GitLab

We found that erasing from the weak container worklist caused timeouts
in some blink_perf benchmarks. The erase operation was meant to avoid
retracing the same backing store multiple times (an iterator to a hash
table would result in at least 2 on-stack references to the backing
store). Instead of erasing, this CL pushes retraced backing stores to
a vector. Backing stores are retraced only if they are not yet in the
vector. Thus iterators should trigger only a single retracing of the
backing store.
The vector is limited in size to avoid making it uncontrollably large.
That means it is possible to retrace the same backing store multiple
times if it was already ejected from the backing store. However this is
very unlikely in practice.

Drive-by: re-enable skipped tests.

Bug: 1142315
Change-Id: I49bd30263db46e66293d046b85699826298aa033
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2516363Reviewed-by: Kentaro Hara <haraken@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Commit-Queue: Omer Katz <omerkatz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#824069}