Stream: t-compiler/wg-prioritization/alerts

Topic: I-prioritize #76213 llvm segfaults during bootstrap of rust…


triagebot (Sep 01 2020 at 19:09, on Zulip):

@WG-prioritization/alerts issue #76213 has been requested for prioritization.

Procedure

apiraino (Sep 02 2020 at 15:22, on Zulip):

The discussion seems to be ongoing - should be wait a bit to assign a priority label? Or something like a P-high is appropriate?

Santiago Pastorino (Sep 02 2020 at 16:08, on Zulip):

why is this happening to them?

Santiago Pastorino (Sep 02 2020 at 16:08, on Zulip):

I meant, what's the specific condition that triggers that because I guess this is not something that is happening to the rest?

matthiaskrgr (Sep 02 2020 at 17:48, on Zulip):

perhaps it's my target-cpu=native ?

matthiaskrgr (Sep 02 2020 at 17:50, on Zulip):

the llvm stack trace shows that llvm might be doing vectorization at the moment of the crash

apiraino (Sep 02 2020 at 18:46, on Zulip):

which version are you trying to build? (Not sure this information is relevant though)

Santiago Pastorino (Sep 02 2020 at 19:31, on Zulip):

matthiaskrgr said:

perhaps it's my target-cpu=native ?

but I meant, unsure if I'm not seeing something but what are you exactly running to compile? what your config.toml looks like?

matthiaskrgr (Sep 02 2020 at 21:14, on Zulip):

I added the config.toml to the ticket

matthiaskrgr (Sep 02 2020 at 21:26, on Zulip):

which version are you trying to build?
The commit hashes i tried to build are noted in the ticket

Santiago Pastorino (Sep 09 2020 at 18:44, on Zulip):

@WG-prioritization this issue is pending prioritization

Santiago Pastorino (Sep 09 2020 at 18:44, on Zulip):

if people want to give opinions are more than welcome

Santiago Pastorino (Sep 09 2020 at 18:45, on Zulip):

I'm going to check it in a bit

apiraino (Sep 09 2020 at 18:47, on Zulip):

Seems like the issue didnt get further attention since last week :thinking: should we slap a P-high on it and have it discussed tomorrow anyway?

apiraino (Sep 09 2020 at 18:47, on Zulip):

I don't have the context for an informed opinion

lcnr (Sep 09 2020 at 18:54, on Zulip):

can we already reproduce this issue?

lcnr (Sep 09 2020 at 18:55, on Zulip):

or more accurately, was someone else other than @matthiaskrgr able to reproduce this crash

apiraino (Sep 09 2020 at 19:04, on Zulip):

not that we know of. reproducing this crash involves bootstrapping rustc which I guess requires a hefty setup in place. Can the bisect team help here?

lcnr (Sep 09 2020 at 19:06, on Zulip):

will try reproducing this myself

apiraino (Sep 09 2020 at 19:10, on Zulip):

yeah, thanks!

matthiaskrgr (Sep 09 2020 at 19:16, on Zulip):

might help if you have a kaby lake cpu

lcnr (Sep 09 2020 at 19:17, on Zulip):

can you trigger this with ./x.py build --stage 2?

matthiaskrgr (Sep 09 2020 at 19:18, on Zulip):

can try in a bit

lcnr (Sep 09 2020 at 19:18, on Zulip):

or what command fails, the command mentioned in the issue depends on full paths afaict

lcnr (Sep 09 2020 at 19:47, on Zulip):
CMake Error at cmake/modules/HandleLLVMOptions.cmake:272 (message):
  Host compiler does not support '-fuse-ld=lld'

ayy

Santiago Pastorino (Sep 09 2020 at 20:23, on Zulip):

this one is pending too but it seems that we don't have a good idea yet

Joshua Nelson (Sep 09 2020 at 20:24, on Zulip):

IMO P-medium or P-low, it only affects contributors with a weird config.toml it seems?

matthiaskrgr (Sep 09 2020 at 20:39, on Zulip):

Hmm, what's weird about my config.toml? :P

Joshua Nelson (Sep 09 2020 at 20:43, on Zulip):

oh not config.toml, sorry, -C target-cpu=native

Joshua Nelson (Sep 09 2020 at 20:43, on Zulip):

hmm I wonder if it goes away without thin-lto

matthiaskrgr (Sep 09 2020 at 20:45, on Zulip):

probably, looking at the stacktrace, llvm was doing vectorization (which could explain why I need target-cpu=native to repro) during lto while it crashed

matthiaskrgr (Sep 09 2020 at 21:31, on Zulip):

@lcnr `./x.py build --stage 2``crashes as well

matthiaskrgr (Sep 09 2020 at 21:31, on Zulip):

(not sure if this differs from my default build command though...

DPC (Sep 11 2020 at 00:24, on Zulip):

i think we should mark this as P-high

triagebot (Sep 11 2020 at 13:48, on Zulip):

Issue #76213's prioritization request has been removed.

apiraino (Dec 09 2020 at 17:08, on Zulip):

Mirroring my comment for issue #79654 Are they just waiting for the LLVm bug to be resolved?

apiraino (Dec 16 2020 at 11:31, on Zulip):

I'm going to flag with I-nominate and try to summarize both #76213 and #79654 for discussion. Perhaps they will be picked up and looked at.
I'd like to also have a word from the team on how the linked LLVM issue (https://bugs.llvm.org/show_bug.cgi?id=48340) is related to these issues and have them under the radar

apiraino (Dec 16 2020 at 11:32, on Zulip):

if anyone has anything to add, I'll be happy to add comments to the agenda summary

Last update: Apr 11 2021 at 18:00UTC