Stream: t-compiler

Topic: PR #63909 fix nounwind logic


pnkfelix (Oct 01 2019 at 12:27, on Zulip):

@centril I discovered today that @RalfJ 's PR fixes #64655. Would you be willing to remove your rfcbot concern based on that discovery?

centril (Oct 01 2019 at 12:28, on Zulip):

I'll look at this later and see

pnkfelix (Oct 01 2019 at 12:28, on Zulip):

(I'm happy to add a commit to the PR that encodes a regression test adapted from #64655 ...)

pnkfelix (Oct 01 2019 at 12:28, on Zulip):

okay thanks

pnkfelix (Oct 02 2019 at 13:31, on Zulip):

hey @rkruppe or @nagisa , would one of you be able/willing to assist me in dissect why LLVM is behaving this way for this case? I have been unsuccessful thus far in reproducing the situation in a #![no_core] style test

pnkfelix (Oct 02 2019 at 13:32, on Zulip):

(that is, I have been trying to simplify things by moving the relevant functionality into its own crate distinct from core. but when I do so, then the bug with unwind and uncalled-drop disappears)

rkruppe (Oct 02 2019 at 14:04, on Zulip):

Can you at least reduce it to just a no_std crate + core?

pnkfelix (Oct 02 2019 at 14:05, on Zulip):

Yeah I actually got it down to something I was able to dissect

pnkfelix (Oct 02 2019 at 14:05, on Zulip):

I'll be posting in the bug soon, but I don't think I need help anymore

rkruppe (Oct 02 2019 at 14:05, on Zulip):

Ok, glad to hear that

pnkfelix (Oct 02 2019 at 14:05, on Zulip):

If you are curious, this gist holds my reduced case: https://gist.github.com/425ddc19aa968bbc36b8cb46fab7d15d

pnkfelix (Oct 02 2019 at 14:06, on Zulip):

(compiling that with -C opt-level=0 -C lto=fat shows the problem.)

rkruppe (Oct 02 2019 at 14:06, on Zulip):

nod nod that looks great

pnkfelix (Oct 02 2019 at 20:02, on Zulip):

okay I filed PR #65020 as a more targeted fix that I think everyone can get on board for (orthogonal to the Q of whether PR #63909 lands) and will be amenable for beta-backport.

Last update: Nov 22 2019 at 04:40UTC