@mw I believe something landed in the last few days that broke summarize; I'd appreciate a ping or so on the PRs to the measureme repo so I can update it on the perf server
I guess I can make it always update to latest
I'd just rather not
since that also seems error prone
hm actually it looks like summarize is just broken:
stderr= Checking issue-32278-big-array-of-strings v0.1.0 (/tmp/.tmpW1HBwV) thread 'main' panicked at 'assertion failed: `(left == right)` left: `"codegen_and_optimize_crate"`, right: `"codegen_crate"`', summarize/src/analysis.rs:121:17 note: Run with `RUST_BACKTRACE=1` environment variable to display a backtrace. thread 'main' panicked at 'summarize failed in "/tmp/.tmpW1HBwV/self-profile-output"; prefix is "issue_32278_big_array_of_strings-12842"', collector/src/bin/rustc-fake.rs:96:21 note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace. error: Could not compile `issue-32278-big-array-of-strings`. To learn more, run the command again with --verbose.
cc @Wesley Wiser
this is kind of painful
I guess we just won't get self profile data for some time
Hmm... that's a little concerning. I don't recall anything landing recently that I would have broken this.
to be clear I'm seeing this on every single perf build
Have not tried to reproduce locally yet, ended up just disabling self profile on perf collection for now
Most of the changes recently have been in the other tools so we should be able to rule them out
I'm 99% sure its the rust-lang/rust PR because I only updated measureme on the perf server _after_ I noticed these errors, in the hope that they'd go away
think that I have found what event causes the panic https://github.com/rust-lang/rust/pull/64840#discussion_r331756092
but do not know how to solve it by any other way then remove the trace
oh sorry, yes that rustc PR changed some of the event names. I didn't realize that summarize was relying a specific names there.
Hm, OK now I see what the problem is.
the backend actually is asynchronous there, so the weird nesting of events is correct ...
I approved @Andreas Jonson's PR (https://github.com/rust-lang/rust/pull/65137) for in order to quickly unbreak summarize (thanks again!). I'll take a closer look as part of the second event review round.
I wonder if we can prevent breakage like this in the future?
maybe we can run cargotest with self profile enabled and summarize that data? probably fairly low overhead time wise