Revert #63649 - "Upgrade Emscripten targets to use upstream LLVM backend" #65151
Conversation
|
Self-approving since this is a revert to unblock bors, and no one seems to be around. |
|
📌 Commit d16b7f7 has been approved by |
bors
added
the
S-waiting-on-bors
…r=tmandry Revert rust-lang#63649 - "Upgrade Emscripten targets to use upstream LLVM backend" This change caused the runtime of the linux-asmjs builder to nearly double from 2+ hours to about 4 hours, which happens to be the bors timeout. (It made it in barely under 4 hours when it was merged.) This is causing timeouts on all new changes. This reverts commit 7870050, reversing changes made to 2e72448.
Rollup of 18 pull requests This contains changes from all the successful runs that bors marked as timed out, plus a revert of #63649 which appears to be the immediate cause of the timeouts. Successful merges: - #64708 (Stabilize `Option::as_deref` and `Option::as_deref_mut`) - #64728 (Stabilize UdpSocket::peer_addr) - #64765 (std: Reduce checks for `feature = "backtrace"`) - #64909 (When encountering chained operators use heuristics to recover from bad turbofish) - #65011 (Do not ICE when dereferencing non-Copy raw pointer) - #65064 (permit asyncawait-ondeck to be added by anyone) - #65066 ([const-prop] Fix ICE when trying to eval polymorphic promoted MIR) - #65100 (Replace GeneratorSubsts with SubstsRef) - #65105 (Split out some passes from librustc) - #65106 (Allow unused attributes to avoid incremental bug) - #65113 (Fix lonely backtick) - #65116 (Remove unneeded visit_statement definition) - #65118 (Update the documented default of -Z mutable-noalias) - #65123 (Account for macro invocation in `let mut $pat` diagnostic.) - #65124 (Replace some instances of `as *[const | mut] _` with `.cast()`) - #65126 (Fix typo on `now()` comments) - #65130 (lint: extern non-exhaustive types are improper) - #65151 (Revert #63649 - "Upgrade Emscripten targets to use upstream LLVM backend") Failed merges: r? @ghost
|
Analysis here looks spot on and agree with reverting while this is figured out. This shows a 1.5 he jump in testing time for the asmjs builder |
|
@alexcrichton I did some profiling and the bulk of the Emscripten link time for the tests is spent in the JS optimizer, optimizing the output of wasm2js. This time can be eliminated or reduced by changing the optimization level Emscripten uses when running the tests. Another option is to replace the asmjs-unknown-emscripten testing with wasm32-unknown-emscripten testing, which would skip the wasm2js and JS optimizing steps entirely. I took some measurements on a sample of 489 tests with various optimization options. Based on this analysis I think it would be best to just switch over to testing wasm32-unknown-emscripten or to switch to using -O1 with asmjs. What do you think would be best? |
|
@tlively I think it should be fine to switch to wasm32-unknown-emscripten and testing with -O1, that sounds like a good sweet spot! |
|
@alexcrichton, rust-lang/cargo#7476 was pulled into the cargo submodule in #65186, but when I test locally it uses the stage0 cargo which does not include that fix. Will the CI use the in-tree cargo to run tests or the old stage0 cargo? If the latter, when can I expect the stage0 cargo to be updated to include rust-lang/cargo#7476? If it will be a while, we probably want to stick with testing asmjs with -O1 for now. |
|
@tlively we always use the beta channel Cargo on the master branch's CI for tests, but if we temporarily ignore testing on wasm/emscripten for now to reenable later that's ok too! |
This change caused the runtime of the linux-asmjs builder to nearly double from 2+ hours to about 4 hours, which happens to be the bors timeout. (It made it in barely under 4 hours when it was merged.) This is causing timeouts on all new changes.
This reverts commit 7870050, reversing
changes made to 2e72448.