Skip to content

Fix #15228 (valueFlowLifetime: subscripted array is not a decayed pointer) - #9039

Open
danmar wants to merge 1 commit into
cppcheck-opensource:mainfrom
cppchecksolutions:fix-15228
Open

danmar wants to merge 1 commit into
cppcheck-opensource:mainfrom
cppchecksolutions:fix-15228

Conversation

@danmar

@danmar danmar commented Nov 17, 2026

Copy link
Copy Markdown
Collaborator

For 'arr[i]' where arr is an array of pointers, isDecayedPointer() returned true because the type of 'arr[i]' is a pointer. A lifetime value was then created and forwarded for each subscript; with many 'arr[N] = ...' statements in one function this is quadratic.

spice rop3.c: ValueFlow 352s => 28s.
In 10000 Debian packages (--check-level=exhaustive) this adds 11 warnings: 9 correct nullPointerOutOfMemory/nullPointerOutOfResources and 2 false positives (uninitvar, arrayIndexOutOfBoundsCond).

@dmcppcheck

Copy link
Copy Markdown
Contributor

There should probably be a trac ticket for this PR. Please create it, or ask us to create it. Then the PR title should be updated.

@danmar danmar changed the title valueFlowLifetime: subscripted array is not a decayed pointer Fix #15228 (valueFlowLifetime: subscripted array is not a decayed pointer) Nov 17, 2026
For 'arr[i]' where arr is an array of pointers, isDecayedPointer()
returned true because the type of 'arr[i]' is a pointer. A lifetime
value was then created and forwarded for each subscript; with many
'arr[N] = ...' statements in one function this is quadratic.

spice rop3.c: ValueFlow 352s => 28s.
In 10000 Debian packages (--check-level=exhaustive) this adds 11
warnings: 9 correct nullPointerOutOfMemory/nullPointerOutOfResources
and 2 false positives (uninitvar, arrayIndexOutOfBoundsCond).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@danmar
danmar force-pushed the fix-15228 branch 2 times, most recently from d1ed991 to 396bb75 Compare November 17, 2026 18:24

@dmcppcheck dmcppcheck left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

AI review

I found no new problems. When the array is the left operand of [, isDecayedPointer() now returns false. That fixes the quadratic slowdown: a function with 3000 arr[i] = "..." statements took 105 s with base and 0.36 s with the PR. I compared the PR with base on many lifetime, dangling-pointer, uninitvar and multi-dimensional-array cases (a[0] of int a[2][3]) and on test/cfg, samples, externals, cli, gui and tools. The results were the same everywhere. I could not reproduce the 2 false positives that the description reports from the Debian run.

Testing

  • I built cppcheck and testrunner in both pr/ and base/.

  • The full ./testrunner -q passes: 5419 tests, 0 failures.

  • I copied the PR's test/testvalueflow.cpp into base. The new assertion fails there, at line 845 (expected 0, got 1). So the test covers the change.

  • Performance: generated void f(void){ const char *arr[3000]; arr[0]="s0"; ... arr[2999]="s2999"; } and ran cppcheck -q perf.c. It took 1m44.9s with base and 0.36s with the PR.

  • Behaviour: I wrote scratch files covering:

    • returning or storing a[0] of 2D arrays in C and C++ (std::string_view, std::vector<int*>)
    • &a[i] aliases followed by uninit reads
    • use after scope of p = a[0] and p = &a[1]
    • p = a[0] + 1
    • pointer arrays filled from malloc and then dereferenced

    Output from base and PR was identical. Some false negatives and one existing uninitvar false positive (&a[0].x) are already present in base, so the PR did not cause them.

  • Corpus comparison: ran base and PR on test/cfg and samples (exhaustive check level, --enable=warning,style --inconclusive) and on externals cli gui/*.cpp tools. There was no difference in the warnings.


Automated review of 396bb75 by Claude. The pull request was built and tested locally.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants