Improve performance of NormalizedFilePath#3067
Conversation
156f1f0 to
17c81ec
Compare
3bb4355 to
160bbd2
Compare
e34e0db to
3fa1858
Compare
|
|
||
| # - 1.8.0.0 | ||
| # - upstream: origin/master | ||
| - upstream: origin/master |
There was a problem hiding this comment.
I think we should compare with origin/master by default, so that we can see if a PR makes the performance better or worse.
There was a problem hiding this comment.
Yeah, I disabled it temporarily when merging the benchmark suite PR because upstream/master lacked some fixes that were required to run it. Thanks for enabling it!
| # - ghcide-code-actions-bindings | ||
| # - ghcide-code-actions-fill-holes | ||
| # - ghcide-code-actions-imports-exports | ||
| # - ghcide-code-actions-type-signatures |
There was a problem hiding this comment.
Enabling all of these is very slow to run, and the generated chart contains too many lines and is too hard to read. A good default is only to enable the "All" group.
We can see the difference.
Before:
After:
…server into kokobd/os-path
This reverts commit 1c2a11d.
|
Current result:
Note that the result is not ALWAYS better. Locally, most benchmarks perform better, as you see above. But on the CI, some benchmark results are not as good as the local ones. See:
|
…server into kokobd/os-path
* upgrade lsp * modify default benchmark config * upgrade lsp * use lsp master * temp: compare benchmark with previous commit * use text in NormalizedFilePath * upgrade to lsp master * fix stack config * remove obsolete dir form ghcide.cabal * run pre-commit without args * Revert "run pre-commit without args" This reverts commit 1c2a11d. * remove unnecessary tests
Upgrade
lspto benefit from the recent optimization inNormalizedFilePath.Upstream PRs:
This PR doesn't take care of Nix, as Nix is already failing due to
hiedbandhie-bios. Maybe someone familiar with Nix could solve them all in one PR.