppAutobuilderSuccess()
         {
-            Actions.RunProcess[@"cmd.exe /C C:\odasa\tools\csharp\nuget\nuget.exe restore C:\Project\test.sln"] = 1;
+            Actions.RunProcess[@"cmd.exe /C nuget restore C:\Project\test.sln -DisableParallelProcessing"] = 1;
+            Actions.RunProcess[@"cmd.exe /C C:\Project\.nuget\nuget.exe restore C:\Project\test.sln -DisableParallelProcessing"] = 0;
             Actions.RunProcess[@"cmd.exe /C CALL ^""C:\Program Files ^(x86^)\Microsoft Visual Studio 14.0\VC\vcvarsall.bat^"" && set Platform=&& type NUL && C:\odasa\tools\odasa index --auto msbuild C:\Project\test.sln /p:UseSharedCompilation=false /t:rebuild /p:Platform=""x86"" /p:Configuration=""Release"" /p:MvcBuildViews=true"] = 0;
             Actions.RunProcessOut[@"C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe -prerelease -legacy -property installationPath"] = "";
             Actions.RunProcess[@"C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe -prerelease -legacy -property installationPath"] = 1;
@@ -286,11 +311,13 @@ namespace Semmle.Autobuild.Cpp.Tests
             Actions.FileExists[@"C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe"] = true;
             Actions.EnumerateFiles[@"C:\Project"] = "foo.cs\ntest.slx";
             Actions.EnumerateDirectories[@"C:\Project"] = "";
+            Actions.CreateDirectories.Add(@"C:\Project\.nuget");
+            Actions.DownloadFiles.Add(("https://dist.nuget.org/win-x86-commandline/latest/nuget.exe", @"C:\Project\.nuget\nuget.exe"));
 
             var autobuilder = CreateAutoBuilder(true);
             var solution = new TestSolution(@"C:\Project\test.sln");
             autobuilder.ProjectsOrSolutionsToBuild.Add(solution);
-            TestAutobuilderScript(autobuilder, 0, 2);
+            TestAutobuilderScript(autobuilder, 0, 3);
         }
     }
 }

cpp/change-notes/2021-04-06-assign-where-compare-meant.md

@@ -0,0 +1,2 @@
+lgtm,codescanning
+* The 'Assignment where comparison was intended' (cpp/assign-where-compare-meant) query has been improved to flag fewer benign assignments in conditionals.

cpp/change-notes/2021-04-13-arithmetic-queries.md

@@ -0,0 +1,2 @@
+lgtm
+* The queries cpp/tainted-arithmetic, cpp/uncontrolled-arithmetic, and cpp/arithmetic-with-extreme-values have been improved to produce fewer false positives.

cpp/change-notes/2021-04-21-return-stack-allocated-object.md

@@ -0,0 +1,2 @@
+codescanning
+* The 'Pointer to stack object used as return value' (cpp/return-stack-allocated-object) query has been deprecated, and any uses should be replaced with `Returning stack-allocated memory` (cpp/return-stack-allocated-memory).

cpp/change-notes/2021-26-04-more-sound-expr-might-overflow.md

@@ -0,0 +1,2 @@
+lgtm,codescanning
+* The `exprMightOverflowPositively` and `exprMightOverflowNegatively` predicates from the `SimpleRangeAnalysis` library now recognize more expressions that might overflow.

cpp/ql/src/Best Practices/Magic Constants/MagicConstantsNumbers.qhelp

@@ -39,7 +39,7 @@ then replace all the relevant occurrences in the code.</p>
 </li>
 <li>
   Mats Henricson and Erik Nyquist, <i>Industrial Strength C++</i>, published by Prentice Hall PTR (1997). 
-  Chapter 5: Object Life Cycle, Rec 5.4 (<a href="http://mongers.org/industrial-c++/">PDF</a>).
+  Chapter 5: Object Life Cycle, Rec 5.4 (<a href="https://web.archive.org/web/20190919025638/https://mongers.org/industrial-c++/">PDF</a>).
 </li>
 <li>
   <a href="https://www.securecoding.cert.org/confluence/display/c/DCL06-C.+Use+meaningful+symbolic+constants+to+represent+literal+values">DCL06-C. Use meaningful symbolic constants to represent literal values</a>

cpp/ql/src/Best Practices/Magic Constants/MagicConstantsString.qhelp

@@ -38,7 +38,7 @@ constant.</p>
 </li>
 <li>
   Mats Henricson and Erik Nyquist, <i>Industrial Strength C++</i>, published by Prentice Hall PTR (1997). 
-  Chapter 5: Object Life Cycle, Rec 5.4 (<a href="http://mongers.org/industrial-c++/">PDF</a>).
+  Chapter 5: Object Life Cycle, Rec 5.4 (<a href="https://web.archive.org/web/20190919025638/https://mongers.org/industrial-c++/">PDF</a>).
 </li>
 <li>
   <a href="https://www.securecoding.cert.org/confluence/display/c/DCL06-C.+Use+meaningful+symbolic+constants+to+represent+literal+values">DCL06-C. Use meaningful symbolic constants to represent literal values</a>

cpp/ql/src/Best Practices/SloppyGlobal.qhelp

@@ -21,7 +21,7 @@ Review the purpose of the each global variable flagged by this rule and update e
 
 <li>
   Mats Henricson and Erik Nyquist, <i>Industrial Strength C++</i>, published by Prentice Hall PTR (1997). 
-  Chapter 1: Naming, Rec 1.1 (<a href="http://mongers.org/industrial-c++/">PDF</a>).
+  Chapter 1: Naming, Rec 1.1 (<a href="https://web.archive.org/web/20190919025638/https://mongers.org/industrial-c++/">PDF</a>).
 </li>
 <li>
   <a href="http://www.learncpp.com/cpp-tutorial/42-global-variables/">Global variables</a>.

cpp/ql/src/Best Practices/UseOfGoto.qhelp

@@ -45,7 +45,7 @@ this rule.
 </li>
 <li>
   Mats Henricson and Erik Nyquist, <i>Industrial Strength C++</i>, Rule 4.6. Prentice Hall PTR, 1997.
-  (<a href="http://mongers.org/industrial-c++/">PDF</a>).
+  (<a href="https://web.archive.org/web/20190919025638/https://mongers.org/industrial-c++/">PDF</a>).
 </li>
 <li>
   cplusplus.com: <a href="http://www.cplusplus.com/doc/tutorial/control/">Control Structures</a>.

cpp/ql/src/Critical/ReturnStackAllocatedObject.ql

@@ -7,6 +7,8 @@
  * @tags reliability
  *       security
  *       external/cwe/cwe-562
+ * @deprecated This query is not suitable for production use and has been deprecated. Use
+ *             cpp/return-stack-allocated-memory instead.
  */
 
 import semmle.code.cpp.pointsto.PointsTo

cpp/ql/src/Critical/ReturnValueIgnored.qhelp

@@ -32,7 +32,7 @@ Check the return value of functions that return status information.
 <references>
 
 <li>
-  M. Henricson and E. Nyquist, <i>Industrial Strength C++</i>, Chapter 12: Error handling. Prentice Hall PTR, 1997 (<a href="http://mongers.org/industrial-c++/">available online</a>).
+  M. Henricson and E. Nyquist, <i>Industrial Strength C++</i>, Chapter 12: Error handling. Prentice Hall PTR, 1997 (<a href="https://web.archive.org/web/20190919025638/https://mongers.org/industrial-c++/">available online</a>).
 </li>
 <li>
   The CERT C Secure Coding Standard: <a href="https://www.securecoding.cert.org/confluence/display/perl/EXP32-PL.+Do+not+ignore+function+return+values">EXP32-PL. Do not ignore function return values</a>.

cpp/ql/src/Likely Bugs/Arithmetic/SignedOverflowCheck.ql

@@ -9,6 +9,8 @@
  * @id cpp/signed-overflow-check
  * @tags correctness
  *       security
+ *       external/cwe/cwe-128
+ *       external/cwe/cwe-190
  */
 
 import cpp

cpp/ql/src/Likely Bugs/Conversion/CastArrayPointerArithmetic.ql

@@ -7,12 +7,12 @@
  * @kind path-problem
  * @problem.severity warning
  * @precision high
+ * @id cpp/upcast-array-pointer-arithmetic
  * @tags correctness
  *       reliability
  *       security
  *       external/cwe/cwe-119
  *       external/cwe/cwe-843
- * @id cpp/upcast-array-pointer-arithmetic
  */
 
 import cpp

cpp/ql/src/Likely Bugs/Format/SnprintfOverflow.ql

@@ -8,6 +8,8 @@
  * @tags reliability
  *       correctness
  *       security
+ *       external/cwe/cwe-190
+ *       external/cwe/cwe-253
  */
 
 import cpp

cpp/ql/src/Likely Bugs/Format/WrongNumberOfFormatArguments.ql

@@ -9,6 +9,7 @@
  * @tags reliability
  *       correctness
  *       security
+ *       external/cwe/cwe-234
  *       external/cwe/cwe-685
  */
 

cpp/ql/src/Likely Bugs/Likely Typos/AssignWhereCompareMeant.ql

@@ -54,7 +54,7 @@ class BooleanControllingAssignmentInExpr extends BooleanControllingAssignment {
   override predicate isWhitelisted() {
     this.getConversion().(ParenthesisExpr).isParenthesised()
     or
-    // whitelist this assignment if all comparison operations in the expression that this
+    // Allow this assignment if all comparison operations in the expression that this
     // assignment is part of, are not parenthesized. In that case it seems like programmer
     // is fine with unparenthesized comparison operands to binary logical operators, and
     // the parenthesis around this assignment was used to call it out as an assignment.
@@ -62,6 +62,21 @@ class BooleanControllingAssignmentInExpr extends BooleanControllingAssignment {
     forex(ComparisonOperation op | op = getComparisonOperand*(this.getParent+()) |
       not op.isParenthesised()
     )
+    or
+    // Match a pattern like:
+    // ```
+    // if((a = b) && use_value(a)) { ... }
+    // ```
+    // where the assignment is meant to update the value of `a` before it's used in some other boolean
+    // subexpression that is guarenteed to be evaluate _after_ the assignment.
+    this.isParenthesised() and
+    exists(LogicalAndExpr parent, Variable var, VariableAccess access |
+      var = this.getLValue().(VariableAccess).getTarget() and
+      access = var.getAnAccess() and
+      not access.isUsedAsLValue() and
+      parent.getRightOperand() = access.getParent*() and
+      parent.getLeftOperand() = this.getParent*()
+    )
   }
 }
 

cpp/ql/src/Likely Bugs/Likely Typos/MissingEnumCaseInSwitch.qhelp

@@ -26,7 +26,7 @@ indication that there may be cases unhandled by the <code>switch</code> statemen
   MSDN Library: <a href="https://docs.microsoft.com/en-us/cpp/cpp/switch-statement-cpp">switch statement (C++)</a>
 </li>
 <li>
-  M. Henricson and E. Nyquist, <i>Industrial Strength C++</i>, Chapter 4: Control Flow, Rec 4.5. Prentice Hall PTR, 1997 (<a href="http://mongers.org/industrial-c++/">available online</a>).
+  M. Henricson and E. Nyquist, <i>Industrial Strength C++</i>, Chapter 4: Control Flow, Rec 4.5. Prentice Hall PTR, 1997 (<a href="https://web.archive.org/web/20190919025638/https://mongers.org/industrial-c++/">available online</a>).
 </li>
 
 

cpp/ql/src/Likely Bugs/Memory Management/PointerOverflow.ql

@@ -8,6 +8,7 @@
  * @id cpp/pointer-overflow-check
  * @tags reliability
  *       security
+ *       external/cwe/cwe-758
  */
 
 import cpp

cpp/ql/src/Likely Bugs/Memory Management/ReturnStackAllocatedMemory.ql

@@ -13,6 +13,7 @@
 
 import cpp
 import semmle.code.cpp.dataflow.EscapesTree
+import semmle.code.cpp.models.interfaces.PointerWrapper
 import semmle.code.cpp.dataflow.DataFlow
 
 /**
@@ -39,6 +40,10 @@ predicate hasNontrivialConversion(Expr e) {
     e instanceof ParenthesisExpr
   )
   or
+  // A smart pointer can be stack-allocated while the data it points to is heap-allocated.
+  // So we exclude such "conversions" from this predicate.
+  e = any(PointerWrapper wrapper).getAnUnwrapperFunction().getACallToThisFunction()
+  or
   hasNontrivialConversion(e.getConversion())
 }
 

cpp/ql/src/Likely Bugs/OO/UnsafeUseOfThis.ql

@@ -10,6 +10,7 @@
  * @tags correctness
  *       language-features
  *       security
+ *       external/cwe/cwe-670
  */
 
 import cpp

cpp/ql/src/Likely Bugs/Underspecified Functions/TooFewArguments.ql

@@ -12,6 +12,8 @@
  * @tags correctness
  *       maintainability
  *       security
+ *       external/cwe/cwe-234
+ *       external/cwe/cwe-685
  */
 
 import cpp

cpp/ql/src/Metrics/Dependencies/ExternalDependencies.ql

@@ -5,7 +5,6 @@
  * @kind treemap
  * @treemap.warnOn highValues
  * @metricType externalDependency
- * @precision medium
  * @id cpp/external-dependencies
  * @tags modularity
  */

cpp/ql/src/Metrics/Files/FLinesOfCode.ql

@@ -7,7 +7,6 @@
  * @treemap.warnOn highValues
  * @metricType file
  * @metricAggregate avg sum max
- * @precision very-high
  * @id cpp/lines-of-code-in-files
  * @tags maintainability
  *       complexity

cpp/ql/src/Metrics/Files/FLinesOfCommentedOutCode.ql

@@ -5,7 +5,6 @@
  * @treemap.warnOn highValues
  * @metricType file
  * @metricAggregate avg sum max
- * @precision high
  * @id cpp/lines-of-commented-out-code-in-files
  * @tags documentation
  */

cpp/ql/src/Metrics/Files/FLinesOfComments.ql

@@ -7,7 +7,6 @@
  * @treemap.warnOn lowValues
  * @metricType file
  * @metricAggregate avg sum max
- * @precision very-high
  * @id cpp/lines-of-comments-in-files
  * @tags maintainability
  *       documentation

cpp/ql/src/Metrics/Files/FLinesOfDuplicatedCode.ql

@@ -8,7 +8,6 @@
  * @treemap.warnOn highValues
  * @metricType file
  * @metricAggregate avg sum max
- * @precision high
  * @id cpp/duplicated-lines-in-files
  * @tags testability
  *       modularity

cpp/ql/src/Metrics/Files/FNumberOfTests.ql

@@ -5,7 +5,6 @@
  * @treemap.warnOn lowValues
  * @metricType file
  * @metricAggregate avg sum max
- * @precision medium
  * @id cpp/tests-in-files
  * @tags maintainability
  */

cpp/ql/src/Metrics/Files/FTransitiveIncludes.qhelp

@@ -29,7 +29,7 @@ build time: the more included files, the longer the compilation time.</p>
   <a href="http://www.drdobbs.com/cpp/decoupling-c-header-files/212701130">Decoupling C Header Files</a>
 </li>
 <li>
-  <a href="https://wiki.hsr.ch/Prog3/files/overload72-FINAL_DesigningHeaderFiles.pdf">C++ Best Practice -
+  <a href="https://accu.org/journals/overload/14/72/griffiths_1995/">C++ Best Practice -
 Designing Header Files</a>
 </li>
 

cpp/ql/src/Metrics/Files/FTransitiveSourceIncludes.qhelp

@@ -35,7 +35,7 @@ they are contributing to unnecessarily long build times and creating artificial
   <a href="http://www.drdobbs.com/cpp/decoupling-c-header-files/212701130">Decoupling C Header Files</a>
 </li>
 <li>
-  <a href="https://wiki.hsr.ch/Prog3/files/overload72-FINAL_DesigningHeaderFiles.pdf">C++ Best Practice -
+  <a href="https://accu.org/journals/overload/14/72/griffiths_1995/">C++ Best Practice -
 Designing Header Files</a>
 </li>
 </references>

cpp/ql/src/Security/CWE/CWE-190/IntegerOverflowTainted.ql

@@ -28,6 +28,7 @@ predicate outOfBoundsExpr(Expr expr, string kind) {
 
 from Expr use, Expr origin, string kind
 where
+  not use.getUnspecifiedType() instanceof PointerType and
   outOfBoundsExpr(use, kind) and
   tainted(origin, use) and
   origin != use and

cpp/ql/src/Security/CWE/CWE-428/UnsafeCreateProcessCall.ql

@@ -93,7 +93,7 @@ class QuotedCommandInCreateProcessFunctionConfiguration extends DataFlow2::Confi
 
 bindingset[s]
 predicate isQuotedOrNoSpaceApplicationNameOnCmd(string s) {
-  s.regexpMatch("\"([^\"])*\"(\\s|.)*") // The first element (path) is quoted
+  s.regexpMatch("\"([^\"])*\"[\\s\\S]*") // The first element (path) is quoted
   or
   s.regexpMatch("[^\\s]+") // There are no spaces in the string
 }

cpp/ql/src/experimental/Security/CWE/CWE-570/WrongInDetectingAndHandlingMemoryAllocationErrors.cpp

@@ -1,35 +1,43 @@
-// BAD: on memory allocation error, the program terminates.
-void badFunction(const int *source, std::size_t length) noexcept {
-  int * dest = new int[length];
+// BAD: the allocation will throw an unhandled exception
+// instead of returning a null pointer.
+void bad1(std::size_t length) noexcept {
+  int* dest = new int[length];
+  if(!dest) {
+    return;
+  }
   std::memset(dest, 0, length);
-// ..
+  // ...
 }
-// GOOD: memory allocation error will be handled.
-void goodFunction(const int *source, std::size_t length) noexcept {
+
+// BAD: the allocation won't throw an exception, but
+// instead return a null pointer.
+void bad2(std::size_t length) noexcept {
   try {
-       int * dest = new int[length];
-  } catch(std::bad_alloc) {
+    int* dest = new(std::nothrow) int[length];
+    std::memset(dest, 0, length);
+    // ...
+  } catch(std::bad_alloc&) {
     // ...
   }
-  std::memset(dest, 0, length);
-// ..
 }
-// BAD: memory allocation error will not be handled.
-void badFunction(const int *source, std::size_t length) noexcept {
+
+// GOOD: the allocation failure is handled appropiately.
+void good1(std::size_t length) noexcept {
   try {
-       int * dest = new (std::nothrow) int[length];
-  } catch(std::bad_alloc) {
+    int* dest = new int[length];
+    std::memset(dest, 0, length);
+    // ...
+  } catch(std::bad_alloc&) {
     // ...
   }
-  std::memset(dest, 0, length);
-// ..
 }
-// GOOD: memory allocation error will be handled.
-void goodFunction(const int *source, std::size_t length) noexcept {
-  int * dest = new (std::nothrow) int[length];
-  if (!dest) {
-      return;
+
+// GOOD: the allocation failure is handled appropiately.
+void good2(std::size_t length) noexcept {
+  int* dest = new int[length];
+  if(!dest) {
+    return;
   }
   std::memset(dest, 0, length);
-// ..
+  // ...
 }

cpp/ql/src/experimental/Security/CWE/CWE-570/WrongInDetectingAndHandlingMemoryAllocationErrors.qhelp

@@ -3,16 +3,20 @@
   "qhelp.dtd">
 <qhelp>
 <overview>
-<p>When using the <code>new</code> operator to allocate memory, you need to pay attention to the different ways of detecting errors. <code>::operator new(std::size_t)</code> throws an exception on error, whereas <code>::operator new(std::size_t, const std::nothrow_t &amp;)</code> returns zero on error. The programmer can get confused and check the error that occurs when allocating memory incorrectly. That can lead to an unhandled program termination or to a violation of the program logic.</p>
+<p>Different overloads of the <code>new</code> operator handle allocation failures in different ways.
+If <code>new T</code> fails for some type <code>T</code>, it throws a <code>std::bad_alloc</code> exception,
+but <code>new(std::nothrow) T</code> returns a null pointer. If the programmer does not use the corresponding
+method of error handling, allocation failure may go unhandled and could cause the program to behave in
+unexpected ways.</p>
 
 </overview>
 <recommendation>
 
-<p>Use the correct error detection method corresponding with the memory allocation.</p>
+<p>Make sure that exceptions are handled appropriately if <code>new T</code> is used. On the other hand,
+make sure to handle the possibility of null pointers if <code>new(std::nothrow) T</code> is used.</p>
 
 </recommendation>
 <example>
-<p>The following example demonstrates various approaches to detecting memory allocation errors using the <code>new</code> operator.</p>
 <sample src="WrongInDetectingAndHandlingMemoryAllocationErrors.cpp" />
 
 </example>

cpp/ql/src/experimental/Security/CWE/CWE-570/WrongInDetectingAndHandlingMemoryAllocationErrors.ql

@@ -1,7 +1,6 @@
 /**
  * @name Detect And Handle Memory Allocation Errors
- * @description --::operator new(std::size_t) throws an exception on error, and ::operator new(std::size_t, const std::nothrow_t &) returns zero on error.
- *              --the programmer can get confused when check the error that occurs when allocating memory incorrectly.
+ * @description `operator new` throws an exception on allocation failures, while `operator new(std::nothrow)` returns a null pointer. Mixing up these two failure conditions can result in unexpected behavior.
  * @kind problem
  * @id cpp/detect-and-handle-memory-allocation-errors
  * @problem.severity warning
@@ -54,27 +53,34 @@ class WrongCheckErrorOperatorNew extends FunctionCall {
    * Holds if results call `operator new` check in `operator if`.
    */
   predicate isExistsIfCondition() {
-    exists(IfCompareWithZero ifc, AssignExpr aex, Initializer it |
+    exists(IfCompareWithZero ifc |
       // call `operator new` directly from the condition of `operator if`.
       this = ifc.getCondition().getAChild*()
       or
-      // check results call `operator new` with variable appropriation
       postDominates(ifc, this) and
-      aex.getAChild() = exp and
-      ifc.getCondition().getAChild().(VariableAccess).getTarget() =
-        aex.getLValue().(VariableAccess).getTarget()
-      or
-      // check results call `operator new` with declaration variable
-      postDominates(ifc, this) and
-      exp = it.getExpr() and
-      it.getDeclaration() = ifc.getCondition().getAChild().(VariableAccess).getTarget()
+      exists(Variable v |
+        v = ifc.getCondition().getAChild().(VariableAccess).getTarget() and
+        (
+          exists(AssignExpr aex |
+            // check results call `operator new` with variable appropriation
+            aex.getAChild() = exp and
+            v = aex.getLValue().(VariableAccess).getTarget()
+          )
+          or
+          exists(Initializer it |
+            // check results call `operator new` with declaration variable
+            exp = it.getExpr() and
+            it.getDeclaration() = v
+          )
+        )
+      )
     )
   }
 
   /**
-   * Holds if `(std::nothrow)` exists in call `operator new`.
+   * Holds if `(std::nothrow)` or `(std::noexcept)` exists in call `operator new`.
    */
-  predicate isExistsNothrow() { this.getAChild().toString() = "nothrow" }
+  predicate isExistsNothrow() { getTarget().isNoExcept() or getTarget().isNoThrow() }
 }
 
 from WrongCheckErrorOperatorNew op

cpp/ql/src/experimental/Security/CWE/CWE-691/InsufficientControlFlowManagementAfterRefactoringTheCode.c

@@ -0,0 +1,17 @@
+while(flagsLoop)
+{
+  ...
+  if(flagsIf) break;
+  ...
+}while(flagsLoop); // BAD: when exiting through `break`, it is possible to get into an eternal loop.
+...
+while(flagsLoop)
+{
+  ...
+  if(flagsIf) break;
+  ...
+} // GOOD: correct cycle
+...
+if(intA+intB) return 1; // BAD: possibly no comparison
+...
+if(intA+intB>intC) return 1; // GOOD: correct comparison

cpp/ql/src/experimental/Security/CWE/CWE-691/InsufficientControlFlowManagementAfterRefactoringTheCode.qhelp

@@ -0,0 +1,28 @@
+<!DOCTYPE qhelp PUBLIC
+  "-//Semmle//qhelp//EN"
+  "qhelp.dtd">
+<qhelp>
+<overview>
+<p>In some situations, after code refactoring, parts of the old constructs may remain. They are correctly accepted by the compiler, but can critically affect program execution. For example, if you switch from `do {...} while ();` to `while () {...}` forgetting to remove the old construct completely, you get `while(){...}while();` which may be vulnerable. These code snippets look suspicious and require the developer's attention.</p>
+
+
+</overview>
+<recommendation>
+
+<p>We recommend that you use more explicit code transformations.</p>
+
+</recommendation>
+<example>
+<p>The following example demonstrates the erroneous and corrected sections of the code.</p>
+<sample src="InsufficientControlFlowManagementAfterRefactoringTheCode.c" />
+
+</example>
+<references>
+
+<li>
+  CWE Common Weakness Enumeration:
+  <a href="https://cwe.mitre.org/data/definitions/691.html"> CWE-691: Insufficient Control Flow Management</a>.
+</li>
+
+</references>
+</qhelp>

cpp/ql/src/experimental/Security/CWE/CWE-691/InsufficientControlFlowManagementAfterRefactoringTheCode.ql

@@ -0,0 +1,119 @@
+/**
+ * @name Errors After Refactoring
+ * @description --In some situations, after code refactoring, parts of the old constructs may remain.
+ *              --They are correctly accepted by the compiler, but can critically affect program execution.
+ *              --For example, if you switch from `do {...} while ();` to `while () {...}` with errors, you run the risk of running out of resources.
+ *              --These code snippets look suspicious and require the developer's attention.
+ * @kind problem
+ * @id cpp/errors-after-refactoring
+ * @problem.severity warning
+ * @precision medium
+ * @tags correctness
+ *       security
+ *       external/cwe/cwe-691
+ */
+
+import cpp
+import semmle.code.cpp.valuenumbering.HashCons
+import semmle.code.cpp.valuenumbering.GlobalValueNumbering
+
+/**
+ * Using `while` directly after the body of another` while`.
+ */
+class UsingWhileAfterWhile extends WhileStmt {
+  /**
+   * Using a loop call after another loop has finished running can result in an eternal loop.
+   * For example, perhaps as a result of refactoring, the `do ... while ()` loop was incorrectly corrected.
+   * Even in the case of deliberate use of such an expression, it is better to correct it.
+   */
+  UsingWhileAfterWhile() {
+    exists(WhileStmt wh1 |
+      wh1.getStmt().getAChild*().(BreakStmt).(ControlFlowNode).getASuccessor().getASuccessor() =
+        this and
+      hashCons(wh1.getCondition()) = hashCons(this.getCondition()) and
+      this.getStmt() instanceof EmptyStmt
+    )
+    or
+    exists(ForStmt fr1 |
+      fr1.getStmt().getAChild*().(BreakStmt).(ControlFlowNode).getASuccessor().getASuccessor() =
+        this and
+      hashCons(fr1.getCondition()) = hashCons(this.getCondition()) and
+      this.getStmt() instanceof EmptyStmt
+    )
+  }
+}
+
+/**
+ * Using arithmetic in a condition.
+ */
+class UsingArithmeticInComparison extends BinaryArithmeticOperation {
+  /**
+   * Using arithmetic operations in a comparison operation can be dangerous.
+   * For example, part of the comparison may have been lost as a result of refactoring.
+   * Even if you deliberately use such an expression, it is better to add an explicit comparison.
+   */
+  UsingArithmeticInComparison() {
+    this.getParent*() instanceof IfStmt and
+    not this.getAChild*().isConstant() and
+    not this.getParent*() instanceof Call and
+    not this.getParent*() instanceof AssignExpr and
+    not this.getParent*() instanceof ArrayExpr and
+    not this.getParent*() instanceof RemExpr and
+    not this.getParent*() instanceof AssignBitwiseOperation and
+    not this.getParent*() instanceof AssignArithmeticOperation and
+    not this.getParent*() instanceof EqualityOperation and
+    not this.getParent*() instanceof RelationalOperation
+  }
+
+  /** Holds when the expression is inside the loop body. */
+  predicate insideTheLoop() { exists(Loop lp | lp.getStmt().getAChild*() = this.getParent*()) }
+
+  /** Holds when the expression is used in binary operations. */
+  predicate workingWithValue() {
+    this.getParent*() instanceof BinaryBitwiseOperation or
+    this.getParent*() instanceof NotExpr
+  }
+
+  /** Holds when the expression contains a pointer. */
+  predicate workingWithPointer() {
+    this.getAChild*().getFullyConverted().getType() instanceof DerivedType
+  }
+
+  /** Holds when a null comparison expression exists. */
+  predicate compareWithZero() {
+    exists(Expr exp |
+      exp instanceof ComparisonOperation and
+      (
+        globalValueNumber(exp.getAChild*()) = globalValueNumber(this) or
+        hashCons(exp.getAChild*()) = hashCons(this)
+      ) and
+      (
+        exp.(ComparisonOperation).getLeftOperand().getValue() = "0" or
+        exp.(ComparisonOperation).getRightOperand().getValue() = "0"
+      )
+    )
+  }
+
+  /** Holds when a comparison expression exists. */
+  predicate compareWithOutZero() {
+    exists(Expr exp |
+      exp instanceof ComparisonOperation and
+      (
+        globalValueNumber(exp.getAChild*()) = globalValueNumber(this) or
+        hashCons(exp.getAChild*()) = hashCons(this)
+      )
+    )
+  }
+}
+
+from Expr exp
+where
+  exp instanceof UsingArithmeticInComparison and
+  not exp.(UsingArithmeticInComparison).workingWithValue() and
+  not exp.(UsingArithmeticInComparison).workingWithPointer() and
+  not exp.(UsingArithmeticInComparison).insideTheLoop() and
+  not exp.(UsingArithmeticInComparison).compareWithZero() and
+  exp.(UsingArithmeticInComparison).compareWithOutZero()
+  or
+  exists(WhileStmt wst | wst instanceof UsingWhileAfterWhile and exp = wst.getCondition())
+select exp, "this expression needs your attention"

cpp/ql/src/experimental/Security/CWE/CWE-691/InsufficientControlFlowManagementWhenUsingBitOperations.c

@@ -0,0 +1,4 @@
+if(len>0 & memset(buf,0,len)) return 1; // BAD: `memset` will be called regardless of the value of the `len` variable. moreover, one cannot be sure that it will happen after verification
+...
+if(len>0 && memset(buf,0,len)) return 1; // GOOD: `memset` will be called after the `len` variable has been checked.
+...

cpp/ql/src/experimental/Security/CWE/CWE-691/InsufficientControlFlowManagementWhenUsingBitOperations.qhelp

@@ -0,0 +1,28 @@
+<!DOCTYPE qhelp PUBLIC
+  "-//Semmle//qhelp//EN"
+  "qhelp.dtd">
+<qhelp>
+<overview>
+<p>Using bitwise operations can be a mistake in some situations. For example, if parameters are evaluated in an expression and the function should be called only upon certain test results. These bitwise operations look suspicious and require developer attention.</p>
+
+
+</overview>
+<recommendation>
+
+<p>We recommend that you evaluate the correctness of using the specified bit operations.</p>
+
+</recommendation>
+<example>
+<p>The following example demonstrates the erroneous and fixed use of bit and logical operations.</p>
+<sample src="InsufficientControlFlowManagementWhenUsingBitOperations.c" />
+
+</example>
+<references>
+
+<li>
+  CWE Common Weakness Enumeration:
+  <a href="https://cwe.mitre.org/data/definitions/691.html"> CWE-691: Insufficient Control Flow Management</a>.
+</li>
+
+</references>
+</qhelp>

cpp/ql/src/experimental/Security/CWE/CWE-691/InsufficientControlFlowManagementWhenUsingBitOperations.ql

@@ -0,0 +1,78 @@
+/**
+ * @name Errors When Using Bit Operations
+ * @description Unlike the binary operations `||` and `&&`, there is no sequence point after evaluating an
+ *              operand of a bitwise operation like `|` or `&`. If left-to-right evaluation is expected this may be confusing.
+ * @kind problem
+ * @id cpp/errors-when-using-bit-operations
+ * @problem.severity warning
+ * @precision medium
+ * @tags correctness
+ *       security
+ *       external/cwe/cwe-691
+ */
+
+import cpp
+import semmle.code.cpp.valuenumbering.GlobalValueNumbering
+
+/**
+ * Dangerous uses of bit operations.
+ * For example: `if(intA>0 & intA<10 & charBuf&myFunc(charBuf[intA]))`.
+ * In this case, the function will be called in any case, and even the sequence of the call is not guaranteed.
+ */
+class DangerousBitOperations extends BinaryBitwiseOperation {
+  FunctionCall bfc;
+
+  /**
+   * The assignment indicates the conscious use of the bit operator.
+   * Use in comparison, conversion, or return value indicates conscious use of the bit operator.
+   * The use of shifts and bitwise operations on any element of an expression indicates a conscious use of the bitwise operator.
+   */
+  DangerousBitOperations() {
+    bfc = this.getRightOperand() and
+    not this.getParent*() instanceof Assignment and
+    not this.getParent*() instanceof Initializer and
+    not this.getParent*() instanceof ReturnStmt and
+    not this.getParent*() instanceof EqualityOperation and
+    not this.getParent*() instanceof UnaryLogicalOperation and
+    not this.getParent*() instanceof BinaryLogicalOperation and
+    not this.getAChild*() instanceof BitwiseXorExpr and
+    not this.getAChild*() instanceof LShiftExpr and
+    not this.getAChild*() instanceof RShiftExpr
+  }
+
+  /** Holds when part of a bit expression is used in a logical operation. */
+  predicate useInLogicalOperations() {
+    exists(BinaryLogicalOperation blop, Expr exp |
+      blop.getAChild*() = exp and
+      exp.(FunctionCall).getTarget() = bfc.getTarget() and
+      not exp.getParent() instanceof ComparisonOperation and
+      not exp.getParent() instanceof BinaryBitwiseOperation
+    )
+  }
+
+  /** Holds when part of a bit expression is used as part of another supply. For example, as an argument to another function. */
+  predicate useInOtherCalls() {
+    bfc.hasQualifier() or
+    bfc.getTarget() instanceof Operator or
+    exists(FunctionCall fc | fc.getAnArgument().getAChild*() = this) or
+    bfc.getTarget() instanceof BuiltInFunction
+  }
+
+  /** Holds when the bit expression contains both arguments and a function call. */
+  predicate dangerousArgumentChecking() {
+    not this.getLeftOperand() instanceof Call and
+    globalValueNumber(this.getLeftOperand().getAChild*()) = globalValueNumber(bfc.getAnArgument())
+  }
+
+  /** Holds when function calls are present in the bit expression. */
+  predicate functionCallsInBitsExpression() {
+    this.getLeftOperand().getAChild*() instanceof FunctionCall
+  }
+}
+
+from DangerousBitOperations dbo
+where
+  not dbo.useInOtherCalls() and
+  dbo.useInLogicalOperations() and
+  (not dbo.functionCallsInBitsExpression() or dbo.dangerousArgumentChecking())
+select dbo, "This bitwise operation appears in a context where a Boolean operation is expected."

cpp/ql/src/experimental/Security/CWE/CWE-783/OperatorPrecedenceLogicErrorWhenUseBoolType.c

@@ -0,0 +1,11 @@
+if(len=funcReadData()==0) return 1; // BAD: variable `len` will not equal the value returned by function `funcReadData()`
+...
+if((len=funcReadData())==0) return 1; // GOOD: variable `len` equal the value returned by function `funcReadData()`
+...
+bool a=true;
+a++;// BAD: variable `a` does not change its meaning
+bool b;
+b=-a;// BAD: variable `b` equal `true`
+...
+a=false;// GOOD: variable `a` equal `false`
+b=!a;// GOOD: variable `b` equal `false`

cpp/ql/src/experimental/Security/CWE/CWE-783/OperatorPrecedenceLogicErrorWhenUseBoolType.qhelp

@@ -0,0 +1,28 @@
+<!DOCTYPE qhelp PUBLIC
+  "-//Semmle//qhelp//EN"
+  "qhelp.dtd">
+<qhelp>
+<overview>
+<p>Finding places of confusing use of boolean type. For example, a unary minus does not work before a boolean type and an increment always gives true.</p>
+
+
+</overview>
+<recommendation>
+
+<p>we recommend making the code simpler.</p>
+
+</recommendation>
+<example>
+<p>The following example demonstrates erroneous and fixed methods for using a boolean data type.</p>
+<sample src="OperatorPrecedenceLogicErrorWhenUseBoolType.c" />
+
+</example>
+<references>
+
+<li>
+  CERT C Coding Standard:
+  <a href="https://wiki.sei.cmu.edu/confluence/display/c/EXP00-C.+Use+parentheses+for+precedence+of+operation">EXP00-C. Use parentheses for precedence of operation</a>.
+</li>
+
+</references>
+</qhelp>

cpp/ql/src/experimental/Security/CWE/CWE-783/OperatorPrecedenceLogicErrorWhenUseBoolType.ql

@@ -0,0 +1,54 @@
+/**
+ * @name Operator Precedence Logic Error When Use Bool Type
+ * @description --Finding places of confusing use of boolean type.
+ *              --For example, a unary minus does not work before a boolean type and an increment always gives true.
+ * @kind problem
+ * @id cpp/operator-precedence-logic-error-when-use-bool-type
+ * @problem.severity warning
+ * @precision medium
+ * @tags correctness
+ *       security
+ *       external/cwe/cwe-783
+ *       external/cwe/cwe-480
+ */
+
+import cpp
+import semmle.code.cpp.valuenumbering.HashCons
+
+/** Holds if `exp` increments a boolean value. */
+predicate incrementBoolType(IncrementOperation exp) {
+  exp.getOperand().getType() instanceof BoolType
+}
+
+/** Holds if `exp` applies the unary minus operator to a boolean type. */
+predicate revertSignBoolType(UnaryMinusExpr exp) {
+  exp.getAnOperand().getType() instanceof BoolType and
+  exp.getFullyConverted().getType() instanceof BoolType
+}
+
+/** Holds, if this is an expression, uses comparison and assignment outside of execution precedence. */
+predicate assignBoolType(Expr exp) {
+  exists(ComparisonOperation co |
+    exp.(AssignExpr).getRValue() = co and
+    exp.isCondition() and
+    not co.isParenthesised() and
+    not exp.(AssignExpr).getLValue().getType() instanceof BoolType and
+    not exists(Expr exbl |
+      hashCons(exbl.(AssignExpr).getLValue()) = hashCons(exp.(AssignExpr).getLValue()) and
+      not exbl.isCondition() and
+      exbl.(AssignExpr).getRValue().getType() instanceof BoolType and
+      exbl.(AssignExpr).getLValue().getType() = exp.(AssignExpr).getLValue().getType()
+    ) and
+    co.getLeftOperand() instanceof FunctionCall and
+    not co.getRightOperand().getType() instanceof BoolType and
+    not co.getRightOperand().getValue() = "0" and
+    not co.getRightOperand().getValue() = "1"
+  )
+}
+
+from Expr exp
+where
+  incrementBoolType(exp) or
+  revertSignBoolType(exp) or
+  assignBoolType(exp)
+select exp, "this expression needs attention"

cpp/ql/src/experimental/Security/CWE/CWE-788/AccessOfMemoryLocationAfterEndOfBufferUsingStrlen.ql

@@ -3,7 +3,7 @@
  * @description The expression `buffer [strlen (buffer)] = 0` is potentially dangerous, if the variable `buffer` does not have a terminal zero, then access beyond the bounds of the allocated memory is possible, which will lead to undefined behavior.
  *              If terminal zero is present, then the specified expression is meaningless.
  * @kind problem
- * @id cpp/access-memory-location-after-end-buffer
+ * @id cpp/access-memory-location-after-end-buffer-strlen
  * @problem.severity warning
  * @precision medium
  * @tags correctness

cpp/ql/src/experimental/Security/CWE/CWE-788/AccessOfMemoryLocationAfterEndOfBufferUsingStrncat.ql

@@ -2,7 +2,7 @@
  * @name Access Of Memory Location After The End Of A Buffer Using Strncat
  * @description Calls of the form `strncat(dest, source, sizeof (dest) - strlen (dest))` set the third argument to one more than possible. So when `dest` is full, the expression `sizeof(dest) - strlen (dest)` will be equal to one, and not zero as the programmer might think. Making a call of this type may result in a zero byte being written just outside the `dest` buffer.
  * @kind problem
- * @id cpp/access-memory-location-after-end-buffer
+ * @id cpp/access-memory-location-after-end-buffer-strncat
  * @problem.severity warning
  * @precision medium
  * @tags correctness
@@ -11,54 +11,32 @@
  */
 
 import cpp
+import semmle.code.cpp.models.implementations.Strcat
 import semmle.code.cpp.valuenumbering.GlobalValueNumbering
 
 /**
- * A call to `strncat` of the form `strncat(buff, str, someExpr - strlen(buf))`, for some expression `someExpr` equal to `sizeof(buff)`.
+ * Holds if `call` is a call to `strncat` such that `sizeArg` and `destArg` are the size and
+ * destination arguments, respectively.
  */
-class WrongCallStrncat extends FunctionCall {
-  Expr leftsomeExpr;
-
-  WrongCallStrncat() {
-    this.getTarget().hasGlobalOrStdName("strncat") and
-    // the expression of the first argument in `strncat` and `strnlen` is identical
-    globalValueNumber(this.getArgument(0)) =
-      globalValueNumber(this.getArgument(2).(SubExpr).getRightOperand().(StrlenCall).getStringExpr()) and
-    // using a string constant often speaks of manually calculating the length of the required buffer.
-    (
-      not this.getArgument(1) instanceof StringLiteral and
-      not this.getArgument(1) instanceof CharLiteral
-    ) and
-    // for use in predicates
-    leftsomeExpr = this.getArgument(2).(SubExpr).getLeftOperand()
-  }
-
-  /**
-   * Holds if the left side of the expression `someExpr` equal to `sizeof(buf)`.
-   */
-  predicate isExpressionEqualSizeof() {
-    // the left side of the expression `someExpr` is `sizeof(buf)`.
-    globalValueNumber(this.getArgument(0)) =
-      globalValueNumber(leftsomeExpr.(SizeofExprOperator).getExprOperand())
-    or
-    // value of the left side of the expression `someExpr` equal  `sizeof(buf)` value, and `buf` is array.
-    leftsomeExpr.getValue().toInt() = this.getArgument(0).getType().getSize()
-  }
-
-  /**
-   * Holds if the left side of the expression `someExpr` equal to variable containing the length of the memory allocated for the buffer.
-   */
-  predicate isVariableEqualValueSizegBuffer() {
-    // the left side of expression `someExpr` is the variable that was used in the function of allocating memory for the buffer`.
-    exists(AllocationExpr alc |
-      leftsomeExpr.(VariableAccess).getTarget() =
-        alc.(FunctionCall).getArgument(0).(VariableAccess).getTarget()
-    )
-  }
+predicate interestringCallWithArgs(Call call, Expr sizeArg, Expr destArg) {
+  exists(StrcatFunction strcat |
+    strcat = call.getTarget() and
+    sizeArg = call.getArgument(strcat.getParamSize()) and
+    destArg = call.getArgument(strcat.getParamDest())
+  )
 }
 
-from WrongCallStrncat sc
+from FunctionCall call, Expr sizeArg, Expr destArg, SubExpr sub, int n
 where
-  sc.isExpressionEqualSizeof() or
-  sc.isVariableEqualValueSizegBuffer()
-select sc, "if the used buffer is full, writing out of the buffer is possible"
+  interestringCallWithArgs(call, sizeArg, destArg) and
+  // The destination buffer is an array of size n
+  destArg.getUnspecifiedType().(ArrayType).getSize() = n and
+  // The size argument is equivalent to a subtraction
+  globalValueNumber(sizeArg).getAnExpr() = sub and
+  // ... where the left side of the subtraction is the constant n
+  globalValueNumber(sub.getLeftOperand()).getAnExpr().getValue().toInt() = n and
+  // ... and the right side of the subtraction is a call to `strlen` where the argument is the
+  // destination buffer.
+  globalValueNumber(sub.getRightOperand()).getAnExpr().(StrlenCall).getStringExpr() =
+    globalValueNumber(destArg).getAnExpr()
+select call, "Possible out-of-bounds write due to incorrect size argument."

cpp/ql/src/external/tests/DefectFilter.ql

@@ -1,12 +0,0 @@
-/**
- * @name Defect filter
- * @description Only include results in large files (200) lines of code, and change the message.
- * @tags filter
- */
-
-import cpp
-import external.DefectFilter
-
-from DefectResult res
-where res.getFile().getMetrics().getNumberOfLinesOfCode() > 200
-select res, "Large files: " + res.getMessage()

cpp/ql/src/external/tests/DefectFromExternalData.ql

@@ -1,18 +0,0 @@
-/**
- * @name Defect from external data
- * @description Insert description here...
- * @kind problem
- * @problem.severity warning
- * @tags external-data
- */
-
-import cpp
-import external.ExternalArtifact
-
-from ExternalData d, File u
-where
-  d.getQueryPath() = "external-data.ql" and
-  u.getShortName() = d.getField(0)
-select u,
-  d.getField(5) + ", " + d.getFieldAsDate(1) + ", " + d.getField(2) + ", " + d.getFieldAsFloat(3) +
-    ", " + d.getFieldAsInt(4) + ": " + d.getNumFields()

cpp/ql/src/external/tests/MetricFilter.ql

@@ -1,12 +0,0 @@
-/**
- * @name Metric filter
- * @description Only include results in large files (200) lines of code.
- * @tags filter
- */
-
-import cpp
-import external.MetricFilter
-
-from MetricResult res
-where res.getFile().getMetrics().getNumberOfLinesOfCode() > 200
-select res, res.getValue()

cpp/ql/src/filters/FilterAutogenerated.ql

@@ -1,16 +0,0 @@
-/**
- * @name Filter: exclude results from files that are autogenerated
- * @description Use this filter to return results only if they are
- *              located in files that are maintained manually.
- * @kind problem
- * @id cpp/autogenerated-filter
- * @tags filter
- */
-
-import cpp
-import semmle.code.cpp.AutogeneratedFile
-import external.DefectFilter
-
-from DefectResult res
-where not res.getFile() instanceof AutogeneratedFile
-select res, res.getMessage()

cpp/ql/src/filters/FilterAutogeneratedForMetric.ql

@@ -1,16 +0,0 @@
-/**
- * @name Metric filter: exclude results from files that are autogenerated
- * @description Use this filter to return results only if they are
- *              located in files that are maintained manually.
- * @kind treemap
- * @id cpp/autogenerated-for-metric-filter
- * @tags filter
- */
-
-import cpp
-import semmle.code.cpp.AutogeneratedFile
-import external.MetricFilter
-
-from MetricResult res
-where not res.getFile() instanceof AutogeneratedFile
-select res, res.getValue()

cpp/ql/src/filters/FromSource.ql

@@ -1,16 +0,0 @@
-/**
- * @name Filter: exclude results from files for which we do not have
- *       source code
- * @description Use this filter to return results only if they are
- *              located in files for which we have source code.
- * @kind problem
- * @id cpp/from-source-filter
- * @tags filter
- */
-
-import cpp
-import external.DefectFilter
-
-from DefectResult res
-where res.getFile().fromSource()
-select res, res.getMessage()

cpp/ql/src/filters/Macros.ql

@@ -1,36 +0,0 @@
-/**
- * @name Filter: exclude results on lines covered by a macro expansion
- * @description Use this filter to return results only when there is no
- *              macro expansion whose location spans all the lines of
- *              the result's location.
- * @kind problem
- * @id cpp/macros-filter
- * @tags filter
- */
-
-import cpp
-import external.DefectFilter
-
-predicate macroLocation(File f, int startLine, int endLine) {
-  exists(MacroInvocation mi, Location l |
-    l = mi.getLocation() and
-    l.getFile() = f and
-    l.getStartLine() = startLine and
-    l.getEndLine() = endLine
-  )
-}
-
-predicate macroCovering(DefectResult r) {
-  exists(File f, int macroStart, int macroEnd, int defectStart, int defectEnd |
-    f = r.getFile() and
-    defectStart = r.getStartLine() and
-    defectEnd = r.getEndLine() and
-    macroLocation(f, macroStart, macroEnd) and
-    macroStart <= defectStart and
-    macroEnd >= defectEnd
-  )
-}
-
-from DefectResult res
-where not macroCovering(res)
-select res, res.getMessage()

cpp/ql/src/jsf/4.10 Classes/AV Rule 79.ql

@@ -91,16 +91,17 @@ private predicate exprReleases(Expr e, Expr released, string kind) {
   // `e` is a call to a release function and `released` is the released argument
   releaseExpr(e, released, kind)
   or
-  exists(Function f, int arg |
+  exists(int arg, VariableAccess access, Function f |
     // `e` is a call to a function that releases one of it's parameters,
     // and `released` is the corresponding argument
     (
       e.(FunctionCall).getTarget() = f or
       e.(FunctionCall).getTarget().(MemberFunction).getAnOverridingFunction+() = f
     ) and
+    access = f.getParameter(arg).getAnAccess() and
     e.(FunctionCall).getArgument(arg) = released and
     exprReleases(_,
-      exprOrDereference(globalValueNumber(f.getParameter(arg).getAnAccess()).getAnExpr()), kind)
+      pragma[only_bind_into](exprOrDereference(globalValueNumber(access).getAnExpr())), kind)
   )
   or
   exists(Function f, ThisExpr innerThis |
@@ -112,7 +113,7 @@ private predicate exprReleases(Expr e, Expr released, string kind) {
     ) and
     e.(FunctionCall).getQualifier() = exprOrDereference(released) and
     innerThis.getEnclosingFunction() = f and
-    exprReleases(_, globalValueNumber(innerThis).getAnExpr(), kind)
+    exprReleases(_, pragma[only_bind_into](globalValueNumber(innerThis).getAnExpr()), kind)
   )
 }
 

cpp/ql/src/jsf/4.24 Control Flow Structures/AV Rule 189.qhelp

@@ -33,7 +33,7 @@ the break statement only exits from one level of the loop.</p>
 </li>
 <li>
   Mats Henricson and Erik Nyquist, <i>Industrial Strength C++</i>, published by Prentice Hall PTR (1997). 
-  Chapter 4: Control Flow, Rule 4.6 (<a href="http://mongers.org/industrial-c++/">PDF</a>).
+  Chapter 4: Control Flow, Rule 4.6 (<a href="https://web.archive.org/web/20190919025638/https://mongers.org/industrial-c++/">PDF</a>).
 </li>
 <li>
   <a href="http://www.cplusplus.com/doc/tutorial/control/">www.cplusplus.com Control Structures</a>

cpp/ql/src/jsf/4.24 Control Flow Structures/AV Rule 201.qhelp

@@ -39,7 +39,7 @@ loop if the loop requires more complicated variable iteration.
 </li>
 <li>
   Mats Henricson and Erik Nyquist, <i>Industrial Strength C++</i>, published by Prentice Hall PTR (1997). 
-  Chapter 4: Control Flow, Rule 4.1 (<a href="http://mongers.org/industrial-c++/">PDF</a>).
+  Chapter 4: Control Flow, Rule 4.1 (<a href="https://web.archive.org/web/20190919025638/https://mongers.org/industrial-c++/">PDF</a>).
 </li>
 
 

cpp/ql/src/semmle/code/cpp/Location.qll

@@ -72,6 +72,7 @@ class Location extends @location {
   }
 
   /** Holds if `this` comes on a line strictly before `l`. */
+  pragma[inline]
   predicate isBefore(Location l) {
     this.getFile() = l.getFile() and this.getEndLine() < l.getStartLine()
   }

cpp/ql/src/semmle/code/cpp/Type.qll

@@ -101,6 +101,7 @@ class Type extends Locatable, @type {
    *
    * For example, starting with `const i64* const` in the context of `typedef long long i64;`, this predicate will return `long long*`.
    */
+  pragma[nomagic]
   Type getUnspecifiedType() { unspecifiedtype(underlyingElement(this), unresolveElement(result)) }
 
   /**

cpp/ql/src/semmle/code/cpp/controlflow/Dominance.qll

@@ -14,6 +14,7 @@ import cpp
  * In rare cases, the same node is used in multiple control-flow scopes. This
  * confuses the dominance analysis, so this predicate is used to exclude them.
  */
+pragma[noinline]
 private predicate hasMultiScopeNode(Function f) {
   exists(ControlFlowNode node |
     node.getControlFlowScope() = f and

cpp/ql/src/semmle/code/cpp/controlflow/internal/CFG.qll

@@ -1307,7 +1307,8 @@ private predicate conditionJumps(Expr test, boolean truth, Node n2, Pos p2) {
   )
 }
 
-// Factored out for performance. See QL-796.
+// Pulled out for performance. See
+// https://github.com/github/codeql-coreql-team/issues/1044.
 private predicate normalGroupMemberBaseCase(Node memberNode, Pos memberPos, Node atNode) {
   memberNode = atNode and
   memberPos.isAt() and

cpp/ql/src/semmle/code/cpp/controlflow/internal/ConstantExprs.qll

@@ -104,9 +104,43 @@ private predicate loopConditionAlwaysUponEntry(ControlFlowNode loop, Expr condit
   )
 }
 
+/**
+ * This relation is the same as the `el instanceof Function`, only obfuscated
+ * so the optimizer will not understand that any `FunctionCall.getTarget()`
+ * should be in this relation.
+ */
+pragma[noinline]
+private predicate isFunction(Element el) {
+  el instanceof Function
+  or
+  el.(Expr).getParent() = el
+}
+
+/**
+ * Holds if `fc` is a `FunctionCall` with no return value for `getTarget`. This
+ * can happen in case of rare database inconsistencies.
+ */
+pragma[noopt]
+private predicate callHasNoTarget(@funbindexpr fc) {
+  exists(Function f |
+    funbind(fc, f) and
+    not isFunction(f)
+  )
+}
+
+// Pulled out for performance. See
+// https://github.com/github/codeql-coreql-team/issues/1044.
+private predicate potentiallyReturningFunctionCall_base(FunctionCall fc) {
+  fc.isVirtual()
+  or
+  callHasNoTarget(fc)
+}
+
 /** A function call that *may* return; if in doubt, we assume it may. */
 private predicate potentiallyReturningFunctionCall(FunctionCall fc) {
-  potentiallyReturningFunction(fc.getTarget()) or fc.isVirtual()
+  potentiallyReturningFunctionCall_base(fc)
+  or
+  potentiallyReturningFunction(fc.getTarget())
 }
 
 /** A function that *may* return; if in doubt, we assume it may. */

cpp/ql/src/semmle/code/cpp/dataflow/internal/AddressFlow.qll

@@ -15,6 +15,7 @@
  */
 
 private import cpp
+private import semmle.code.cpp.models.interfaces.PointerWrapper
 
 /**
  * Holds if `f` is an instantiation of the `std::move` or `std::forward`
@@ -94,6 +95,12 @@ private predicate pointerToPointerStep(Expr pointerIn, Expr pointerOut) {
 
 private predicate lvalueToReferenceStep(Expr lvalueIn, Expr referenceOut) {
   lvalueIn.getConversion() = referenceOut.(ReferenceToExpr)
+  or
+  exists(PointerWrapper wrapper, Call call | call = referenceOut |
+    referenceOut.getUnspecifiedType() instanceof ReferenceType and
+    call = wrapper.getAnUnwrapperFunction().getACallToThisFunction() and
+    lvalueIn = call.getQualifier().getFullyConverted()
+  )
 }
 
 private predicate referenceToLvalueStep(Expr referenceIn, Expr lvalueOut) {
@@ -106,6 +113,13 @@ private predicate referenceToPointerStep(Expr referenceIn, Expr pointerOut) {
       stdAddressOf(call.getTarget()) and
       referenceIn = call.getArgument(0).getFullyConverted()
     )
+  or
+  exists(CopyConstructor copy, Call call | call = pointerOut |
+    copy.getDeclaringType() instanceof PointerWrapper and
+    call.getTarget() = copy and
+    // The 0'th argument is the value being copied.
+    referenceIn = call.getArgument(0).getFullyConverted()
+  )
 }
 
 private predicate referenceToReferenceStep(Expr referenceIn, Expr referenceOut) {
@@ -190,6 +204,19 @@ private predicate pointerToUpdate(Expr pointer, Expr outer, ControlFlowNode node
         // See the `lvalueToUpdate` case for an explanation of this conjunct.
         call.getType().isDeeplyConstBelow()
       )
+      or
+      // Pointer wrappers behave as raw pointers for dataflow purposes.
+      outer = call.getAnArgument().getFullyConverted() and
+      exists(PointerWrapper wrapper | wrapper = outer.getType().stripTopLevelSpecifiers() |
+        not wrapper.pointsToConst()
+      )
+      or
+      outer = call.getQualifier().getFullyConverted() and
+      outer.getUnspecifiedType() instanceof PointerWrapper and
+      not (
+        call.getTarget().hasSpecifier("const") and
+        call.getType().isDeeplyConstBelow()
+      )
     )
     or
     exists(PointerFieldAccess fa |
@@ -218,7 +245,9 @@ private predicate referenceToUpdate(Expr reference, Expr outer, ControlFlowNode
     not stdIdentityFunction(call.getTarget()) and
     not stdAddressOf(call.getTarget()) and
     exists(ReferenceType rt | rt = outer.getType().stripTopLevelSpecifiers() |
-      not rt.getBaseType().isConst()
+      not rt.getBaseType().isConst() or
+      rt.getBaseType().getUnspecifiedType() =
+        any(PointerWrapper wrapper | not wrapper.pointsToConst())
     )
   ) and
   reference = outer

cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImpl.qll

@@ -2133,11 +2133,8 @@ private module Stage4 {
 
   bindingset[node, cc, config]
   private LocalCc getLocalCc(Node node, Cc cc, Configuration config) {
-    exists(Cc cc0 |
-      cc = pragma[only_bind_into](cc0) and
-      localFlowEntry(node, config) and
-      result = getLocalCallContext(cc0, getNodeEnclosingCallable(node))
-    )
+    localFlowEntry(node, config) and
+    result = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(node))
   }
 
   private predicate localStep(
@@ -3132,7 +3129,7 @@ private predicate pathStep(PathNodeMid mid, Node node, CallContext cc, SummaryCt
     conf = mid.getConfiguration() and
     cc = mid.getCallContext() and
     sc = mid.getSummaryCtx() and
-    localCC = getLocalCallContext(cc, getNodeEnclosingCallable(midnode)) and
+    localCC = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(midnode)) and
     ap0 = mid.getAp()
   |
     localFlowBigStep(midnode, node, true, _, conf, localCC) and

cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImpl2.qll

@@ -2133,11 +2133,8 @@ private module Stage4 {
 
   bindingset[node, cc, config]
   private LocalCc getLocalCc(Node node, Cc cc, Configuration config) {
-    exists(Cc cc0 |
-      cc = pragma[only_bind_into](cc0) and
-      localFlowEntry(node, config) and
-      result = getLocalCallContext(cc0, getNodeEnclosingCallable(node))
-    )
+    localFlowEntry(node, config) and
+    result = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(node))
   }
 
   private predicate localStep(
@@ -3132,7 +3129,7 @@ private predicate pathStep(PathNodeMid mid, Node node, CallContext cc, SummaryCt
     conf = mid.getConfiguration() and
     cc = mid.getCallContext() and
     sc = mid.getSummaryCtx() and
-    localCC = getLocalCallContext(cc, getNodeEnclosingCallable(midnode)) and
+    localCC = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(midnode)) and
     ap0 = mid.getAp()
   |
     localFlowBigStep(midnode, node, true, _, conf, localCC) and

cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImpl3.qll

@@ -2133,11 +2133,8 @@ private module Stage4 {
 
   bindingset[node, cc, config]
   private LocalCc getLocalCc(Node node, Cc cc, Configuration config) {
-    exists(Cc cc0 |
-      cc = pragma[only_bind_into](cc0) and
-      localFlowEntry(node, config) and
-      result = getLocalCallContext(cc0, getNodeEnclosingCallable(node))
-    )
+    localFlowEntry(node, config) and
+    result = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(node))
   }
 
   private predicate localStep(
@@ -3132,7 +3129,7 @@ private predicate pathStep(PathNodeMid mid, Node node, CallContext cc, SummaryCt
     conf = mid.getConfiguration() and
     cc = mid.getCallContext() and
     sc = mid.getSummaryCtx() and
-    localCC = getLocalCallContext(cc, getNodeEnclosingCallable(midnode)) and
+    localCC = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(midnode)) and
     ap0 = mid.getAp()
   |
     localFlowBigStep(midnode, node, true, _, conf, localCC) and

cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImpl4.qll

@@ -2133,11 +2133,8 @@ private module Stage4 {
 
   bindingset[node, cc, config]
   private LocalCc getLocalCc(Node node, Cc cc, Configuration config) {
-    exists(Cc cc0 |
-      cc = pragma[only_bind_into](cc0) and
-      localFlowEntry(node, config) and
-      result = getLocalCallContext(cc0, getNodeEnclosingCallable(node))
-    )
+    localFlowEntry(node, config) and
+    result = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(node))
   }
 
   private predicate localStep(
@@ -3132,7 +3129,7 @@ private predicate pathStep(PathNodeMid mid, Node node, CallContext cc, SummaryCt
     conf = mid.getConfiguration() and
     cc = mid.getCallContext() and
     sc = mid.getSummaryCtx() and
-    localCC = getLocalCallContext(cc, getNodeEnclosingCallable(midnode)) and
+    localCC = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(midnode)) and
     ap0 = mid.getAp()
   |
     localFlowBigStep(midnode, node, true, _, conf, localCC) and

cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImplCommon.qll

@@ -31,7 +31,7 @@ predicate accessPathCostLimits(int apLimit, int tupleLimit) {
  * currently excludes read-steps, store-steps, and flow-through.
  *
  * The analysis uses non-linear recursion: When computing a flow path in or out
- * of a call, we use the results of the analysis recursively to resolve lamba
+ * of a call, we use the results of the analysis recursively to resolve lambda
  * calls. For this reason, we cannot reuse the code from `DataFlowImpl.qll` directly.
  */
 private module LambdaFlow {

cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowImplLocal.qll

@@ -2133,11 +2133,8 @@ private module Stage4 {
 
   bindingset[node, cc, config]
   private LocalCc getLocalCc(Node node, Cc cc, Configuration config) {
-    exists(Cc cc0 |
-      cc = pragma[only_bind_into](cc0) and
-      localFlowEntry(node, config) and
-      result = getLocalCallContext(cc0, getNodeEnclosingCallable(node))
-    )
+    localFlowEntry(node, config) and
+    result = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(node))
   }
 
   private predicate localStep(
@@ -3132,7 +3129,7 @@ private predicate pathStep(PathNodeMid mid, Node node, CallContext cc, SummaryCt
     conf = mid.getConfiguration() and
     cc = mid.getCallContext() and
     sc = mid.getSummaryCtx() and
-    localCC = getLocalCallContext(cc, getNodeEnclosingCallable(midnode)) and
+    localCC = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(midnode)) and
     ap0 = mid.getAp()
   |
     localFlowBigStep(midnode, node, true, _, conf, localCC) and

cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowPrivate.qll

@@ -321,5 +321,5 @@ predicate lambdaCreation(Node creation, LambdaCallKind kind, DataFlowCallable c)
 /** 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 lamba flow analysis. */
+/** Extra data-flow steps needed for lambda flow analysis. */
 predicate additionalLambdaFlowStep(Node nodeFrom, Node nodeTo, boolean preservesValue) { none() }

cpp/ql/src/semmle/code/cpp/dataflow/internal/DataFlowUtil.qll

@@ -46,7 +46,7 @@ class Node extends TNode {
   /**
    * INTERNAL: Do not use. Alternative name for `getFunction`.
    */
-  final Function getEnclosingCallable() { result = unique(Function f | f = this.getFunction() | f) }
+  final Function getEnclosingCallable() { result = this.getFunction() }
 
   /** Gets the type of this node. */
   Type getType() { none() } // overridden in subclasses
@@ -324,7 +324,7 @@ private class VariablePartialDefinitionNode extends PartialDefinitionNode {
  * A synthetic data flow node used for flow into a collection when an iterator
  * write occurs in a callee.
  */
-class IteratorPartialDefinitionNode extends PartialDefinitionNode {
+private class IteratorPartialDefinitionNode extends PartialDefinitionNode {
   override IteratorPartialDefinition pd;
 
   override Node getPreUpdateNode() { pd.definesExpressions(_, result.asExpr()) }
@@ -694,7 +694,12 @@ private predicate exprToExprStep_nocfg(Expr fromExpr, Expr toExpr) {
           fromExpr = call.getQualifier()
         ) and
         call.getTarget() = f and
-        outModel.isReturnValue()
+        // AST dataflow treats a reference as if it were the referred-to object, while the dataflow
+        // models treat references as pointers. If the return type of the call is a reference, then
+        // look for data flow the the referred-to object, rather than the reference itself.
+        if call.getType().getUnspecifiedType() instanceof ReferenceType
+        then outModel.isReturnValueDeref()
+        else outModel.isReturnValue()
       )
     )
 }
@@ -715,6 +720,7 @@ private predicate exprToDefinitionByReferenceStep(Expr exprIn, Expr argOut) {
 }
 
 private module FieldFlow {
+  private import DataFlowImplCommon
   private import DataFlowImplLocal
   private import DataFlowPrivate
 
@@ -747,7 +753,7 @@ private module FieldFlow {
     exists(FieldConfiguration cfg | cfg.hasFlow(node1, node2)) and
     // This configuration should not be able to cross function boundaries, but
     // we double-check here just to be sure.
-    node1.getEnclosingCallable() = node2.getEnclosingCallable()
+    getNodeEnclosingCallable(node1) = getNodeEnclosingCallable(node2)
   }
 }
 

cpp/ql/src/semmle/code/cpp/dataflow/internal/FlowVar.qll

@@ -7,6 +7,7 @@ private import semmle.code.cpp.controlflow.SSA
 private import semmle.code.cpp.dataflow.internal.SubBasicBlocks
 private import semmle.code.cpp.dataflow.internal.AddressFlow
 private import semmle.code.cpp.models.implementations.Iterator
+private import semmle.code.cpp.models.interfaces.PointerWrapper
 
 /**
  * A conceptual variable that is assigned only once, like an SSA variable. This
@@ -158,18 +159,14 @@ private module PartialDefinitions {
     Expr innerDefinedExpr;
 
     IteratorPartialDefinition() {
-      exists(Expr convertedInner |
-        not this instanceof Conversion and
-        valueToUpdate(convertedInner, this.getFullyConverted(), node) and
-        innerDefinedExpr = convertedInner.getUnconverted() and
-        (
-          innerDefinedExpr.(Call).getQualifier() = getAnIteratorAccess(collection)
-          or
-          innerDefinedExpr.(Call).getQualifier() = collection.getAnAccess() and
-          collection instanceof IteratorParameter
-        ) and
-        innerDefinedExpr.(Call).getTarget() instanceof IteratorPointerDereferenceMemberOperator
-      )
+      innerDefinedExpr = getInnerDefinedExpr(this, node) and
+      (
+        innerDefinedExpr.(Call).getQualifier() = getAnIteratorAccess(collection)
+        or
+        innerDefinedExpr.(Call).getQualifier() = collection.getAnAccess() and
+        collection instanceof IteratorParameter
+      ) and
+      innerDefinedExpr.(Call).getTarget() instanceof IteratorPointerDereferenceMemberOperator
       or
       // iterators passed by value without a copy constructor
       exists(Call call |
@@ -207,16 +204,18 @@ private module PartialDefinitions {
     }
   }
 
+  private Expr getInnerDefinedExpr(Expr e, ControlFlowNode node) {
+    not e instanceof Conversion and
+    exists(Expr convertedInner |
+      valueToUpdate(convertedInner, e.getFullyConverted(), node) and
+      result = convertedInner.getUnconverted()
+    )
+  }
+
   class VariablePartialDefinition extends PartialDefinition {
     Expr innerDefinedExpr;
 
-    VariablePartialDefinition() {
-      not this instanceof Conversion and
-      exists(Expr convertedInner |
-        valueToUpdate(convertedInner, this.getFullyConverted(), node) and
-        innerDefinedExpr = convertedInner.getUnconverted()
-      )
-    }
+    VariablePartialDefinition() { innerDefinedExpr = getInnerDefinedExpr(this, node) }
 
     deprecated override predicate partiallyDefines(Variable v) {
       innerDefinedExpr = v.getAnAccess()
@@ -296,7 +295,8 @@ module FlowVar_internal {
     // treating them as immutable, but for data flow it gives better results in
     // practice to make the variable synonymous with its contents.
     not v.getUnspecifiedType() instanceof ReferenceType and
-    not v instanceof IteratorParameter
+    not v instanceof IteratorParameter and
+    not v instanceof PointerWrapperParameter
   }
 
   /**
@@ -644,10 +644,19 @@ module FlowVar_internal {
   predicate parameterIsNonConstReference(Parameter p) {
     exists(ReferenceType refType |
       refType = p.getUnderlyingType() and
-      not refType.getBaseType().isConst()
+      (
+        not refType.getBaseType().isConst()
+        or
+        // A field of a parameter of type `const std::shared_ptr<A>& p` can still be changed even though
+        // the base type of the reference is `const`.
+        refType.getBaseType().getUnspecifiedType() =
+          any(PointerWrapper wrapper | not wrapper.pointsToConst())
+      )
     )
     or
     p instanceof IteratorParameter
+    or
+    p instanceof PointerWrapperParameter
   }
 
   /**
@@ -836,6 +845,10 @@ module FlowVar_internal {
     IteratorParameter() { this.getUnspecifiedType() instanceof Iterator }
   }
 
+  class PointerWrapperParameter extends Parameter {
+    PointerWrapperParameter() { this.getUnspecifiedType() instanceof PointerWrapper }
+  }
+
   /**
    * Holds if `v` is initialized to have value `assignedExpr`.
    */

cpp/ql/src/semmle/code/cpp/dataflow/internal/TaintTrackingUtil.qll

@@ -11,6 +11,7 @@
 private import semmle.code.cpp.models.interfaces.DataFlow
 private import semmle.code.cpp.models.interfaces.Taint
 private import semmle.code.cpp.models.interfaces.Iterator
+private import semmle.code.cpp.models.interfaces.PointerWrapper
 
 private module DataFlow {
   import semmle.code.cpp.dataflow.internal.DataFlowUtil
@@ -141,7 +142,10 @@ private predicate noFlowFromChildExpr(Expr e) {
   or
   e instanceof LogicalOrExpr
   or
-  e instanceof Call
+  // Allow taint from `operator*` on smart pointers.
+  exists(Call call | e = call |
+    not call.getTarget() = any(PointerWrapper wrapper).getAnUnwrapperFunction()
+  )
   or
   e instanceof SizeofOperator
   or

cpp/ql/src/semmle/code/cpp/exprs/Call.qll

@@ -300,17 +300,25 @@ class FunctionCall extends Call, @funbindexpr {
   }
 }
 
+/** A _user-defined_ unary `operator*` function. */
+class OverloadedPointerDereferenceFunction extends Function {
+  OverloadedPointerDereferenceFunction() {
+    this.hasName("operator*") and
+    this.getEffectiveNumberOfParameters() = 1
+  }
+}
+
 /**
  * An instance of a _user-defined_ unary `operator*` applied to its argument.
  * ```
  * T1 operator*(const T2 &);
  * T1 a; T2 b;
  * a = *b;
+ * ```
  */
 class OverloadedPointerDereferenceExpr extends FunctionCall {
   OverloadedPointerDereferenceExpr() {
-    getTarget().hasName("operator*") and
-    getTarget().getEffectiveNumberOfParameters() = 1
+    this.getTarget() instanceof OverloadedPointerDereferenceFunction
   }
 
   override string getAPrimaryQlClass() { result = "OverloadedPointerDereferenceExpr" }

cpp/ql/src/semmle/code/cpp/exprs/Expr.qll

@@ -1271,7 +1271,8 @@ private predicate convparents(Expr child, int idx, Element parent) {
   )
 }
 
-// Pulled out for performance. See QL-796.
+// Pulled out for performance. See
+// https://github.com/github/codeql-coreql-team/issues/1044.
 private predicate hasNoConversions(Expr e) { not e.hasConversion() }
 
 /**

cpp/ql/src/semmle/code/cpp/ir/dataflow/DefaultTaintTracking.qll

@@ -2,13 +2,16 @@ import cpp
 import semmle.code.cpp.security.Security
 private import semmle.code.cpp.ir.dataflow.DataFlow
 private import semmle.code.cpp.ir.dataflow.internal.DataFlowUtil
-private import semmle.code.cpp.ir.dataflow.DataFlow2
 private import semmle.code.cpp.ir.dataflow.DataFlow3
 private import semmle.code.cpp.ir.IR
 private import semmle.code.cpp.ir.dataflow.internal.DataFlowDispatch as Dispatch
 private import semmle.code.cpp.controlflow.IRGuards
 private import semmle.code.cpp.models.interfaces.Taint
 private import semmle.code.cpp.models.interfaces.DataFlow
+private import semmle.code.cpp.ir.dataflow.TaintTracking
+private import semmle.code.cpp.ir.dataflow.TaintTracking2
+private import semmle.code.cpp.ir.dataflow.TaintTracking3
+private import semmle.code.cpp.ir.dataflow.internal.ModelUtil
 
 /**
  * A predictable instruction is one where an external user can predict
@@ -65,23 +68,19 @@ private DataFlow::Node getNodeForExpr(Expr node) {
   not argv(node.(VariableAccess).getTarget())
 }
 
-private class DefaultTaintTrackingCfg extends DataFlow::Configuration {
+private class DefaultTaintTrackingCfg extends TaintTracking::Configuration {
   DefaultTaintTrackingCfg() { this = "DefaultTaintTrackingCfg" }
 
   override predicate isSource(DataFlow::Node source) { source = getNodeForSource(_) }
 
   override predicate isSink(DataFlow::Node sink) { exists(adjustedSink(sink)) }
 
-  override predicate isAdditionalFlowStep(DataFlow::Node n1, DataFlow::Node n2) {
-    commonTaintStep(n1, n2)
-  }
+  override predicate isSanitizer(DataFlow::Node node) { nodeIsBarrier(node) }
 
-  override predicate isBarrier(DataFlow::Node node) { nodeIsBarrier(node) }
-
-  override predicate isBarrierIn(DataFlow::Node node) { nodeIsBarrierIn(node) }
+  override predicate isSanitizerIn(DataFlow::Node node) { nodeIsBarrierIn(node) }
 }
 
-private class ToGlobalVarTaintTrackingCfg extends DataFlow::Configuration {
+private class ToGlobalVarTaintTrackingCfg extends TaintTracking::Configuration {
   ToGlobalVarTaintTrackingCfg() { this = "GlobalVarTaintTrackingCfg" }
 
   override predicate isSource(DataFlow::Node source) { source = getNodeForSource(_) }
@@ -90,20 +89,18 @@ private class ToGlobalVarTaintTrackingCfg extends DataFlow::Configuration {
     sink.asVariable() instanceof GlobalOrNamespaceVariable
   }
 
-  override predicate isAdditionalFlowStep(DataFlow::Node n1, DataFlow::Node n2) {
-    commonTaintStep(n1, n2)
-    or
+  override predicate isAdditionalTaintStep(DataFlow::Node n1, DataFlow::Node n2) {
     writesVariable(n1.asInstruction(), n2.asVariable().(GlobalOrNamespaceVariable))
     or
     readsVariable(n2.asInstruction(), n1.asVariable().(GlobalOrNamespaceVariable))
   }
 
-  override predicate isBarrier(DataFlow::Node node) { nodeIsBarrier(node) }
+  override predicate isSanitizer(DataFlow::Node node) { nodeIsBarrier(node) }
 
-  override predicate isBarrierIn(DataFlow::Node node) { nodeIsBarrierIn(node) }
+  override predicate isSanitizerIn(DataFlow::Node node) { nodeIsBarrierIn(node) }
 }
 
-private class FromGlobalVarTaintTrackingCfg extends DataFlow2::Configuration {
+private class FromGlobalVarTaintTrackingCfg extends TaintTracking2::Configuration {
   FromGlobalVarTaintTrackingCfg() { this = "FromGlobalVarTaintTrackingCfg" }
 
   override predicate isSource(DataFlow::Node source) {
@@ -114,18 +111,16 @@ private class FromGlobalVarTaintTrackingCfg extends DataFlow2::Configuration {
 
   override predicate isSink(DataFlow::Node sink) { exists(adjustedSink(sink)) }
 
-  override predicate isAdditionalFlowStep(DataFlow::Node n1, DataFlow::Node n2) {
-    commonTaintStep(n1, n2)
-    or
+  override predicate isAdditionalTaintStep(DataFlow::Node n1, DataFlow::Node n2) {
     // Additional step for flow out of variables. There is no flow _into_
     // variables in this configuration, so this step only serves to take flow
     // out of a variable that's a source.
     readsVariable(n2.asInstruction(), n1.asVariable())
   }
 
-  override predicate isBarrier(DataFlow::Node node) { nodeIsBarrier(node) }
+  override predicate isSanitizer(DataFlow::Node node) { nodeIsBarrier(node) }
 
-  override predicate isBarrierIn(DataFlow::Node node) { nodeIsBarrierIn(node) }
+  override predicate isSanitizerIn(DataFlow::Node node) { nodeIsBarrierIn(node) }
 }
 
 private predicate readsVariable(LoadInstruction load, Variable var) {
@@ -202,206 +197,26 @@ private predicate nodeIsBarrierIn(DataFlow::Node node) {
     // `getNodeForSource`.
     node = DataFlow::definitionByReferenceNodeFromArgument(source)
   )
-}
-
-cached
-private predicate commonTaintStep(DataFlow::Node fromNode, DataFlow::Node toNode) {
-  operandToInstructionTaintStep(fromNode.asOperand(), toNode.asInstruction())
   or
-  instructionToOperandTaintStep(fromNode.asInstruction(), toNode.asOperand())
-}
-
-private predicate instructionToOperandTaintStep(Instruction fromInstr, Operand toOperand) {
-  // Propagate flow from the definition of an operand to the operand, even when the overlap is inexact.
-  // We only do this in certain cases:
-  // 1. The instruction's result must not be conflated, and
-  // 2. The instruction's result type is one the types where we expect element-to-object flow. Currently
-  // this is array types and union types. This matches the other two cases of element-to-object flow in
-  // `DefaultTaintTracking`.
-  toOperand.getAnyDef() = fromInstr and
-  not fromInstr.isResultConflated() and
-  (
-    fromInstr.getResultType() instanceof ArrayType or
-    fromInstr.getResultType() instanceof Union
+  // don't use dataflow into binary instructions if both operands are unpredictable
+  exists(BinaryInstruction iTo |
+    iTo = node.asInstruction() and
+    not predictableInstruction(iTo.getLeft()) and
+    not predictableInstruction(iTo.getRight()) and
+    // propagate taint from either the pointer or the offset, regardless of predictability
+    not iTo instanceof PointerArithmeticInstruction
   )
   or
-  exists(ReadSideEffectInstruction readInstr |
-    fromInstr = readInstr.getArgumentDef() and
-    toOperand = readInstr.getSideEffectOperand()
-  )
-}
-
-private predicate operandToInstructionTaintStep(Operand fromOperand, Instruction toInstr) {
-  // Expressions computed from tainted data are also tainted
-  exists(CallInstruction call, int argIndex | call = toInstr |
-    isPureFunction(call.getStaticCallTarget().getName()) and
-    fromOperand = getACallArgumentOrIndirection(call, argIndex) and
-    forall(Operand argOperand | argOperand = call.getAnArgumentOperand() |
-      argOperand = getACallArgumentOrIndirection(call, argIndex) or
-      predictableInstruction(argOperand.getAnyDef())
-    ) and
-    // flow through `strlen` tends to cause dubious results, if the length is
-    // bounded.
-    not call.getStaticCallTarget().getName() = "strlen"
-  )
-  or
-  // Flow from argument to return value
-  toInstr =
-    any(CallInstruction call |
-      exists(int indexIn |
-        modelTaintToReturnValue(call.getStaticCallTarget(), indexIn) and
-        fromOperand = getACallArgumentOrIndirection(call, indexIn) and
-        not predictableOnlyFlow(call.getStaticCallTarget().getName())
-      )
-    )
-  or
-  // Flow from input argument to output argument
-  // TODO: This won't work in practice as long as all aliased memory is tracked
-  // together in a single virtual variable.
-  // TODO: Will this work on the test for `TaintedPath.ql`, where the output arg
-  // is a pointer addition expression?
-  toInstr =
-    any(WriteSideEffectInstruction outInstr |
-      exists(CallInstruction call, int indexIn, int indexOut |
-        modelTaintToParameter(call.getStaticCallTarget(), indexIn, indexOut) and
-        fromOperand = getACallArgumentOrIndirection(call, indexIn) and
-        outInstr.getIndex() = indexOut and
-        outInstr.getPrimaryInstruction() = call
-      )
-    )
-  or
-  // Flow through pointer dereference
-  toInstr.(LoadInstruction).getSourceAddressOperand() = fromOperand
-  or
-  // Flow through partial reads of arrays and unions
-  toInstr.(LoadInstruction).getSourceValueOperand() = fromOperand and
-  exists(Instruction fromInstr | fromInstr = fromOperand.getAnyDef() |
-    not fromInstr.isResultConflated() and
-    (
-      fromInstr.getResultType() instanceof ArrayType or
-      fromInstr.getResultType() instanceof Union
-    )
-  )
-  or
-  // Unary instructions tend to preserve enough information in practice that we
-  // want taint to flow through.
-  // The exception is `FieldAddressInstruction`. Together with the rule for
-  // `LoadInstruction` above and for `ChiInstruction` below, 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.
-  toInstr.(UnaryInstruction).getUnaryOperand() = fromOperand and
-  (
-    not toInstr instanceof FieldAddressInstruction
-    or
-    toInstr.(FieldAddressInstruction).getField().getDeclaringType() instanceof Union
-  )
-  or
-  // Flow from an element to an array or union that contains it.
-  toInstr.(ChiInstruction).getPartialOperand() = fromOperand and
-  not toInstr.isResultConflated() and
-  exists(Type t | toInstr.getResultLanguageType().hasType(t, false) |
-    t instanceof Union
-    or
-    t instanceof ArrayType
-  )
-  or
-  exists(BinaryInstruction bin |
-    bin = toInstr and
-    predictableInstruction(toInstr.getAnOperand().getDef()) and
-    fromOperand = toInstr.getAnOperand()
-  )
-  or
-  // This is part of the translation of `a[i]`, where we want taint to flow
-  // from `a`.
-  toInstr.(PointerAddInstruction).getLeftOperand() = fromOperand
-  or
-  // Until we have flow through indirections across calls, we'll take flow out
-  // of the indirection and into the argument.
-  // When we get proper flow through indirections across calls, this code can be
-  // moved to `adjusedSink` or possibly into the `DataFlow::ExprNode` class.
-  exists(ReadSideEffectInstruction read |
-    read.getSideEffectOperand() = fromOperand and
-    read.getArgumentDef() = toInstr
-  )
-  or
-  // Until we have from through indirections across calls, we'll take flow out
-  // of the parameter and into its indirection.
-  // `InitializeIndirectionInstruction` only has a single operand: the address of the
-  // value whose indirection we are initializing. When initializing an indirection of a parameter `p`,
-  // the IR looks like this:
-  // ```
-  // m1 = InitializeParameter[p] : &r1
-  // r2 = Load[p] : r2, m1
-  // m3 = InitializeIndirection[p] : &r2
-  // ```
-  // So by having flow from `r2` to `m3` we're enabling flow from `m1` to `m3`. This relies on the
-  // `LoadOperand`'s overlap being exact.
-  toInstr.(InitializeIndirectionInstruction).getAnOperand() = fromOperand
-}
-
-/**
- * Returns the index of the side effect instruction corresponding to the specified function output,
- * if one exists.
- */
-private int getWriteSideEffectIndex(FunctionOutput output) {
-  output.isParameterDeref(result)
-  or
-  output.isQualifierObject() and result = -1
-}
-
-/**
- * Get an operand that goes into argument `argumentIndex` of `call`. This
- * can be either directly or through one pointer indirection.
- */
-private Operand getACallArgumentOrIndirection(CallInstruction call, int argumentIndex) {
-  result = call.getPositionalArgumentOperand(argumentIndex)
-  or
-  exists(ReadSideEffectInstruction readSE |
-    // TODO: why are read side effect operands imprecise?
-    result = readSE.getSideEffectOperand() and
-    readSE.getPrimaryInstruction() = call and
-    readSE.getIndex() = argumentIndex
-  )
-}
-
-private predicate modelTaintToParameter(Function f, int parameterIn, int parameterOut) {
-  exists(FunctionInput modelIn, FunctionOutput modelOut |
-    (
-      f.(DataFlowFunction).hasDataFlow(modelIn, modelOut)
-      or
-      f.(TaintFunction).hasTaintFlow(modelIn, modelOut)
-    ) and
-    (modelIn.isParameter(parameterIn) or modelIn.isParameterDeref(parameterIn)) and
-    parameterOut = getWriteSideEffectIndex(modelOut)
-  )
-}
-
-private predicate modelTaintToReturnValue(Function f, int parameterIn) {
-  // Taint flow from parameter to return value
-  exists(FunctionInput modelIn, FunctionOutput modelOut |
-    f.(TaintFunction).hasTaintFlow(modelIn, modelOut) and
-    (modelIn.isParameter(parameterIn) or modelIn.isParameterDeref(parameterIn)) and
-    (modelOut.isReturnValue() or modelOut.isReturnValueDeref())
-  )
-  or
-  // Data flow (not taint flow) to where the return value points. For the time
-  // being we will conflate pointers and objects in taint tracking.
-  exists(FunctionInput modelIn, FunctionOutput modelOut |
-    f.(DataFlowFunction).hasDataFlow(modelIn, modelOut) and
-    (modelIn.isParameter(parameterIn) or modelIn.isParameterDeref(parameterIn)) and
-    modelOut.isReturnValueDeref()
-  )
-  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(int parameterMid, InParameter modelMid, OutReturnValue returnOut |
-    modelTaintToParameter(f, parameterIn, parameterMid) and
-    modelMid.isParameter(parameterMid) and
-    f.(DataFlowFunction).hasDataFlow(modelMid, returnOut)
+  // don't use dataflow through calls to pure functions if two or more operands
+  // are unpredictable
+  exists(Instruction iFrom1, Instruction iFrom2, CallInstruction iTo |
+    iTo = node.asInstruction() and
+    isPureFunction(iTo.getStaticCallTarget().getName()) and
+    iFrom1 = iTo.getAnArgument() and
+    iFrom2 = iTo.getAnArgument() and
+    not predictableInstruction(iFrom1) and
+    not predictableInstruction(iFrom2) and
+    iFrom1 != iFrom2
   )
 }
 
@@ -440,6 +255,14 @@ private Element adjustedSink(DataFlow::Node sink) {
   or
   // Taint `e1 += e2`, `e &= e2` and friends when `e1` or `e2` is tainted.
   result.(AssignOperation).getAnOperand() = sink.asExpr()
+  or
+  result =
+    sink.asOperand()
+        .(SideEffectOperand)
+        .getUse()
+        .(ReadSideEffectInstruction)
+        .getArgumentDef()
+        .getUnconvertedResultExpression()
 }
 
 /**
@@ -558,7 +381,7 @@ module TaintedWithPath {
     string toString() { result = "TaintTrackingConfiguration" }
   }
 
-  private class AdjustedConfiguration extends DataFlow3::Configuration {
+  private class AdjustedConfiguration extends TaintTracking3::Configuration {
     AdjustedConfiguration() { this = "AdjustedConfiguration" }
 
     override predicate isSource(DataFlow::Node source) {
@@ -571,21 +394,34 @@ module TaintedWithPath {
       exists(TaintTrackingConfiguration cfg | cfg.isSink(adjustedSink(sink)))
     }
 
-    override predicate isAdditionalFlowStep(DataFlow::Node n1, DataFlow::Node n2) {
-      commonTaintStep(n1, n2)
-      or
+    override predicate isAdditionalTaintStep(DataFlow::Node n1, DataFlow::Node n2) {
+      // Steps into and out of global variables
       exists(TaintTrackingConfiguration cfg | cfg.taintThroughGlobals() |
         writesVariable(n1.asInstruction(), n2.asVariable().(GlobalOrNamespaceVariable))
         or
         readsVariable(n2.asInstruction(), n1.asVariable().(GlobalOrNamespaceVariable))
       )
+      or
+      // Step to return value of a modeled function when an input taints the
+      // dereference of the return value
+      exists(CallInstruction call, Function func, FunctionInput modelIn, FunctionOutput modelOut |
+        n1.asOperand() = callInput(call, modelIn) and
+        (
+          func.(TaintFunction).hasTaintFlow(modelIn, modelOut)
+          or
+          func.(DataFlowFunction).hasDataFlow(modelIn, modelOut)
+        ) and
+        call.getStaticCallTarget() = func and
+        modelOut.isReturnValueDeref() and
+        call = n2.asInstruction()
+      )
     }
 
-    override predicate isBarrier(DataFlow::Node node) {
+    override predicate isSanitizer(DataFlow::Node node) {
       exists(TaintTrackingConfiguration cfg, Expr e | cfg.isBarrier(e) and node = getNodeForExpr(e))
     }
 
-    override predicate isBarrierIn(DataFlow::Node node) { nodeIsBarrierIn(node) }
+    override predicate isSanitizerIn(DataFlow::Node node) { nodeIsBarrierIn(node) }
   }
 
   /*

cpp/ql/src/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 semmle.code.cpp.ir.dataflow.internal.tainttracking3.TaintTrackingImpl
+}

cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/DataFlowImpl.qll

@@ -2133,11 +2133,8 @@ private module Stage4 {
 
   bindingset[node, cc, config]
   private LocalCc getLocalCc(Node node, Cc cc, Configuration config) {
-    exists(Cc cc0 |
-      cc = pragma[only_bind_into](cc0) and
-      localFlowEntry(node, config) and
-      result = getLocalCallContext(cc0, getNodeEnclosingCallable(node))
-    )
+    localFlowEntry(node, config) and
+    result = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(node))
   }
 
   private predicate localStep(
@@ -3132,7 +3129,7 @@ private predicate pathStep(PathNodeMid mid, Node node, CallContext cc, SummaryCt
     conf = mid.getConfiguration() and
     cc = mid.getCallContext() and
     sc = mid.getSummaryCtx() and
-    localCC = getLocalCallContext(cc, getNodeEnclosingCallable(midnode)) and
+    localCC = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(midnode)) and
     ap0 = mid.getAp()
   |
     localFlowBigStep(midnode, node, true, _, conf, localCC) and

cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/DataFlowImpl2.qll

@@ -2133,11 +2133,8 @@ private module Stage4 {
 
   bindingset[node, cc, config]
   private LocalCc getLocalCc(Node node, Cc cc, Configuration config) {
-    exists(Cc cc0 |
-      cc = pragma[only_bind_into](cc0) and
-      localFlowEntry(node, config) and
-      result = getLocalCallContext(cc0, getNodeEnclosingCallable(node))
-    )
+    localFlowEntry(node, config) and
+    result = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(node))
   }
 
   private predicate localStep(
@@ -3132,7 +3129,7 @@ private predicate pathStep(PathNodeMid mid, Node node, CallContext cc, SummaryCt
     conf = mid.getConfiguration() and
     cc = mid.getCallContext() and
     sc = mid.getSummaryCtx() and
-    localCC = getLocalCallContext(cc, getNodeEnclosingCallable(midnode)) and
+    localCC = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(midnode)) and
     ap0 = mid.getAp()
   |
     localFlowBigStep(midnode, node, true, _, conf, localCC) and

cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/DataFlowImpl3.qll

@@ -2133,11 +2133,8 @@ private module Stage4 {
 
   bindingset[node, cc, config]
   private LocalCc getLocalCc(Node node, Cc cc, Configuration config) {
-    exists(Cc cc0 |
-      cc = pragma[only_bind_into](cc0) and
-      localFlowEntry(node, config) and
-      result = getLocalCallContext(cc0, getNodeEnclosingCallable(node))
-    )
+    localFlowEntry(node, config) and
+    result = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(node))
   }
 
   private predicate localStep(
@@ -3132,7 +3129,7 @@ private predicate pathStep(PathNodeMid mid, Node node, CallContext cc, SummaryCt
     conf = mid.getConfiguration() and
     cc = mid.getCallContext() and
     sc = mid.getSummaryCtx() and
-    localCC = getLocalCallContext(cc, getNodeEnclosingCallable(midnode)) and
+    localCC = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(midnode)) and
     ap0 = mid.getAp()
   |
     localFlowBigStep(midnode, node, true, _, conf, localCC) and

cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/DataFlowImpl4.qll

@@ -2133,11 +2133,8 @@ private module Stage4 {
 
   bindingset[node, cc, config]
   private LocalCc getLocalCc(Node node, Cc cc, Configuration config) {
-    exists(Cc cc0 |
-      cc = pragma[only_bind_into](cc0) and
-      localFlowEntry(node, config) and
-      result = getLocalCallContext(cc0, getNodeEnclosingCallable(node))
-    )
+    localFlowEntry(node, config) and
+    result = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(node))
   }
 
   private predicate localStep(
@@ -3132,7 +3129,7 @@ private predicate pathStep(PathNodeMid mid, Node node, CallContext cc, SummaryCt
     conf = mid.getConfiguration() and
     cc = mid.getCallContext() and
     sc = mid.getSummaryCtx() and
-    localCC = getLocalCallContext(cc, getNodeEnclosingCallable(midnode)) and
+    localCC = getLocalCallContext(pragma[only_bind_out](cc), getNodeEnclosingCallable(midnode)) and
     ap0 = mid.getAp()
   |
     localFlowBigStep(midnode, node, true, _, conf, localCC) and

cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/DataFlowImplCommon.qll

@@ -31,7 +31,7 @@ predicate accessPathCostLimits(int apLimit, int tupleLimit) {
  * currently excludes read-steps, store-steps, and flow-through.
  *
  * The analysis uses non-linear recursion: When computing a flow path in or out
- * of a call, we use the results of the analysis recursively to resolve lamba
+ * of a call, we use the results of the analysis recursively to resolve lambda
  * calls. For this reason, we cannot reuse the code from `DataFlowImpl.qll` directly.
  */
 private module LambdaFlow {

cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/DataFlowPrivate.qll

@@ -557,5 +557,5 @@ predicate lambdaCreation(Node creation, LambdaCallKind kind, DataFlowCallable c)
 /** 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 lamba flow analysis. */
+/** Extra data-flow steps needed for lambda flow analysis. */
 predicate additionalLambdaFlowStep(Node nodeFrom, Node nodeTo, boolean preservesValue) { none() }

cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/DataFlowUtil.qll

@@ -362,15 +362,22 @@ private class ExplicitFieldStoreQualifierNode extends PartialDefinitionNode {
 
 /**
  * Not every store instruction generates a chi instruction that we can attach a PostUpdateNode to.
- * For instance, an update to a field of a struct containing only one field. For these cases we
- * attach the PostUpdateNode to the store instruction. There's no obvious pre update node for this case
- * (as the entire memory is updated), so `getPreUpdateNode` is implemented as `none()`.
+ * For instance, an update to a field of a struct containing only one field. Even if the store does
+ * have a chi instruction, a subsequent use of the result of the store may be linked directly to the
+ * result of the store as an inexact definition if the store totally overlaps the use. For these
+ * cases we attach the PostUpdateNode to the store instruction. There's no obvious pre update node
+ * for this case (as the entire memory is updated), so `getPreUpdateNode` is implemented as
+ * `none()`.
  */
 private class ExplicitSingleFieldStoreQualifierNode extends PartialDefinitionNode {
   override StoreInstruction instr;
 
   ExplicitSingleFieldStoreQualifierNode() {
-    not exists(ChiInstruction chi | chi.getPartial() = instr) and
+    (
+      instr.getAUse().isDefinitionInexact()
+      or
+      not exists(ChiInstruction chi | chi.getPartial() = instr)
+    ) and
     // Without this condition any store would create a `PostUpdateNode`.
     instr.getDestinationAddress() instanceof FieldAddressInstruction
   }

cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/ModelUtil.qll

@@ -9,30 +9,18 @@ private import semmle.code.cpp.ir.dataflow.DataFlow
 /**
  * Gets the instruction that goes into `input` for `call`.
  */
-DataFlow::Node callInput(CallInstruction call, FunctionInput input) {
-  // A positional argument
+Operand callInput(CallInstruction call, FunctionInput input) {
+  // An argument or qualifier
   exists(int index |
-    result.asInstruction() = call.getPositionalArgument(index) and
-    input.isParameter(index)
+    result = call.getArgumentOperand(index) and
+    input.isParameterOrQualifierAddress(index)
   )
   or
-  // A value pointed to by a positional argument
+  // A value pointed to by an argument or qualifier
   exists(ReadSideEffectInstruction read |
-    result.asOperand() = read.getSideEffectOperand() and
+    result = read.getSideEffectOperand() and
     read.getPrimaryInstruction() = call and
-    input.isParameterDeref(read.getIndex())
-  )
-  or
-  // The qualifier pointer
-  result.asInstruction() = call.getThisArgument() and
-  input.isQualifierAddress()
-  or
-  // The qualifier object
-  exists(ReadSideEffectInstruction read |
-    result.asOperand() = read.getSideEffectOperand() and
-    read.getPrimaryInstruction() = call and
-    read.getIndex() = -1 and
-    input.isQualifierObject()
+    input.isParameterDerefOrQualifierObject(read.getIndex())
   )
 }
 
@@ -44,19 +32,11 @@ Instruction callOutput(CallInstruction call, FunctionOutput output) {
   result = call and
   output.isReturnValue()
   or
-  // The side effect of a call on the value pointed to by a positional argument
+  // The side effect of a call on the value pointed to by an argument or qualifier
   exists(WriteSideEffectInstruction effect |
     result = effect and
     effect.getPrimaryInstruction() = call and
-    output.isParameterDeref(effect.getIndex())
-  )
-  or
-  // The side effect of a call on the qualifier object
-  exists(WriteSideEffectInstruction effect |
-    result = effect and
-    effect.getPrimaryInstruction() = call and
-    effect.getIndex() = -1 and
-    output.isQualifierObject()
+    output.isParameterDerefOrQualifierObject(effect.getIndex())
   )
   // TODO: return value dereference
 }

cpp/ql/src/semmle/code/cpp/ir/dataflow/internal/PrintIRLocalFlow.qll

@@ -6,34 +6,7 @@ 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
-
-/**
- * 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.
- */
-private 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
-}
+private import PrintIRUtilities
 
 /**
  * Gets the local dataflow from other nodes in the same function to this node.

cpp/ql/src/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/src/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/src/semmle/code/cpp/ir/dataflow/internal/TaintTrackingUtil.qll

@@ -21,53 +21,104 @@ predicate localTaintStep(DataFlow::Node nodeFrom, DataFlow::Node nodeTo) {
  */
 cached
 predicate localAdditionalTaintStep(DataFlow::Node nodeFrom, DataFlow::Node nodeTo) {
-  localInstructionTaintStep(nodeFrom.asInstruction(), nodeTo.asInstruction())
+  operandToInstructionTaintStep(nodeFrom.asOperand(), nodeTo.asInstruction())
   or
-  modeledTaintStep(nodeFrom, nodeTo)
+  instructionToOperandTaintStep(nodeFrom.asInstruction(), nodeTo.asOperand())
+}
+
+private predicate instructionToOperandTaintStep(Instruction fromInstr, Operand toOperand) {
+  // Propagate flow from the definition of an operand to the operand, even when the overlap is inexact.
+  // We only do this in certain cases:
+  // 1. The instruction's result must not be conflated, and
+  // 2. The instruction's result type is one the types where we expect element-to-object flow. Currently
+  // this is array types and union types. This matches the other two cases of element-to-object flow in
+  // `DefaultTaintTracking`.
+  toOperand.getAnyDef() = fromInstr and
+  not fromInstr.isResultConflated() and
+  (
+    fromInstr.getResultType() instanceof ArrayType or
+    fromInstr.getResultType() instanceof Union
+  )
+  or
+  exists(ReadSideEffectInstruction readInstr |
+    fromInstr = readInstr.getArgumentDef() and
+    toOperand = readInstr.getSideEffectOperand()
+  )
+  or
+  toOperand.(LoadOperand).getAnyDef() = fromInstr
 }
 
 /**
  * Holds if taint propagates from `nodeFrom` to `nodeTo` in exactly one local
  * (intra-procedural) step.
  */
-private predicate localInstructionTaintStep(Instruction nodeFrom, Instruction nodeTo) {
+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.
-  nodeTo.getAnOperand().getAnyDef() = nodeFrom and
+  instrTo.getAnOperand() = opFrom and
   (
-    nodeTo instanceof ArithmeticInstruction
+    instrTo instanceof ArithmeticInstruction
     or
-    nodeTo instanceof BitwiseInstruction
+    instrTo instanceof BitwiseInstruction
     or
-    nodeTo instanceof PointerArithmeticInstruction
-    or
-    nodeTo instanceof FieldAddressInstruction
+    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.
-    nodeTo instanceof CopyInstruction
+    instrTo instanceof CopyInstruction
   )
   or
-  nodeTo.(LoadInstruction).getSourceAddress() = nodeFrom
-  or
-  // Flow through partial reads of arrays and unions
-  nodeTo.(LoadInstruction).getSourceValueOperand().getAnyDef() = nodeFrom and
-  not nodeFrom.isResultConflated() and
+  // 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
   (
-    nodeFrom.getResultType() instanceof ArrayType or
-    nodeFrom.getResultType() instanceof Union
+    not instrTo instanceof FieldAddressInstruction
+    or
+    instrTo.(FieldAddressInstruction).getField().getDeclaringType() instanceof Union
   )
   or
+  instrTo.(LoadInstruction).getSourceAddressOperand() = opFrom
+  or
   // Flow from an element to an array or union that contains it.
-  nodeTo.(ChiInstruction).getPartial() = nodeFrom and
-  not nodeTo.isResultConflated() and
-  exists(Type t | nodeTo.getResultLanguageType().hasType(t, false) |
+  instrTo.(ChiInstruction).getPartialOperand() = opFrom and
+  not instrTo.isResultConflated() and
+  exists(Type t | instrTo.getResultLanguageType().hasType(t, false) |
     t instanceof Union
     or
     t instanceof ArrayType
   )
+  or
+  // Until we have flow through indirections across calls, we'll take flow out
+  // of the indirection and into the argument.
+  // When we get proper flow through indirections across calls, this code can be
+  // moved to `adjusedSink` or possibly into the `DataFlow::ExprNode` class.
+  exists(ReadSideEffectInstruction read |
+    read.getSideEffectOperand() = opFrom and
+    read.getArgumentDef() = instrTo
+  )
+  or
+  // Until we have from through indirections across calls, we'll take flow out
+  // of the parameter and into its indirection.
+  // `InitializeIndirectionInstruction` only has a single operand: the address of the
+  // value whose indirection we are initializing. When initializing an indirection of a parameter `p`,
+  // the IR looks like this:
+  // ```
+  // m1 = InitializeParameter[p] : &r1
+  // r2 = Load[p] : r2, m1
+  // m3 = InitializeIndirection[p] : &r2
+  // ```
+  // So by having flow from `r2` to `m3` we're enabling flow from `m1` to `m3`. This relies on the
+  // `LoadOperand`'s overlap being exact.
+  instrTo.(InitializeIndirectionInstruction).getAnOperand() = opFrom
+  or
+  modeledTaintStep(opFrom, instrTo)
 }
 
 /**
@@ -110,17 +161,19 @@ 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) {
+predicate modeledTaintStep(Operand nodeIn, Instruction nodeOut) {
   exists(CallInstruction call, TaintFunction func, FunctionInput modelIn, FunctionOutput modelOut |
     (
       nodeIn = callInput(call, modelIn)
       or
       exists(int n |
-        modelIn.isParameterDeref(n) and
-        nodeIn = callInput(call, any(InParameter inParam | inParam.getIndex() = n))
+        modelIn.isParameterDerefOrQualifierObject(n) and
+        if n = -1
+        then nodeIn = callInput(call, any(InQualifierObject inQualifier))
+        else nodeIn = callInput(call, any(InParameter inParam | inParam.getIndex() = n))
       )
     ) and
-    nodeOut.asInstruction() = callOutput(call, modelOut) and
+    nodeOut = callOutput(call, modelOut) and
     call.getStaticCallTarget() = func and
     func.hasTaintFlow(modelIn, modelOut)
   )
@@ -135,11 +188,29 @@ predicate modeledTaintStep(DataFlow::Node nodeIn, DataFlow::Node nodeOut) {
     int indexMid, InParameter modelMidIn, OutReturnValue modelOut
   |
     nodeIn = callInput(call, modelIn) and
-    nodeOut.asInstruction() = callOutput(call, modelOut) 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, ReadSideEffectInstruction read, Function func, FunctionInput modelIn,
+    FunctionOutput modelOut
+  |
+    read.getSideEffectOperand() = callInput(call, modelIn) and
+    read.getArgumentDef() = nodeIn.getDef() and
+    not read.getSideEffect().isResultModeled() and
+    call.getStaticCallTarget() = func and
+    (
+      func.(DataFlowFunction).hasDataFlow(modelIn, modelOut)
+      or
+      func.(TaintFunction).hasTaintFlow(modelIn, modelOut)
+    ) and
+    nodeOut = callOutput(call, modelOut)
+  )
 }

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

@@ -0,0 +1,115 @@
+/**
+ * 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
+  abstract override predicate isSource(DataFlow::Node source);
+
+  /**
+   * 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
+  abstract override predicate isSink(DataFlow::Node sink);
+
+  /** Holds if the node `node` is a taint sanitizer. */
+  predicate isSanitizer(DataFlow::Node node) { none() }
+
+  final override predicate isBarrier(DataFlow::Node node) {
+    isSanitizer(node) or
+    defaultTaintSanitizer(node)
+  }
+
+  /** Holds if taint propagation into `node` is prohibited. */
+  predicate isSanitizerIn(DataFlow::Node node) { none() }
+
+  final override predicate isBarrierIn(DataFlow::Node node) { isSanitizerIn(node) }
+
+  /** Holds if taint propagation out of `node` is prohibited. */
+  predicate isSanitizerOut(DataFlow::Node node) { none() }
+
+  final override predicate isBarrierOut(DataFlow::Node node) { isSanitizerOut(node) }
+
+  /** Holds if taint propagation through nodes guarded by `guard` is prohibited. */
+  predicate isSanitizerGuard(DataFlow::BarrierGuard guard) { none() }
+
+  final override predicate isBarrierGuard(DataFlow::BarrierGuard guard) { isSanitizerGuard(guard) }
+
+  /**
+   * Holds if the additional taint propagation step from `node1` to `node2`
+   * must be taken into account in the analysis.
+   */
+  predicate isAdditionalTaintStep(DataFlow::Node node1, DataFlow::Node node2) { none() }
+
+  final override predicate isAdditionalFlowStep(DataFlow::Node node1, DataFlow::Node node2) {
+    isAdditionalTaintStep(node1, node2) or
+    defaultAdditionalTaintStep(node1, node2)
+  }
+
+  /**
+   * 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/src/semmle/code/cpp/ir/dataflow/internal/tainttracking3/TaintTrackingParameter.qll

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

cpp/ql/src/semmle/code/cpp/ir/implementation/aliased_ssa/Instruction.qll

@@ -1645,6 +1645,19 @@ class CallInstruction extends Instruction {
    * Gets the number of arguments of the call, including the `this` pointer, if any.
    */
   final int getNumberOfArguments() { result = count(this.getAnArgumentOperand()) }
+
+  /**
+   * Holds if the result is a side effect for the argument at the specified index, or `this` if
+   * `index` is `-1`.
+   *
+   * This helper predicate makes it easy to join on both of these columns at once, avoiding
+   * pathological join orders in case the argument index should get joined first.
+   */
+  pragma[noinline]
+  final SideEffectInstruction getAParameterSideEffect(int index) {
+    this = result.getPrimaryInstruction() and
+    index = result.(IndexedInstruction).getIndex()
+  }
 }
 
 /**

cpp/ql/src/semmle/code/cpp/ir/implementation/aliased_ssa/internal/AliasAnalysis.qll

@@ -4,6 +4,71 @@ private import AliasAnalysisImports
 
 private class IntValue = Ints::IntValue;
 
+/**
+ * If `instr` is a `SideEffectInstruction`, gets the primary `CallInstruction` that caused the side
+ * effect. If `instr` is a `CallInstruction`, gets that same `CallInstruction`.
+ */
+private CallInstruction getPrimaryCall(Instruction instr) {
+  result = instr
+  or
+  result = instr.(SideEffectInstruction).getPrimaryInstruction()
+}
+
+/**
+ * Holds if `operand` serves as an input argument (or indirection) to `call`, in the position
+ * specified by `input`.
+ */
+private predicate isCallInput(
+  CallInstruction call, Operand operand, AliasModels::FunctionInput input
+) {
+  call = getPrimaryCall(operand.getUse()) and
+  (
+    exists(int index |
+      input.isParameterOrQualifierAddress(index) and
+      operand = call.getArgumentOperand(index)
+    )
+    or
+    exists(int index, ReadSideEffectInstruction read |
+      input.isParameterDerefOrQualifierObject(index) and
+      read = call.getAParameterSideEffect(index) and
+      operand = read.getSideEffectOperand()
+    )
+  )
+}
+
+/**
+ * Holds if `instr` serves as a return value or output argument indirection for `call`, in the
+ * position specified by `output`.
+ */
+private predicate isCallOutput(
+  CallInstruction call, Instruction instr, AliasModels::FunctionOutput output
+) {
+  call = getPrimaryCall(instr) and
+  (
+    output.isReturnValue() and instr = call
+    or
+    exists(int index, WriteSideEffectInstruction write |
+      output.isParameterDerefOrQualifierObject(index) and
+      write = call.getAParameterSideEffect(index) and
+      instr = write
+    )
+  )
+}
+
+/**
+ * Holds if the address in `operand` flows directly to the result of `resultInstr` due to modeled
+ * address flow through a function call.
+ */
+private predicate hasAddressFlowThroughCall(Operand operand, Instruction resultInstr) {
+  exists(
+    CallInstruction call, AliasModels::FunctionInput input, AliasModels::FunctionOutput output
+  |
+    call.getStaticCallTarget().(AliasModels::AliasFunction).hasAddressFlow(input, output) and
+    isCallInput(call, operand, input) and
+    isCallOutput(call, resultInstr, output)
+  )
+}
+
 /**
  * Holds if the operand `tag` of instruction `instr` is used in a way that does
  * not result in any address held in that operand from escaping beyond the
@@ -34,7 +99,7 @@ private predicate operandIsConsumedWithoutEscaping(Operand operand) {
 
 private predicate operandEscapesDomain(Operand operand) {
   not operandIsConsumedWithoutEscaping(operand) and
-  not operandIsPropagated(operand, _) and
+  not operandIsPropagated(operand, _, _) and
   not isArgumentForParameter(_, operand, _) and
   not isOnlyEscapesViaReturnArgument(operand) and
   not operand.getUse() instanceof ReturnValueInstruction and
@@ -69,67 +134,67 @@ IntValue getPointerBitOffset(PointerOffsetInstruction instr) {
 }
 
 /**
- * Holds if any address held in operand `tag` of instruction `instr` is
- * propagated to the result of `instr`, offset by the number of bits in
- * `bitOffset`. If the address is propagated, but the offset is not known to be
- * a constant, then `bitOffset` is unknown.
+ * Holds if any address held in operand `operand` is propagated to the result of `instr`, offset by
+ * the number of bits in `bitOffset`. If the address is propagated, but the offset is not known to
+ * be a constant, then `bitOffset` is `unknown()`.
  */
-private predicate operandIsPropagated(Operand operand, IntValue bitOffset) {
-  exists(Instruction instr |
-    instr = operand.getUse() and
-    (
-      // Converting to a non-virtual base class adds the offset of the base class.
-      exists(ConvertToNonVirtualBaseInstruction convert |
-        convert = instr and
-        bitOffset = Ints::mul(convert.getDerivation().getByteOffset(), 8)
-      )
-      or
-      // Conversion using dynamic_cast results in an unknown offset
-      instr instanceof CheckedConvertOrNullInstruction and
-      bitOffset = Ints::unknown()
-      or
-      // Converting to a derived class subtracts the offset of the base class.
-      exists(ConvertToDerivedInstruction convert |
-        convert = instr and
-        bitOffset = Ints::neg(Ints::mul(convert.getDerivation().getByteOffset(), 8))
-      )
-      or
-      // Converting to a virtual base class adds an unknown offset.
-      instr instanceof ConvertToVirtualBaseInstruction and
-      bitOffset = Ints::unknown()
-      or
-      // Conversion to another pointer type propagates the source address.
-      exists(ConvertInstruction convert, IRType resultType |
-        convert = instr and
-        resultType = convert.getResultIRType() and
-        resultType instanceof IRAddressType and
-        bitOffset = 0
-      )
-      or
-      // Adding an integer to or subtracting an integer from a pointer propagates
-      // the address with an offset.
-      exists(PointerOffsetInstruction ptrOffset |
-        ptrOffset = instr and
-        operand = ptrOffset.getLeftOperand() and
-        bitOffset = getPointerBitOffset(ptrOffset)
-      )
-      or
-      // Computing a field address from a pointer propagates the address plus the
-      // offset of the field.
-      bitOffset = Language::getFieldBitOffset(instr.(FieldAddressInstruction).getField())
-      or
-      // A copy propagates the source value.
-      operand = instr.(CopyInstruction).getSourceValueOperand() and bitOffset = 0
-      or
-      // Some functions are known to propagate an argument
-      isAlwaysReturnedArgument(operand) and bitOffset = 0
+private predicate operandIsPropagated(Operand operand, IntValue bitOffset, Instruction instr) {
+  // Some functions are known to propagate an argument
+  hasAddressFlowThroughCall(operand, instr) and
+  bitOffset = 0
+  or
+  instr = operand.getUse() and
+  (
+    // Converting to a non-virtual base class adds the offset of the base class.
+    exists(ConvertToNonVirtualBaseInstruction convert |
+      convert = instr and
+      bitOffset = Ints::mul(convert.getDerivation().getByteOffset(), 8)
     )
+    or
+    // Conversion using dynamic_cast results in an unknown offset
+    instr instanceof CheckedConvertOrNullInstruction and
+    bitOffset = Ints::unknown()
+    or
+    // Converting to a derived class subtracts the offset of the base class.
+    exists(ConvertToDerivedInstruction convert |
+      convert = instr and
+      bitOffset = Ints::neg(Ints::mul(convert.getDerivation().getByteOffset(), 8))
+    )
+    or
+    // Converting to a virtual base class adds an unknown offset.
+    instr instanceof ConvertToVirtualBaseInstruction and
+    bitOffset = Ints::unknown()
+    or
+    // Conversion to another pointer type propagates the source address.
+    exists(ConvertInstruction convert, IRType resultType |
+      convert = instr and
+      resultType = convert.getResultIRType() and
+      resultType instanceof IRAddressType and
+      bitOffset = 0
+    )
+    or
+    // Adding an integer to or subtracting an integer from a pointer propagates
+    // the address with an offset.
+    exists(PointerOffsetInstruction ptrOffset |
+      ptrOffset = instr and
+      operand = ptrOffset.getLeftOperand() and
+      bitOffset = getPointerBitOffset(ptrOffset)
+    )
+    or
+    // Computing a field address from a pointer propagates the address plus the
+    // offset of the field.
+    bitOffset = Language::getFieldBitOffset(instr.(FieldAddressInstruction).getField())
+    or
+    // A copy propagates the source value.
+    operand = instr.(CopyInstruction).getSourceValueOperand() and bitOffset = 0
   )
 }
 
 private predicate operandEscapesNonReturn(Operand operand) {
-  // The address is propagated to the result of the instruction, and that result itself is returned
-  operandIsPropagated(operand, _) and resultEscapesNonReturn(operand.getUse())
+  exists(Instruction instr |
+    // The address is propagated to the result of the instruction, and that result itself is returned
+    operandIsPropagated(operand, _, instr) and resultEscapesNonReturn(instr)
+  )
   or
   // The operand is used in a function call which returns it, and the return value is then returned
   exists(CallInstruction ci, Instruction init |
@@ -151,9 +216,11 @@ private predicate operandEscapesNonReturn(Operand operand) {
 }
 
 private predicate operandMayReachReturn(Operand operand) {
-  // The address is propagated to the result of the instruction, and that result itself is returned
-  operandIsPropagated(operand, _) and
-  resultMayReachReturn(operand.getUse())
+  exists(Instruction instr |
+    // The address is propagated to the result of the instruction, and that result itself is returned
+    operandIsPropagated(operand, _, instr) and
+    resultMayReachReturn(instr)
+  )
   or
   // The operand is used in a function call which returns it, and the return value is then returned
   exists(CallInstruction ci, Instruction init |
@@ -173,9 +240,9 @@ private predicate operandMayReachReturn(Operand operand) {
 
 private predicate operandReturned(Operand operand, IntValue bitOffset) {
   // The address is propagated to the result of the instruction, and that result itself is returned
-  exists(IntValue bitOffset1, IntValue bitOffset2 |
-    operandIsPropagated(operand, bitOffset1) and
-    resultReturned(operand.getUse(), bitOffset2) and
+  exists(Instruction instr, IntValue bitOffset1, IntValue bitOffset2 |
+    operandIsPropagated(operand, bitOffset1, instr) and
+    resultReturned(instr, bitOffset2) and
     bitOffset = Ints::add(bitOffset1, bitOffset2)
   )
   or
@@ -214,13 +281,6 @@ private predicate isArgumentForParameter(
   )
 }
 
-private predicate isAlwaysReturnedArgument(Operand operand) {
-  exists(AliasModels::AliasFunction f |
-    f = operand.getUse().(CallInstruction).getStaticCallTarget() and
-    f.parameterIsAlwaysReturned(operand.(PositionalArgumentOperand).getIndex())
-  )
-}
-
 private predicate isOnlyEscapesViaReturnArgument(Operand operand) {
   exists(AliasModels::AliasFunction f |
     f = operand.getUse().(CallInstruction).getStaticCallTarget() and
@@ -270,12 +330,15 @@ predicate allocationEscapes(Configuration::Allocation allocation) {
 /**
  * Equivalent to `operandIsPropagated()`, but includes interprocedural propagation.
  */
-private predicate operandIsPropagatedIncludingByCall(Operand operand, IntValue bitOffset) {
-  operandIsPropagated(operand, bitOffset)
+private predicate operandIsPropagatedIncludingByCall(
+  Operand operand, IntValue bitOffset, Instruction instr
+) {
+  operandIsPropagated(operand, bitOffset, instr)
   or
   exists(CallInstruction call, Instruction init |
     isArgumentForParameter(call, operand, init) and
-    resultReturned(init, bitOffset)
+    resultReturned(init, bitOffset) and
+    instr = call
   )
 }
 
@@ -292,8 +355,7 @@ private predicate hasBaseAndOffset(AddressOperand addrOperand, Instruction base,
     // We already have an offset from `middle`.
     hasBaseAndOffset(addrOperand, middle, previousBitOffset) and
     // `middle` is propagated from `base`.
-    middleOperand = middle.getAnOperand() and
-    operandIsPropagatedIncludingByCall(middleOperand, additionalBitOffset) and
+    operandIsPropagatedIncludingByCall(middleOperand, additionalBitOffset, middle) and
     base = middleOperand.getDef() and
     bitOffset = Ints::add(previousBitOffset, additionalBitOffset)
   )

cpp/ql/src/semmle/code/cpp/ir/implementation/aliased_ssa/internal/AliasConfiguration.qll

@@ -105,7 +105,21 @@ class DynamicAllocation extends Allocation, TDynamicAllocation {
   DynamicAllocation() { this = TDynamicAllocation(call) }
 
   final override string toString() {
-    result = call.toString() + " at " + call.getLocation() // This isn't performant, but it's only used in test/dump code right now.
+    // This isn't performant, but it's only used in test/dump code right now.
+    // Dynamic allocations within a function are numbered in the order by start
+    // line number. This keeps them stable when the function moves within the
+    // file, or when non-allocating lines are added and removed within the
+    // function.
+    exists(int i |
+      result = "dynamic{" + i.toString() + "}" and
+      call =
+        rank[i](CallInstruction rangeCall |
+          exists(TDynamicAllocation(rangeCall)) and
+          rangeCall.getEnclosingIRFunction() = call.getEnclosingIRFunction()
+        |
+          rangeCall order by rangeCall.getLocation().getStartLine()
+        )
+    )
   }
 
   final override CallInstruction getABaseInstruction() { result = call }

cpp/ql/src/semmle/code/cpp/ir/implementation/raw/Instruction.qll

@@ -1645,6 +1645,19 @@ class CallInstruction extends Instruction {
    * Gets the number of arguments of the call, including the `this` pointer, if any.
    */
   final int getNumberOfArguments() { result = count(this.getAnArgumentOperand()) }
+
+  /**
+   * Holds if the result is a side effect for the argument at the specified index, or `this` if
+   * `index` is `-1`.
+   *
+   * This helper predicate makes it easy to join on both of these columns at once, avoiding
+   * pathological join orders in case the argument index should get joined first.
+   */
+  pragma[noinline]
+  final SideEffectInstruction getAParameterSideEffect(int index) {
+    this = result.getPrimaryInstruction() and
+    index = result.(IndexedInstruction).getIndex()
+  }
 }
 
 /**

cpp/ql/src/semmle/code/cpp/ir/implementation/raw/internal/SideEffects.qll

@@ -0,0 +1,109 @@
+/**
+ * Predicates to compute the modeled side effects of calls during IR construction.
+ *
+ * These are used in `TranslatedElement.qll` to generate the `TTranslatedSideEffect` instances, and
+ * also in `TranslatedCall.qll` to inject the actual side effect instructions.
+ */
+
+private import cpp
+private import semmle.code.cpp.ir.implementation.Opcode
+private import semmle.code.cpp.models.interfaces.SideEffect
+
+/**
+ * Holds if the specified call has a side effect that does not come from a `SideEffectFunction`
+ * model.
+ */
+private predicate hasDefaultSideEffect(Call call, ParameterIndex i, boolean buffer, boolean isWrite) {
+  not call.getTarget() instanceof SideEffectFunction and
+  (
+    exists(MemberFunction mfunc |
+      // A non-static member function, including a constructor or destructor, may write to `*this`,
+      // and may also read from `*this` if it is not a constructor.
+      i = -1 and
+      mfunc = call.getTarget() and
+      not mfunc.isStatic() and
+      buffer = false and
+      (
+        isWrite = false and not mfunc instanceof Constructor
+        or
+        isWrite = true and not mfunc instanceof ConstMemberFunction
+      )
+    )
+    or
+    exists(Expr expr |
+      // A pointer-like argument is assumed to read from the pointed-to buffer, and may write to the
+      // buffer as well unless the pointer points to a `const` value.
+      i >= 0 and
+      buffer = true and
+      expr = call.getArgument(i).getFullyConverted() and
+      exists(Type t | t = expr.getUnspecifiedType() |
+        t instanceof ArrayType or
+        t instanceof PointerType or
+        t instanceof ReferenceType
+      ) and
+      (
+        isWrite = true and
+        not call.getTarget().getParameter(i).getType().isDeeplyConstBelow()
+        or
+        isWrite = false
+      )
+    )
+  )
+}
+
+/**
+ * Returns a side effect opcode for parameter index `i` of the specified call.
+ *
+ * This predicate will return at most two results: one read side effect, and one write side effect.
+ */
+Opcode getASideEffectOpcode(Call call, ParameterIndex i) {
+  exists(boolean buffer |
+    (
+      call.getTarget().(SideEffectFunction).hasSpecificReadSideEffect(i, buffer)
+      or
+      not call.getTarget() instanceof SideEffectFunction and
+      hasDefaultSideEffect(call, i, buffer, false)
+    ) and
+    if exists(call.getTarget().(SideEffectFunction).getParameterSizeIndex(i))
+    then (
+      buffer = true and
+      result instanceof Opcode::SizedBufferReadSideEffect
+    ) else (
+      buffer = false and result instanceof Opcode::IndirectReadSideEffect
+      or
+      buffer = true and result instanceof Opcode::BufferReadSideEffect
+    )
+  )
+  or
+  exists(boolean buffer, boolean mustWrite |
+    (
+      call.getTarget().(SideEffectFunction).hasSpecificWriteSideEffect(i, buffer, mustWrite)
+      or
+      not call.getTarget() instanceof SideEffectFunction and
+      hasDefaultSideEffect(call, i, buffer, true) and
+      mustWrite = false
+    ) and
+    if exists(call.getTarget().(SideEffectFunction).getParameterSizeIndex(i))
+    then (
+      buffer = true and
+      mustWrite = false and
+      result instanceof Opcode::SizedBufferMayWriteSideEffect
+      or
+      buffer = true and
+      mustWrite = true and
+      result instanceof Opcode::SizedBufferMustWriteSideEffect
+    ) else (
+      buffer = false and
+      mustWrite = false and
+      result instanceof Opcode::IndirectMayWriteSideEffect
+      or
+      buffer = false and
+      mustWrite = true and
+      result instanceof Opcode::IndirectMustWriteSideEffect
+      or
+      buffer = true and mustWrite = false and result instanceof Opcode::BufferMayWriteSideEffect
+      or
+      buffer = true and mustWrite = true and result instanceof Opcode::BufferMustWriteSideEffect
+    )
+  )
+}

cpp/ql/src/semmle/code/cpp/ir/implementation/raw/internal/TranslatedCall.qll

@@ -4,6 +4,7 @@ private import semmle.code.cpp.ir.implementation.internal.OperandTag
 private import semmle.code.cpp.ir.internal.CppType
 private import semmle.code.cpp.models.interfaces.SideEffect
 private import InstructionTag
+private import SideEffects
 private import TranslatedElement
 private import TranslatedExpr
 private import TranslatedFunction
@@ -424,12 +425,15 @@ class TranslatedCallSideEffects extends TranslatedSideEffects, TTranslatedCallSi
 }
 
 class TranslatedStructorCallSideEffects extends TranslatedCallSideEffects {
-  TranslatedStructorCallSideEffects() { getParent().(TranslatedStructorCall).hasQualifier() }
+  TranslatedStructorCallSideEffects() {
+    getParent().(TranslatedStructorCall).hasQualifier() and
+    getASideEffectOpcode(expr, -1) instanceof WriteSideEffectOpcode
+  }
 
   override predicate hasInstruction(Opcode opcode, InstructionTag tag, CppType t) {
-    opcode instanceof Opcode::IndirectMayWriteSideEffect and
     tag instanceof OnlyInstructionTag and
-    t = getTypeForPRValue(expr.getTarget().getDeclaringType())
+    t = getTypeForPRValue(expr.getTarget().getDeclaringType()) and
+    opcode = getASideEffectOpcode(expr, -1).(WriteSideEffectOpcode)
   }
 
   override Instruction getInstructionSuccessor(InstructionTag tag, EdgeKind kind) {
@@ -460,9 +464,11 @@ class TranslatedSideEffect extends TranslatedElement, TTranslatedArgumentSideEff
   Call call;
   Expr arg;
   int index;
-  boolean write;
+  SideEffectOpcode sideEffectOpcode;
 
-  TranslatedSideEffect() { this = TTranslatedArgumentSideEffect(call, arg, index, write) }
+  TranslatedSideEffect() {
+    this = TTranslatedArgumentSideEffect(call, arg, index, sideEffectOpcode)
+  }
 
   override Locatable getAST() { result = arg }
 
@@ -472,13 +478,13 @@ class TranslatedSideEffect extends TranslatedElement, TTranslatedArgumentSideEff
 
   int getArgumentIndex() { result = index }
 
-  predicate isWrite() { write = true }
+  predicate isWrite() { sideEffectOpcode instanceof WriteSideEffectOpcode }
 
   override string toString() {
-    write = true and
+    isWrite() and
     result = "(write side effect for " + arg.toString() + ")"
     or
-    write = false and
+    not isWrite() and
     result = "(read side effect for " + arg.toString() + ")"
   }
 
@@ -489,29 +495,29 @@ class TranslatedSideEffect extends TranslatedElement, TTranslatedArgumentSideEff
   override Instruction getFirstInstruction() { result = getInstruction(OnlyInstructionTag()) }
 
   override predicate hasInstruction(Opcode opcode, InstructionTag tag, CppType type) {
-    isWrite() and
-    hasSpecificWriteSideEffect(opcode) and
     tag = OnlyInstructionTag() and
+    opcode = sideEffectOpcode and
     (
-      opcode instanceof BufferAccessOpcode and
-      type = getUnknownType()
-      or
-      not opcode instanceof BufferAccessOpcode and
-      exists(Type baseType | baseType = arg.getUnspecifiedType().(DerivedType).getBaseType() |
-        if baseType instanceof VoidType
-        then type = getUnknownType()
-        else type = getTypeForPRValueOrUnknown(baseType)
+      isWrite() and
+      (
+        opcode instanceof BufferAccessOpcode and
+        type = getUnknownType()
+        or
+        not opcode instanceof BufferAccessOpcode and
+        exists(Type baseType | baseType = arg.getUnspecifiedType().(DerivedType).getBaseType() |
+          if baseType instanceof VoidType
+          then type = getUnknownType()
+          else type = getTypeForPRValueOrUnknown(baseType)
+        )
+        or
+        index = -1 and
+        not arg.getUnspecifiedType() instanceof DerivedType and
+        type = getTypeForPRValueOrUnknown(arg.getUnspecifiedType())
       )
       or
-      index = -1 and
-      not arg.getUnspecifiedType() instanceof DerivedType and
-      type = getTypeForPRValueOrUnknown(arg.getUnspecifiedType())
+      not isWrite() and
+      type = getVoidType()
     )
-    or
-    not isWrite() and
-    hasSpecificReadSideEffect(opcode) and
-    tag = OnlyInstructionTag() and
-    type = getVoidType()
   }
 
   override Instruction getInstructionSuccessor(InstructionTag tag, EdgeKind kind) {
@@ -535,7 +541,7 @@ class TranslatedSideEffect extends TranslatedElement, TTranslatedArgumentSideEff
 
   override CppType getInstructionMemoryOperandType(InstructionTag tag, TypedOperandTag operandTag) {
     not isWrite() and
-    if hasSpecificReadSideEffect(any(BufferAccessOpcode op))
+    if sideEffectOpcode instanceof BufferAccessOpcode
     then
       result = getUnknownType() and
       tag instanceof OnlyInstructionTag and
@@ -557,56 +563,6 @@ class TranslatedSideEffect extends TranslatedElement, TTranslatedArgumentSideEff
       )
   }
 
-  predicate hasSpecificWriteSideEffect(Opcode op) {
-    exists(boolean buffer, boolean mustWrite |
-      if exists(call.getTarget().(SideEffectFunction).getParameterSizeIndex(index))
-      then
-        call.getTarget().(SideEffectFunction).hasSpecificWriteSideEffect(index, true, mustWrite) and
-        buffer = true and
-        (
-          mustWrite = false and op instanceof Opcode::SizedBufferMayWriteSideEffect
-          or
-          mustWrite = true and op instanceof Opcode::SizedBufferMustWriteSideEffect
-        )
-      else (
-        call.getTarget().(SideEffectFunction).hasSpecificWriteSideEffect(index, buffer, mustWrite) and
-        (
-          buffer = true and mustWrite = false and op instanceof Opcode::BufferMayWriteSideEffect
-          or
-          buffer = false and mustWrite = false and op instanceof Opcode::IndirectMayWriteSideEffect
-          or
-          buffer = true and mustWrite = true and op instanceof Opcode::BufferMustWriteSideEffect
-          or
-          buffer = false and mustWrite = true and op instanceof Opcode::IndirectMustWriteSideEffect
-        )
-      )
-    )
-    or
-    not call.getTarget() instanceof SideEffectFunction and
-    getArgumentIndex() != -1 and
-    op instanceof Opcode::BufferMayWriteSideEffect
-    or
-    not call.getTarget() instanceof SideEffectFunction and
-    getArgumentIndex() = -1 and
-    op instanceof Opcode::IndirectMayWriteSideEffect
-  }
-
-  predicate hasSpecificReadSideEffect(Opcode op) {
-    exists(boolean buffer |
-      call.getTarget().(SideEffectFunction).hasSpecificReadSideEffect(index, buffer) and
-      if exists(call.getTarget().(SideEffectFunction).getParameterSizeIndex(index))
-      then buffer = true and op instanceof Opcode::SizedBufferReadSideEffect
-      else (
-        buffer = true and op instanceof Opcode::BufferReadSideEffect
-        or
-        buffer = false and op instanceof Opcode::IndirectReadSideEffect
-      )
-    )
-    or
-    not call.getTarget() instanceof SideEffectFunction and
-    op instanceof Opcode::BufferReadSideEffect
-  }
-
   override Instruction getPrimaryInstructionForSideEffect(InstructionTag tag) {
     tag = OnlyInstructionTag() and
     result = getTranslatedCallInstruction(call)

cpp/ql/src/semmle/code/cpp/ir/implementation/raw/internal/TranslatedElement.qll

@@ -12,6 +12,7 @@ private import TranslatedStmt
 private import TranslatedExpr
 private import IRConstruction
 private import semmle.code.cpp.models.interfaces.SideEffect
+private import SideEffects
 
 /**
  * Gets the "real" parent of `expr`. This predicate treats conversions as if
@@ -41,7 +42,8 @@ IRTempVariable getIRTempVariable(Locatable ast, TempVariableTag tag) {
  */
 predicate isIRConstant(Expr expr) { exists(expr.getValue()) }
 
-// Pulled out to work around QL-796
+// Pulled out for performance. See
+// https://github.com/github/codeql-coreql-team/issues/1044.
 private predicate isOrphan(Expr expr) { not exists(getRealParent(expr)) }
 
 /**
@@ -635,46 +637,15 @@ newtype TTranslatedElement =
   // The side effects of an allocation, i.e. `new`, `new[]` or `malloc`
   TTranslatedAllocationSideEffects(AllocationExpr expr) { not ignoreExpr(expr) } or
   // A precise side effect of an argument to a `Call`
-  TTranslatedArgumentSideEffect(Call call, Expr expr, int n, boolean isWrite) {
-    (
-      expr = call.getArgument(n).getFullyConverted()
-      or
-      expr = call.getQualifier().getFullyConverted() and
-      n = -1 and
-      // Exclude calls to static member functions. They don't modify the qualifier
-      not exists(MemberFunction func | func = call.getTarget() and func.isStatic())
-    ) and
-    (
-      call.getTarget().(SideEffectFunction).hasSpecificReadSideEffect(n, _) and
-      isWrite = false
-      or
-      call.getTarget().(SideEffectFunction).hasSpecificWriteSideEffect(n, _, _) and
-      isWrite = true
-      or
-      not call.getTarget() instanceof SideEffectFunction and
-      exists(Type t | t = expr.getUnspecifiedType() |
-        t instanceof ArrayType or
-        t instanceof PointerType or
-        t instanceof ReferenceType
-      ) and
-      (
-        isWrite = true and
-        not call.getTarget().getParameter(n).getType().isDeeplyConstBelow()
-        or
-        isWrite = false
-      )
-      or
-      not call.getTarget() instanceof SideEffectFunction and
-      n = -1 and
-      (
-        isWrite = true and
-        not call.getTarget() instanceof ConstMemberFunction
-        or
-        isWrite = false
-      )
-    ) and
+  TTranslatedArgumentSideEffect(Call call, Expr expr, int n, SideEffectOpcode opcode) {
     not ignoreExpr(expr) and
-    not ignoreExpr(call)
+    not ignoreExpr(call) and
+    (
+      n >= 0 and expr = call.getArgument(n).getFullyConverted()
+      or
+      n = -1 and expr = call.getQualifier().getFullyConverted()
+    ) and
+    opcode = getASideEffectOpcode(call, n)
   }
 
 /**