https://github.com/nikic requested changes to this pull request.

Yes, we should not use alloca types for this purpose.

The correct type to use here is the [byte 
type](https://blog.llvm.org/posts/2025-08-29-gsoc-byte-type/) -- this is 
necessary for correctness, otherwise the transform may be incorrect wrt pointer 
provenance and poison propagation. We currently get away with using integers 
through a combination of conservative handling elsewhere and poison in memory 
being de-facto unsupported (in significant part due to exactly this issue).

The byte type does not exist yet (though I expect to see movement here soon), 
but I think that the change proposed here is a dead end.

(What we *could* do is inspect users of the pointer to check whether it is only 
used with a single type and use that type. Not sure whether doing that actually 
makes sense or not.)

https://github.com/llvm/llvm-project/pull/169966
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to