Files
codeql/python/ql/test/library-tests/ControlFlow/evaluation-order/NewCfgBranchTimestamps.ql
Copilot 9cdeb8e7ee Python: add new shared-CFG-backed control flow graph
Preparatory refactor for the shared-CFG dataflow migration. Adds the
new Python CFG library additively, without changing any production
behaviour.

Library additions:

- semmle.python.controlflow.internal.AstNodeImpl — mediates between
  the Python AST and the shared codeql.controlflow.ControlFlowGraph
  signature. Wraps Python's Stmt/Expr/Scope/Pattern and adds two
  synthetic kinds of node (BlockStmt for body slots, intermediate
  nodes for multi-operand boolean expressions).

- semmle.python.controlflow.internal.Cfg — public facade
  re-exposing the same API surface as semmle/python/Flow.qll
  (ControlFlowNode, CallNode, BasicBlock, NameNode, DefinitionNode,
  CompareNode, ...), backed by the shared CFG.

- lib/printCfgNew.ql — debug/visualisation query for the new CFG.

- consistency-queries/CfgConsistency.ql — consistency query running
  the shared CFG's standard checks against Python.

Shared library:

- shared.controlflow.ControlFlowGraph — adds two defaulted
  getWhileElse / getForeachElse predicates to AstSig so Python can
  model while-else / for-else (no behavioural change for other
  languages).

Test additions:

- ControlFlow/bindings/* — annotation-driven SSA-binding tests for
  the new CFG (annassign, compound, comprehension, decorated,
  except_handler, imports, match_pattern, parameters, simple,
  type_params, walrus_starred, with_stmt, dead_under_no_raise).

- ControlFlow/store-load/* — basic store/load coverage.

- ControlFlow/evaluation-order/NewCfg*.ql — mirrors of the existing
  OldCfg evaluation-order self-validation suite, run against the
  new CFG via NewCfgImpl.qll.

- Minor extensions to existing test_if.py / test_boolean.py +
  cosmetic .expected churn on a handful of OldCfg tests.

No dataflow, SSA, or production query is migrated yet — that lands in
follow-up PRs. The new CFG library has zero callers in lib/ and src/.

Verified by:
- All lib + src + consistency-queries compile clean (367 queries).
- All 56 ControlFlow library-tests pass.
- All 474 dataflow + PointsTo library-tests + consistency tests pass.
- syntax_error/CONSISTENCY/CfgConsistency passes.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
2026-06-22 13:38:30 +00:00

81 lines
2.8 KiB
Plaintext

/**
* New-CFG version of BranchTimestamps.
*
* Checks that when a node has both a true and false successor, the
* live timestamps on one branch are marked as dead on the other.
* This ensures that boolean branches are fully annotated with dead()
* markers for the paths not taken.
*
* Limitation: the `@ t[ts, ...]` / `dead(ts)` annotation scheme can only
* model branch-dead-ness for plain boolean control flow that reconverges
* linearly after the split — i.e. `if`-with-else and `if`-expression.
* It cannot model:
*
* * loops (`while` / `for`): body timestamps repeat across iterations,
* so the loop-exit annotation can't list them as dead;
* * `match` statements: each `case` body is a syntactically distinct
* sub-tree, and the branches don't reconverge through a common
* annotation point in the timeline;
* * `try` / `with` and `raise` / `assert`: exception edges are modelled
* as true/false but flow to syntactically distinct handlers, with no
* reconvergence in the linear annotation order;
* * short-circuit `and` / `or` (`BoolExpr`): the branches reconverge at
* the BoolExpr's after-node, so timestamps on one branch are live
* downstream of the other rather than dead;
* * `if` without an `else` clause, and `if`/`elif` chains: the false
* branch reconverges with the true branch at the post-if statement
* (no-else) or fans out across multiple elif-test annotations,
* neither of which fit the binary annotation scheme.
*
* Branch nodes inside those constructs are therefore whitelisted out
* below. The check still fires (and is useful) for plain `if`/`else`
* and conditional-expression branching.
*/
import python
import TimerUtils
import NewCfgImpl
private module Utils = EvalOrderCfgUtils<NewCfg>;
private import Utils
private import Utils::CfgTests
/**
* Holds if `f` contains a construct whose branches the linear-timestamp
* annotation scheme cannot describe (see file-level comment).
*/
private predicate hasUnmodellableBranching(Function f) {
exists(AstNode bad |
bad.getScope() = f and
(
bad instanceof While
or
bad instanceof For
or
bad instanceof MatchStmt
or
bad instanceof Try
or
bad instanceof With
or
bad instanceof Raise
or
bad instanceof Assert
or
bad instanceof BoolExpr
or
bad instanceof If and
(not exists(bad.(If).getAnOrelse()) or bad.(If).isElif())
)
)
}
from TimerCfgNode node, int ts, string branch
where
missingBranchTimestamp(node, ts, branch) and
not hasUnmodellableBranching(node.getTestFunction())
select node,
"Timestamp " + ts + " on true/false branch is missing a dead() annotation on the " + branch +
" successor in $@", node.getTestFunction(), node.getTestFunction().getName()