|
From: Kevin K. <kev...@gm...> - 2017-05-29 23:30:38
|
Ok, thanks. I expect to get all the tests passing Real Soon Now, at which point I'll land the callframe-motion branch. It's badly needed. There seems to be a procedure size that makes LLVM fall off a cliff, and removing callframe churn is necessary. On May 29, 2017 6:50 PM, "Donal K. Fellows" < don...@ma...> wrote: > On 29/05/2017 18:23, Kevin Kenny wrote: > >> I can see where this is going astray, but not what you intended in the >> pre-callframe implementation. I'm hoping to avoid debugging it. I'll >> keep developing on your branch and plan to backport into trunk as needed. >> > > Oops. I put code in to check this, but forgot to actually stamp out the > flaw it was put in to stop. I got fed up of undetected problems due to > putting function types in where they really didn't belong (as opposed to > pointers to functions, which are fine) but didn't do a basic rerun after > putting the check in. Comes of working on a branch where I'd actually fixed > it and then deciding that this was too important to leave on the branch. :-) > > Fixed trunk. > > Donal. > |