This is an archive of the discontinued LLVM Phabricator instance.

[WebAssembly] Don't assume that strongly defined symbol are DSO-local
ClosedPublic

Authored by sbc100 on May 9 2019, 5:45 PM.

Diff Detail

Repository
rLLD LLVM Linker

Event Timeline

sbc100 created this revision.May 9 2019, 5:45 PM
Herald added a project: Restricted Project. · View Herald TranscriptMay 9 2019, 5:45 PM
sbc100 updated this revision to Diff 198962.May 9 2019, 5:46 PM

cleanup

Harbormaster completed remote builds in B31722: Diff 198962.
sbc100 edited the summary of this revision. (Show Details)May 9 2019, 5:53 PM
sbc100 added a reviewer: dschuff.
sbc100 retitled this revision from [WebAssembly] Don't assume that strongly defined functions are DSO-local to [WebAssembly] Don't assume that strongly defined symbol are DSO-local.
sbc100 updated this revision to Diff 198964.May 9 2019, 6:05 PM

Update tests

sbc100 edited the summary of this revision. (Show Details)May 9 2019, 6:06 PM
sbc100 updated this revision to Diff 198965.May 9 2019, 6:07 PM
sbc100 edited the summary of this revision. (Show Details)

Remove newline

sbc100 edited the summary of this revision. (Show Details)May 9 2019, 6:10 PM
ruiu accepted this revision.May 9 2019, 6:33 PM
ruiu added a subscriber: ruiu.

LGTM

This revision is now accepted and ready to land.May 9 2019, 6:33 PM
This revision was automatically updated to reflect the committed changes.

This may have broken CI, as it shows as one of the commits in the first broken roll here:

https://chromium-review.googlesource.com/c/emscripten-releases/+/1604139

other.test_binaryen_metadce_hello is broken there. Maybe we just need to update expectations? Looks like they rose though, which is odd?

Waterfall CI also shows wasm2.test_dylink_raii_exceptions is broken on Module.dynCall_ifdi is not a function which is maybe also related to this, but just a guess.

This may have broken CI, as it shows as one of the commits in the first broken roll here:

https://chromium-review.googlesource.com/c/emscripten-releases/+/1604139

other.test_binaryen_metadce_hello is broken there. Maybe we just need to update expectations? Looks like they rose though, which is odd?

Waterfall CI also shows wasm2.test_dylink_raii_exceptions is broken on Module.dynCall_ifdi is not a function which is maybe also related to this, but just a guess.

There are a couple of different issues right now with the waterfall.

The test_binaryen_metadce_hello failure is very strange. I've put this fix for it, but I can't see how it broke, certainly doesn't seem to be this change that did it: https://github.com/emscripten-core/emscripten/pull/8581

The test_dylink_raii_exceptions failure has a solution in binaryen: https://github.com/WebAssembly/binaryen/pull/2095. However I'm not sure that test is still failing on ToT.