Skip to content

Fix #15219 FP constVariablePointer (pointer assignment with offset) - #9033

Open
chrchr-github wants to merge 5 commits into
cppcheck-opensource:mainfrom
chrchr-github:chr_15219
Open

chrchr-github wants to merge 5 commits into
cppcheck-opensource:mainfrom
chrchr-github:chr_15219

Conversation

@chrchr-github

Copy link
Copy Markdown
Collaborator

No description provided.

@dmcppcheck

Copy link
Copy Markdown
Contributor

Test results for commit 77ceb53 (tools/test-my-pr.py, main compared to this PR):

Test: http://ec2-16-170-140-253.eu-north-1.compute.amazonaws.com/pr-9033/
packages: 100
Differing warnings: 23 (+0 -23)
AI review: 12 acceptable regression, 11 improvement (http://ec2-16-170-140-253.eu-north-1.compute.amazonaws.com/pr-9033/2026-11-16_06-59-34-77ceb53a9ebc-review.txt)
Timing: +0.1% (http://ec2-16-170-140-253.eu-north-1.compute.amazonaws.com/pr-9033/2026-11-16_06-59-34-77ceb53a9ebc_timing.html)

Posted automatically by the cppcheck PR test runner. +N: warnings only with this PR, -N: warnings only with main. The AI review is written by Claude and can be wrong.

@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

The FP from the ticket is fixed, and so are the related FPs for writes through p - 1 (*(p - 1) = 0;, int* q = p - 1; *q = 0;). But the PR adds new false negatives. A pointer subtraction that is only read is no longer reported: return *(p - 1);, char c = *(p - 1); and return (p - 1)[0]; all warned before and are silent now, while *(p + 1) still warns. The deref detection at the top of the loop handles + (hasIncDecPlus) but not -, so a deref of p - n now goes to the non-deref branch and marks p as non-const. The new valueType condition also drops warnings when the - has no valueType, for example p - n > 0 where n has an unknown type. Most of the 23 warnings that the PR test runner lists as removed are probably these cases.

Testing

I built cppcheck and testrunner in pr/ and cppcheck in base/. ./testrunner TestOther passes: 0 failures, 255 tests.

I compared the two builds with cppcheck -q --enable=style --template='{line}: {id} {message}' FILE.

a.cpp

int f1(int* p) { return *(p - 1); }
int f2(int* p) { return *(p + 1); }
void f3(int* p) { *(p - 1) = 0; }
int f4(int* p) { const int* q = p - 1; return *q; }
int f5(int* p) { const int* q = p + 1; return *q; }
long f6(int* p, int* q) { return p - q; }
int f7(int* p, int n) { return p[-n]; }
int g(const int*);
int f8(int* p) { return g(p - 1); }
int f9(int* p) { return g(p + 1); }
int* f10(int* s) { int* q = s - 1; return q; }
int f11(int* p) { int* q = p - 1; return *q; }
void f12(int* p) { int* q = p - 1; *q = 0; }
  • base warns on p in lines 1, 2, 4, 6, 7, 9, 10, 11, 12, 13, on q in line 6, on s in line 11, and gives constVariablePointer for q in line 12. The warnings on lines 3, 11 and 13 are false positives.
  • pr warns only on lines 2, 6 (p and q), 7, 9, 10, and gives constVariablePointer for q in line 12.
  • So the PR fixes the FPs on lines 3, 11 and 13. Line 1 (return *(p - 1);) is a new false negative, while line 2 (*(p + 1)) is still reported.

b.cpp

int f1(int* p, unknown_t n) { return *(p - n); }
int f2(int* p, unknown_t n) { int x = p - n > 0; return x; }
int f3(char* p) { char c = *(p - 1); return c; }
int f4(int* p) { return (p - 1)[0]; }
  • base gives constParameterPointer for p on all 4 lines.
  • pr gives no warning at all. These are new false negatives.

Automated review of 77ceb53 by Claude. The pull request was built and tested locally.

Comment thread lib/checkother.cpp
if (Token::Match(parent, "%oror%|%comp%|&&|?|!|-|<<|;"))
if (Token::Match(parent, "%oror%|%comp%|&&|?|!|<<|;"))
continue;
if (Token::simpleMatch(parent, "-") && parent->valueType() && parent->valueType()->pointer == 0)

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

This drops warnings when the pointer subtraction is only read. int f(int* p) { return *(p - 1); }, char c = *(p - 1); and return (p - 1)[0]; all gave constParameterPointer before this PR and give nothing now, while *(p + 1) is still reported. The deref detection above (around line 2021) handles + through hasIncDecPlus but not -, so *(p - 1) reaches this branch and marks p as non-const.

Treating - like + there (parent->str() == "+" || parent->str() == "-") should keep these warnings and still fix #15219.

Also, when the - token has no valueType (e.g. p - n > 0 with n of an unknown type), this no longer continues, so that case is a new false negative too.

@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. The FP from #15219 is fixed: int* q = s - 1; return q; and writes through p - 1 are no longer reported. The main point from the earlier review is also fixed, because *(p - 1) that is only read warns again. Two false negatives from that review are still there. (p - 1)[0] that is only read gets no warning, and neither does p - n > 0 where - has no valueType. The [ case is not specific to -: (p + 1)[0] is not reported on main either. The same holds for const int* q = p - 1;, return p - 1; from a function that returns const int*, and p - 1 == q. The PR now handles - the same way as +, which keeps these existing false negatives.

Earlier comments

# Comment By Where Status
1 Read-only deref of pointer subtraction (*(p - 1)) no longer reported AI review general ✅ Fixed
2 (p - 1)[0] read-only no longer reported AI review general ⚠️ Open (false negative)
3 p - n with no valueType (unknown n) no longer reported AI review general ⚠️ Open (false negative)
4 Deref detection ignores '-', so *(p - 1) reads mark p non-const AI review lib/checkother.cpp:2094 ✅ Fixed
5 '-' without valueType no longer continues (p - n > 0 with unknown n) AI review lib/checkother.cpp:2094 ⚠️ Open (false negative)

@chrchr-github, please fix, resolve or reply to the open comments: #2, #3, #5.
For #2, #3, #5 (false negative), it is enough to create a Trac ticket and link it here. It does not need to be fixed in this PR.

Testing

I built cppcheck and testrunner in pr/ and cppcheck in base/. ./testrunner TestOther passes with 0 failures.

I compared the two builds with cppcheck -q --enable=style --template='{line}: {id} {message}' on scratch files with about 30 functions that use p - n in different ways.

False positives on main that the PR fixes:

  • int* q = s - 1; return q;
  • int* q = p - 1; *q = 0;
  • (p - 1)[0] = 1;
  • (*(p - 1))++;
  • int* f(int* p) { return p - 1; }

Cases where both builds give the same result:

  • These still warn: return *(p - 1);, char c = *(p - 1);, g(p - 1) where g takes const int*, *(p - n) with an unknown n, and e - p.

New true positive in the PR:

  • return (p - 1)->x;

Warnings that main gives and the PR drops:

Code Same code with + on main
return (p - 1)[0]; (p + 1)[0] is not reported
int x = p - n > 0; with unknown n —
const int* q = p - 1; not reported
const int* f(int* p) { return p - 1; } not reported
return p - 1 == q; not reported

Automated review of 5db5d7f by Claude. The pull request was built and tested locally.

@chrchr-github

Copy link
Copy Markdown
Collaborator Author

@dmcppcheck

Copy link
Copy Markdown
Contributor

Test results for commit 5db5d7f (tools/test-my-pr.py, main compared to this PR):

Test: http://ec2-16-170-140-253.eu-north-1.compute.amazonaws.com/pr-9033/
packages: 100
Differing warnings: 13 (+0 -13)
AI review: 12 acceptable regression, 1 improvement (http://ec2-16-170-140-253.eu-north-1.compute.amazonaws.com/pr-9033/2026-11-17_07-00-29-5db5d7f30982-review.txt)
Timing: -0.1% (http://ec2-16-170-140-253.eu-north-1.compute.amazonaws.com/pr-9033/2026-11-17_07-00-29-5db5d7f30982_timing.html)

Posted automatically by the cppcheck PR test runner. +N: warnings only with this PR, -N: warnings only with main. The AI review is written by Claude and can be wrong.

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