Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
rustc: Work around `DICompileUnit` bugs in LLVM #46772
referenced this pull request
Dec 16, 2017
Do we have some kind of guarantee that the first
I'm a bit wary of the consequences of this patch. LLVM does not crash any more but how much will it break line information? I'm OK with merging for now but before we make ThinLTO the default for release builds we should verify that backtraces and basic profiling still work.
Also, we should really push for this being properly fixed in LLVM.
Heh I was hoping you'd know more than I on those topics :)
I'm not actually sure at all what
So given that the file doesn't exist I'd imagine that this wouldn't impact the experience too much. That being said though were there particular things you're worried about? The backtraces do indeed seem to work on OSX (don't have access to Linux right now) but this would also be pretty easy to verify after-the-fact of landing.
As far as paths go, we might be in the clear though, since all compile units from the same compilation session with have the same directory there.
Ah excellent points @michaelwoerister! Useful to know for sure that
I think in that case this may end up working out well? In debug mode this'll never happen because we only generate one codegen unit and ThinLTO isn't used, and then in optimized mode debuginfo is sort of sketchy anyway, but it'd only worsen the experience with debugging the inlined functions from other libraries I think? (which may not have been that great to begin with...)
In that sense perhaps we should merge and see what happens? It's surely easy to turn off quickly :)