llvmorg-github-actions[bot] wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-clang
Author: Akash Manna (akash-manna-sky)
<details>
<summary>Changes</summary>
Fixes #<!-- -->173728
An array of zero-sized elements like `struct T {} s[-sizeof(0)][0]` slipped
past Sema's size check, because that check only looks at the total size in
bytes, and zero times anything is zero. So we ended up with an array of 2^64 -
4 elements. The constant evaluator, which runs on this code on its own (e.g.
`isEvaluatable` in codegen or range checks for `-W` warnings), then tried to
default-construct it: the element count got truncated to `unsigned` and we
either ran out of memory allocating an `APValue` per element, or hit the
"bounds check failed for in-bounds index" assertion in `adjustIndex` first.
This goes back to at least Clang 3.4.
Sema now checks the element count against the same limit as the byte size, so
these arrays are rejected with the usual "array is too large" error, like GCC
does. For counts that are still allowed, the array paths in
`VisitCXXConstructExpr` and `VisitArrayInitLoopExpr` now go through the
existing `CheckArraySize` guard, the same one used for `new` and array
destruction, so evaluation just gives up instead of trying to build billions of
elements.
---
Full diff: https://github.com/llvm/llvm-project/pull/226899.diff
7 Files Affected:
- (modified) clang/docs/ReleaseNotes.md (+4)
- (modified) clang/lib/AST/ExprConstant.cpp (+4)
- (modified) clang/lib/Sema/SemaType.cpp (+8-6)
- (modified) clang/test/CodeGenCXX/stmtexpr.cpp (+5)
- (modified) clang/test/Sema/array-size-64.c (+7)
- (modified) clang/test/SemaCXX/cxx2a-constexpr-dynalloc-limits.cpp (+34)
- (modified) clang/test/SemaCXX/zero-length-arrays.cpp (+41-2)
``````````diff
The server is unavailable at this time. Please wait a few minutes before you
try again.
``````````
</details>
https://github.com/llvm/llvm-project/pull/226899
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
