On Mon, Aug 17, 2020 at 11:36:49AM -0700, Nick Desaulniers wrote:
That said, this libcall optimization/transformation (sprintf->stpcpy) does look useful to me. Kees, do you have thoughts on me providing the implementation without exposing it in a header vs using -fno-builtin-stpcpy? (I would need to add the missing EXPORT_SYMBOL, as pointed out by 0day bot and on the github thread). I don't care either way; I'd just like your input before sending a V+1. Maybe better to just not implement it and never implement it?
I think I would ultimately prefer -fno-builtin-stpcpy, but for now, sure, an implementation without a header (and a biiig comment above it detailing why and a reference to the bug) would be fine by me.