Summary:
Fixes T5097. When linting a large list of paths (e.g., with `--everything`), do this internally:
$chunks = array_chunk($paths, 32);
foreach ($chunks as $chunk) {
$this->lintChunk($chunk);
}
This keeps the advantages of parallelism and artifact sharing for future-based linters, without having memory usage grow in an unbounded way.
These callbacks changed:
- `willLintPath()`: Useless, no meaningful implementations. Internalized the required side effect and broke the hook.
- `didRunLinters()`: Now useless, with no meaningful implementations. Broke the hook.
- `didLintPaths()`: New hook which executes opposite `willLintPaths()`.
- `lintPath()`: Linters no longer need to implement this method.
XHPAST now has an explicit way to release shared futures.
These minor changes also happened:
- Formalized the "linter ID", which is a semi-durable identifier for the cache.
- Removed linter -> exception explicit mapping, which was unused. We now just collect exceptions.
- We do the `canRun()` checks first (and separately) now.
- Share more service call profiling code.
- Fix an issue where the test harness would use the path on disk, even if configuration set a different path.
Test Plan:
- Ran `arc lint --everything` in `arcanist/`.
- With no chunking, saw **unstable** memory usage with a peak at 941 MB.
- With chunk size 32, saw **stable** memory usage with a peak at 269 MB.
- With chunk size 8, saw **stable** memory usage with a peak at 180 MB.
- Ran with `--trace` and saw profiling information.
- Created this diff.
Reviewers: joshuaspence
Reviewed By: joshuaspence
Subscribers: epriestley
Maniphest Tasks: T5097
Differential Revision: https://secure.phabricator.com/D12501
Summary: To me, it seems that these methods should never be overwritten in subclasses.
Test Plan: `arc unit`
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: aran, epriestley, Korvin
Differential Revision: https://secure.phabricator.com/D7958
