AST importer crash with CTU analysis enabled. #133144
Labels
clang:frontend
Language frontend issues, e.g. anything involving "Sema"
crash
Prefer [crash-on-valid] or [crash-on-invalid]
Comments
EugeneZelenko
added
clang:frontend
|
@llvm/issue-subscribers-clang-frontend Author: None (earnol) I have encountered compiler crash with the following stack trace signifying error in the ASTImporter. Please look at the place with While investigating the issue i was able to create a fix which apparently addresses the issue: But it seems what it addressess is not a root cause. Stack trace: |
mizvekov
added a commit
that referenced
this issue
Mar 27, 2025
Fixes a regression introduced in #130537 and reported here #133144 This fixes a crash in ASTStructuralEquivalence where the non-null precondition for IsStructurallyEquivalent would be violated, when comparing member pointers with a dependent class. This also drive-by fixes the ast node traverser for member pointers so it doesn't traverse into the qualifier in case it's not a type, or the class declaration in case there is none. This avoids printing of `<<<NULL>>>` on the text node dumper, which is redundant. No release notes since the regression was never released. Fixes #133144
|
#133343 should fix it. However, I can't reproduce the analyzer crash as the command line invocation is system specific. Can you provide a pre-processed reproducer? |
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this issue
Mar 28, 2025
…er pointer (#133343) Fixes a regression introduced in llvm/llvm-project#130537 and reported here llvm/llvm-project#133144 This fixes a crash in ASTStructuralEquivalence where the non-null precondition for IsStructurallyEquivalent would be violated, when comparing member pointers with a dependent class. This also drive-by fixes the ast node traverser for member pointers so it doesn't traverse into the qualifier in case it's not a type, or the class declaration in case it would be equivalent to what the qualifier refers. This avoids printing of `<<<NULL>>>` on the text node dumper, which is redundant. No release notes since the regression was never released. Fixes llvm/llvm-project#133144
|
Verified the fix with revision 772173f, which includes #133343. Thank you. |
|
The ast-dump showing null values is not the same thing as the crash. It would be good to include a regression test for the analyzer. |
|
I see your point. The crash happened ONLY if serialized-binary AST has NULL values and class with NULL values got compared. I will think how such test can be constructed.
|
|
The AST dump containing NULL was not the problem. The nulls are still there, my patch changed the ast-dumper so it doesn't print those properties which are null. The patch fixed a case where the null was unexpected. However, I could see the problem, but not reproduce it, because I am not so familiar with the analyzer. If we could add a regression test for this bug, that would be appreciated. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I have encountered compiler crash with the following stack trace signifying error in the ASTImporter (reproducible with 9269aae).
The reason of the error is improper import of some AST subtrees:
Please look at the place with
<<<NULL>>>. This is what causing issue while processingclang::StructuralEquivalenceContext::CheckKindSpecificEquivalenceinvokation and having identical subtrees.While investigating the issue i was able to create a fix which apparently addresses the issue:
But it seems what it addressess is not a root cause.
Created minimal reproducer showing the issue with AST: https://gcc.godbolt.org/z/5fx54Yx6W . The isseu will be after mark
line:29:1, line:31:46.Complex explanation how i run into the issue in the first place:
The whole issue reproduction involved code checker and that it what i have done:
The actual crash will happen during the follow invocation:
Another way to reproduce the problem without CodeChecker is to call:
to get a problematic AST in human readable form in file bstrwrap.cpp.hhast and search for the line
line:496:3, line:498:47. The NULL issue will be a few lines below.Stack trace:
The text was updated successfully, but these errors were encountered: