https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101765
--- Comment #3 from Iain Sandoe <iains at gcc dot gnu.org> --- (In reply to Kacper Słomiński from comment #2) > (In reply to Iain Sandoe from comment #1) > > I am not sure that a VLA can be used in a coroutine (neither can alloca, if > > I remember correctly) [so not sure that this is ICE on valid, it could be > > ICE on invalid] > > > > Either way, we should not ICE from it. > > Using alloca or __builtin_alloca inside of coroutines compiles fine, That's probably an error ;-) ... I recall discussing this with Gor after Jakub raised the question at a GNU cauldron - with the outcome that the original implementation (MSVC) errors-out on alloca. > but > despite that I think this should be ICE-on-invalid, yeah, a VLA would mean a (potentially) different frame layout for each call - which the current implementations do not support. > as I don't see how stack > allocations would work inside of coroutines, since AFAIK the coro frame size > must be known up-front. It must be a compile-time constant, known before lowering to IR - but we should not ICE in any case. ( I have not had a chance yet to process this further )
