[llvm-branch-commits] [clang] release/18.x [X86_64] fix SSE type error in vaarg (PR #86698)

[Eli Friedman via llvm-branch-commits]

efriedma-quic wrote:

Right, the policy doesn't say we can only take regression fixes.  We just need to weight the impact vs. the risk.

Looking at the latest conversation on the bug report this case is pretty clearly still broken.  It's improved in the sense that after the va_arg of the struct, subsequent va_arg calls produce the right value.  But the va_arg iteslf doesn't produce the right value (probably we aren't copying the struct correctly).  So that would be a regression for some cases.

Given that, we probably don't want to pull this into 18.x as-is.

Also, given that we're making other fixes to the surrounding code, pulling any one fix into 18.x seems risky to me.  And probably low-impact, given the testcases appear to be generated by a fuzzer.

https://github.com/llvm/llvm-project/pull/86698