зеркало из https://github.com/mozilla/gecko-dev.git
af44ad742c
This assert was wrong. The assert may fire if we resurrect the node from a different thread and insert a kid fast enough. We allow resurrecting nodes (bumping the nodes from zero to one) to avoid allocation churn. In particular, while the thread dropping the node gets to read the children (so after the fetch_sub from the refcount, but before the read() of the children), another thread could plausibly bumped the refcount back, and added a children. This is a very big edge case of course, but I'm kinda sad I hadn't realized before. Differential Revision: https://phabricator.services.mozilla.com/D63286 --HG-- extra : moz-landing-system : lando |
||
|---|---|---|
| .. | ||
| components | ||
| ports/geckolib | ||
| tests/unit | ||
| moz.build | ||
| rustfmt.toml | ||