[PATCH] D49403: More aggressively complete RecordTypes with Function Pointers

[Andy Kaylor via Phabricator via cfe-commits]

andrew.w.kaylor added a comment.

Fair enough.

FYI, I've spent most of the day poking at the IR linker and I've found all sorts of ways that I can get it to make a complete mess of structure types and pointers to them, including some simple cases where it will mess it up in different ways depending on the order in which you link files, but I think I've convinced myself that there is no way to get the linker to cause incorrect code to be generated -- just lots of extra types, pointer casts, and function casts. This seems to be entirely consistent with what you are saying and sounds like a pretty solid argument for getting rid of typed pointers.

I guess maybe I'll take a step back and think about whether or not I could solve my current problems by pretending that all pointers are i8* and reasoning from there based on uses.

Thanks for taking the time to share your thoughts on this with me.


https://reviews.llvm.org/D49403