Parallelize the generation part of webidl
Categories
(Core :: DOM: Bindings (WebIDL), task)
Tracking
()
Tracking | Status | |
---|---|---|
firefox126 | --- | fixed |
People
(Reporter: sergesanspaille, Assigned: sergesanspaille)
References
(Blocks 1 open bug)
Details
Attachments
(1 file, 1 obsolete file)
Once the sequential parsing is done, dumping individual files can be done in parallel (at least conceptually). It would be great if the code could take advantage of that property
Assignee | ||
Updated•9 months ago
|
Assignee | ||
Comment 1•9 months ago
|
||
Parallelization is process based, which implies a pickle-compatible IR,
thus the few code changes.
It is mandatory to serialize the IR only once per process for the
parallelization to be beneficial.
The parsing is still single-threaded.
Updated•8 months ago
|
Backed out for causing WebIDL py3 failures
- Backout link
- Push with failures
- Failure Log
Failure line: dom/bindings/mozwebidlcodegen/test/test_mozwebidlcodegen.py::TestWebIDLCodegenManager::test_only_rebuild_self TEST-UNEXPECTED-FAIL
Comment 5•8 months ago
|
||
bugherder |
Other than the faulty Python API issue, forking 60 processes is quite expensive on Windows and takes 40sec on my Threadripper machine. This limits on 4 processes because any more than that actually makes things slower.
(Giving 4 processes causes no difference from not spawning extra process at all, but for now I'm not disabling parallel pool here.)
Comment 7•8 months ago
|
||
Comment on attachment 9395201 [details]
Bug 1886167 - Limit to 4 processes on Windows r=peterv,sergesanspaille
Revision D206740 was moved to bug 1889842. Setting attachment 9395201 [details] to obsolete.
Description
•