Repository navigation
EXC-1937 Measure the fork point without blocking Neovim - #42
Merged
Merged
Conversation
- After gh names a PR target, the re-measure runs through Git.async rather than on the main loop. - The gutter's sync measures through fork_point.get_async, and holds every buffer move until that measure lands, so no buffer moves onto the base of the branch or repository just left. - A sync replaced by another repository's still puts its own repository's buffers on the base it measured. - tests/support/spawns.lua records the processes a spec starts, and tells a blocking git call that only a measure runs.
- build_at measures through fork_point.get_async and calls back once: true when built, false with no merge base, nil when an ask for another repository replaced it. An ask for the same repository joins the build in flight and has it measure once more, so it never lands on a point measured before the ask. - The sidebar opens in the tabpage it was asked from once the fork point lands; the no-base warning still comes before any window. - A second toggle before it lands cancels the open, a toggle after a cold ]c focuses it, and ]c presses meanwhile still add up. - rows(), diff(), refresh(), restore() and a session load follow the async build. - A sidebar window closed another way no longer cancels an open made since.
- Delete fork_point.get: nothing outside the specs called it, and a blocking get left in place invites a caller to block again. - ask returns only the point; get_async takes over get's contract. - Specs await get_async through tests/support/fork_point.lua; the parity specs that only pinned get against get_async are pruned.
- The sidebar opens once its fork point lands, in the tabpage it was asked from; the no-base warning still comes before any window. - The picker waits a moment for the fork point as well as the diff. - Keys pressed while it measures act on the editor as it is, so a second toggle cancels the open it would have closed. - The settle loop's three-measure cap keeps a measure from spawning git for as long as another process keeps moving HEAD.
- The build callback names its outcome ("built", "no_base",
"replaced"); the picker answers a replaced build at once rather
than waiting it out.
- A build lands after three measures, however many asks join it or
however often HEAD moves, and a background rebuild yields to a
build in flight for another repository rather than replace it.
- A step from a closed sidebar on a kept tree says nothing about
reading; a second :Changeset diff before the build lands says the
diff is off.
- The gutter threads the repository and base it moves buffers onto
through parameters rather than swapping module state.
- The pending open is a changeset.Opening the open hands back; token
checks are named predicates rather than casts.
- Tests share Sidebar.shown and one process spy, check a measure did
run, and drop a redundant sleep; docs cover the gutter's held
moves.
- build_spec's build callbacks receive an outcome string, not a readiness flag. - review.lua's token check reads in_flight, as build.lua's twin does.
- rows() waits up to 10 s for the fork point, then up to 2 s for the diff, rather than both inside one 2 s budget: under load the measure alone can run for seconds, which left the picker answering "still reading the diff". - Specs written for the blocking measure wait long enough for an async rebuild, the no-base spec waits for its own warning rather than any note, and the kept-tree spec waits for its own open.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Part of: EXC-1937
Change Kind: perf
What changed
Four paths measured the branch's fork point with blocking git calls on Neovim's main loop: the first sidebar open, the gutter's first sync when gitsigns attaches, the re-measure after
ghnames a PR target, and the rebuild after a branch switch. All four now go through the existing asyncfork_point.get_async, and the blockingfork_point.getis deleted, so the editor never freezes while the fork point is measured. The sidebar opens once the fork point lands, tens of ms later. Keys pressed meanwhile act as typeahead used to: a second toggle cancels the open, and]cpresses still add up.Measured before the change, each git spawn costs about 5 ms whatever the repository's size. A measure makes 4–5 spawns, or about 13 on a stacked branch, and all of it blocked the editor:
maintrunkWhat to test
:Changesetopens with its header and ref, and keys typed right after it reach the file window without a hitch.]cfrom a closed sidebar opens it and steps.