The reveal that advances the focused main view's selection after staging
captured the next/previous change block as a patch identity and found it
again in the re-rendered diff via SamePatchLine. SamePatchLine keys a
deletion on its old-file (index-side) line number, which staging shifts:
staging a hunk that changes the line count moves the index-side numbers of
every hunk below it. So a deletion-led "next hunk" candidate no longer
matched, and the reveal fell back to a worse candidate — often colliding
with a header or context row (SamePatchLine doesn't require a change line)
and, in hunk mode, snapping to the first change block. The visible result
was the selection jumping to an *earlier* hunk after staging.
Match the reveal's candidates by their worktree (new-file) line number,
which staging never moves, and require a change line of the same side. This
is safe here because the reveal only ever targets change blocks and
selectHunkAround expands to the whole block, so the new-file number's
ambiguity between two consecutive deletions (the reason SamePatchLine uses
the old-file number) doesn't matter.
The escape restore and the -U context-size preserve keep SamePatchLine:
they re-render the same staged/unstaged state, so no index-side shift, and
the latter deliberately anchors on context lines.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Staging or unstaging can move the acted-on side to the other half of the
focused main view — unstaging the first hunk of an only-staged file splits
it, pushing the staged remainder to the secondary pane. The reveal that
re-selects the nearest surviving change after the op then has to ride the
re-render of that other pane, not the one the user acted in.
Split RevealSelectionAfterStaging's single view into a source (where the
candidate lines are read) and a target (where they're found again and
re-selected), and get-or-create the target's buffer manager so the restore
can be installed on a pane that hasn't rendered yet. Behavior-preserving:
the sole caller passes the same view for both, as before.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Staging or unstaging a selection re-renders the focused main view's diff (the
staged lines move to the other side), which left the selection at a now-stale
view position — often off the new, shorter content, so it looked like there was
no selection at all.
Install a restore before the re-render that lands the selection on the change
nearest the one just acted on, the same way the staging view advances: the
change block after the selection, else the one before it, else the acted-on line
itself (which only survives when the whole side was staged and the view flips to
the other side). It re-selects in the current mode, so hunk staging walks hunk to
hunk. A range selection collapses to a single line first, since staging consumes
it. This rides the existing restore-by-identity machinery (as the escape restore
does), so it works over any conforming rendering.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A side-by-side rendering (patched delta in side-by-side mode) carries more than
one diff-line record per rendered row: the deletion on the left, the addition
replacing it on the right. DiffLineContent.Metadata keeps only the first payload
— enough to identify a single-column row, but it drops the right side — so a
range staged from such a view would miss half the changes.
Add a gocui accessor for all the distinct metadata payloads a row's cells carry,
and have the range collector resolve every one of them, so staging a
side-by-side row includes both sides. You can't stage just one side of a
side-by-side row; that's an accepted restriction (switch to a single-column
rendering to do it). Rows without metadata (no pager, or the buffer-parse /
hyperlink backends) keep falling back to their single resolved record, so
single-column staging is unchanged.
The full delta-side-by-side chain can't run in the integration harness (no real
pager), so it's covered by a gocui unit test for the multi-payload accessor;
end-to-end behaviour needs interactive verification with the patched delta.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Step 1 made the focused main view a staging surface for a single line. This
adds the range and hunk selection the staging view has, so you can stage (or
unstage) a range or a whole hunk in place — `v` starts a range, shift-up/down
extend one, `a` toggles hunk selection — without diving into the staging view.
The selection mode (line / range / hunk, plus sticky-range and "the user turned
hunk mode on themselves") lives on MainContext rather than the controller,
because three places need a pane's mode: the main view controller that drives
it, the focus controller that resets it when you focus the view, and togglePanel
which sets it on the *other* pane. The selected line and the range anchor stay
in the gocui view itself (its cursor and rangeSelectStartY), so the existing
native range-select rendering draws the highlight — no new highlight machinery.
Hunk bounds come from the diff-line metadata (the isChange run around the
cursor, via ChangeBlockBounds), not from a parsed patch, so they work over any
conforming rendering. The up/down keys are mode-aware: hunk mode steps hunk to
hunk, a non-sticky range collapses on a plain move, a sticky range extends.
GetOnStageFocusedMainView now takes the selected view-line range instead of one
line; the files handler collects the change lines in the range and applies a
single patch. It identifies each change line's patch line by scanning the parsed
patch and matching (file line number, deletion?) identities, rather than looking
each number up with PatchLineForLineNumber. That lookup can't tell an addition at
the very start of a hunk from the deletion above it, nor an addition from the
deletion it replaces on a modified line — both of which a range routinely spans.
As a side effect this also fixes staging a change on the first line of a file.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The focused main view already lets you point at a diff line and act on it
(dive into staging, edit it, open it in a PR). This makes it a staging surface
in its own right: press space to stage — or unstage — the selected line without
diving into the separate staging view.
To make that usable, the selection is now shown automatically whenever the main
view holds a diff, anchored on the first change line already visible (so the
view doesn't jump), rather than being toggled on demand from the middle of the
view. That frees space for staging and means there's always a line to act on.
Which contexts show a diff is marked by a new types.DiffMainViewContext, because
"shows a diff" is the right signal, not "is stageable": reflog shows a diff (its
selection drives edit/PR/navigation) but can't be staged, while a branch's log
or the status dashboard show no diff and get no selection at all.
Staging is delegated to the side panel beneath via a GetOnStageFocusedMainView
handler mirroring GetOnClickFocusedMainView, so what "stage" means stays the
panel's concern (the working tree stages/unstages; commits will later add to a
custom patch). The files handler resolves the line's patch identity from the
diff-line metadata, maps it to a patch line, and applies a one-line patch,
choosing stage vs unstage from whether the shown diff is the unstaged or staged
side (diffSplitState).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
On a position-preserving re-render the first paint swapped the off-screen
content in and only then ran Apply, which for the buffer-parse backend (no
pager) scans the whole diff to locate the line to land on. That scan takes
tens of milliseconds on a large diff, during which the new content was already
displayed at the *old* scroll position — a layout draw landing in that window
showed a frame at the stale (and now out-of-range) scroll, a pronounced
flicker when changing context size while scrolled down. The metadata/hyperlink
backends didn't show it because they resolve the target during the load, so
their Apply is instant.
Let Apply own the swap: it locates the target against the still-off-screen
(and, at end of input, complete) buffer first, then calls swapIn and settles
the scroll. The scan now runs while the previous content is still displayed, so
the new content is revealed already at the right position — matching what the
early-resolving backends already did.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Changing the -U context size or switching pagers preserves the scroll
position by scanning the re-rendered diff for the line to land on. Each scan
resolved every buffer line through the per-line resolver, whose buffer-parse
backend re-parses that line's whole file section on every call — so a scan was
O(n2) in the diff length. On a 9600-line diff, changing context took ~33s with
no pager (worse with delta); the file/hunk navigation scans had the same cost.
Route the whole-buffer scans (the position restore's nearbyDiffLines and
end-of-load resolution, and the file/hunk navigation) through a new
resolveDiffLines, which parses each file section once for the whole buffer and
applies the metadata/buffer/hyperlink precedence on top — O(n). The incremental
restore scan now resolves only the rows that loaded since it last looked, using
the per-row backends (metadata/hyperlink) that don't need surrounding context;
the buffer-parse case still resolves once the diff is complete. The single-line
resolver (clicks) is unchanged.
Measured on a synthetic single-file diff, the whole-buffer scan drops from
1.7s to 1ms at 1700 lines and from ~107s to 15ms at 6800 lines.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The diff-line buffer-parse backend (mechanism #1) resolves one line at a
time, re-parsing that line's whole file section on every call. The position-
restore and navigation scans resolve every line of the buffer, so they pay
that whole-section parse once per line — O(n2) on a large single-file diff.
Extract the per-section parse (parseFileSection) and the section-bounds
search (fileSectionBounds) out of parseDiffLineFromBuffer, and add
parseAllDiffLinesFromBuffer, which walks the file sections once and resolves
every line in a single O(n) pass. Behavior-preserving: parseDiffLineFromBuffer
now delegates to the same per-section parser, so a single-line lookup is
unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
When {/} change the -U context size with no selection showing, we re-anchor
the re-rendered diff on the top visible line. Anchor on the middle visible
line instead, so the diff appears to pivot around the line you're most
likely looking at rather than around its top edge.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The terminal-allocation audit settled the protocol's OSC number on 1717 (unused
by every surveyed terminal — see diff-line-metadata-osc-spec.md), retiring the
456 placeholder. Rename the host side to match: the gocui carrier that accumulates
and reads back the sequence, the parser, and the handshake env var the pager
subprocess is given (now EMIT_OSC1717_METADATA). Flip the design notes and spec
from 'rename pending' to done.
The delta and difftastic emitters are renamed in their own repos.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add file and change-block ("hunk") navigation to the focused main view,
mirroring the staging view's hunk keys: `<left>`/`<right>` jump to the
previous/next hunk and `n`/`N` to the next/previous file. A "hunk" here is
lazygit's notion — a run of consecutive added/deleted lines separated by
context, not a git `@@` section — matching what the staging view jumps
between.
This is a consumer of the diff-line primitive in its forward direction:
resolve each rendered row's patch identity, then scan for the next/previous
change block (by the line type) or file boundary (by the path changing). The
file scan lands on the top of the neighbouring file even when a restructuring
pager leaves the header rows untagged, by backing up over them from the
file's first identifiable row — which is impossible without the per-line
metadata once the pager stops emitting a parseable unified diff. The
boundary arithmetic is pulled out into pure functions and unit-tested.
The anchor is the selected line if a selection is showing, else the top
visible line. With a selection we move it to the target and scroll it into
view, like the staging view; with none we stay in scroll mode, bringing the
target to the top without creating a selection.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Increasing or decreasing the diff context size (the `{`/`}` keybindings)
re-renders the diff with a different `git diff -U<n>` command. Because the
command key changes, the render reset the main view to the top — losing the
spot the user was reading, which is exactly the spot the context change is
about.
Preserve it instead, reusing the identity-based restore built for the escape
path. This is its sibling consumer: capture the lines around the anchor (the
selection, or the top visible line when there's none) as restore candidates,
and after the re-render land on the nearest one that survived, put back at the
same screen row. Prefer the anchor line itself, falling back outward only when
it didn't survive: a context line vanishes when the context size shrinks,
whereas additions and deletions always survive, so expansion stops at the
first change line in each direction and the candidate list always contains a
survivor. Landing on the nearest survivor keeps scrolling to a minimum, and a
context line that is still in the patch stays put (or stays selected).
This generalizes the shared restore helper to take a prioritized candidate
list (the escape path passes a single candidate). It covers the focused main
view and every side panel's diff, since they all render into the same "main"
view through the same path. A showing selection is re-established on the landed
line; otherwise the view stays in scroll mode.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The escape restore's mechanism — set a RenderRestore that scans the
re-rendering content for a target patch identity and, once it loads,
positions the view on the matched row — is about to gain a second caller:
preserving a diff view's scroll/selection when its -U context size changes
re-renders it (the sibling consumer of the diff-line primitive). That caller
positions the row differently (put it back where it was, rather than scroll to
and select it), so split the positioning out behind a `place` callback and
keep the scan/swap machinery shared.
Behaviour-preserving: RestoreFocusedMainViewOnEscape passes the same
FocusPoint-and-select closure the inline Apply used.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Diving into staging from a focused main view records a snapshot so escape can
return there. But it was recorded only on the half we entered (unstaged), and
staging the last unstaged hunk moves the selection to the staged half
(RefreshStagingPanel pushes StagingSecondary when the unstaged state goes empty).
Escaping from there found no snapshot and fell back to the files panel, instead
of returning to the focused main view.
Record the snapshot on both staging halves at entry, and clear both on escape.
Escaping from either half now returns to the focused main view, and the
identity-based restore lands on the line the explorer ended up selecting — which,
with the last unstaged hunk gone, is shown in the main view (the file now has only
staged changes, so no split). Routing the split-and-tabbed-to-staged case to the
secondary focused main view is left for follow-up; see focused-main-view-notes.md
§14.3.
Escaping a patch explorer (staging / patch building) back to the focused main
view it was entered from used to replay a numeric scroll position and selection
index captured on the way in. But the reason to escape after staging or dropping
a hunk is that the content changed, so a saved index points at the wrong line —
and the host auto-advances the explorer's selection to a still-valid line anyway,
which is the line the user actually cares about returning to.
Restore by *patch identity* instead. On escape, read the (file, type, source
line) the explorer currently has selected, then have the main view's re-render
land on the row that matches it: scan the incoming content as it loads (the
inverse of the diff-line primitive), and once the matching row plus a screenful
below it have loaded, swap the off-screen render in and scroll to / select that
row in one step. FocusPoint with scrollIntoView centres the row only if it's
off-screen, so the common unchanged-content escape — where the row is already
where it was — doesn't move at all. If the line is gone (the content really
changed), nothing is forced.
This generalizes the scroll restore from a fixed origin to a predicate
(RenderRestore: FirstPaintReady decides when the saved position is reachable,
Apply re-establishes it), folding the separate selection restore into the same
first paint — so it no longer rides a post-load callback that could fire early.
The restore also now survives task replacement, which the numeric version did
not: a periodic refresh can stop the escape's re-render before it first-paints.
The pending restore is held on the buffer manager and is *not* cleared when a
task starts, so the replacement task picks it up. It is not gated on the command
key — staging the last unstaged hunk re-renders `git diff` as `git diff --cached`,
a different command, yet the line to land on is still in the new content — but
validates itself: the scan finds the target line only when the content still
contains it, so applying it to a different item is a harmless no-op. A task
clears it once it has applied it (found or not), so it lives for exactly one
re-render. Because the restore is anchored on content identity and is idempotent,
"survive replacement" and "restore by identity" are one mechanism, not two.
With the identity in hand the snapshot no longer needs the captured scroll/index;
they're derived from the explorer's live selection.
The diff-line primitive (recover a rendered row's patch-space identity) is about
to gain a second, inverse consumer: the escape restore scans a focused main
view's rows as it re-renders, looking for the row that matches a target patch
identity. That scan runs over the *loading* off-screen buffer, not the displayed
view, so the resolver can't be tied to the displayed view's per-view-line readers.
Pull the three backends (OSC metadata, buffer parse, lazygit-edit hyperlink) onto
a single buffer-agnostic resolver that takes a snapshot of a diff's per-line
content — text, metadata and hyperlink per unwrapped buffer line — and the line
to resolve. The forward consumers (click/enter/edit/PR) feed it a snapshot of the
displayed buffer (gocui.DiffLineContents) after mapping the wrapped view line to
its buffer line; the upcoming scan will feed it the off-screen buffer's loaded
rows. Behavior is unchanged.
Escaping to a focused main view restored the selection by scheduling, on
the next UI tick, a ReadToEnd whose callback re-selected the saved line.
But ReadToEnd fires its callback synchronously when the manager has no
live read channel, and the re-render task triggered by the push creates
that channel later, inside its own goroutine (after stopping the previous
task). If the UI tick won that race, the restore ran before any content
was loaded, FocusPoint no-oped against the unloaded line, and the
selection was silently dropped — intermittently, and more often under
load.
Thread the restore through the task instead: a thenForNextTask hook on
the buffer manager, folded into the next cmd/pty task's initial-read Then,
mirroring scrollToOriginYForNextTask. It runs once the task has read
enough to place the selection, and can't fire before the task exists. The
scroll restore already applies at the task's first paint, which precedes
the initial read's end, so the origin is in place when the selection is
restored.
This needs interactive verification (LAZYGIT_SLOW_RENDER + a real pager);
see focused-main-view-notes.md §13.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add mechanism #2 to GetDiffLineInfo: when a patched pager annotated each diff
line with OSC 456 metadata, read it back as the first backend, ahead of the
buffer parser (#1) and the lazygit-edit hyperlink. It is strictly higher
fidelity -- it carries the side explicitly, so it serves the renderings #1
cannot parse (delta's default mode, --line-numbers, diff-so-fancy) and conveys
deletions, which the hyperlink can't.
The host advertises the protocol versions it understands by setting
EMIT_OSC456_METADATA on the pager subprocess; a pager that doesn't understand
it ignores the variable, so this is safe to set unconditionally.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The focused main view's click/enter/e/G handlers all need the same thing: given
a rendered diff row, the patch-space line it corresponds to. Until now that came
solely from delta's lazygit-edit:// hyperlinks, which only carry a path and a
single line number — no side. That's lossy: for a deletion the number is the old
line, but the consumers fed it into new-file lookups, and two consecutive
deletions (which share a new-file line number) couldn't be told apart at all.
Replace GetFileAndLineForClickedDiffLine with GetDiffLineInfo, returning the
fuller (file, type, new-line, old-line) record from diff-line-metadata-notes.md.
This is mechanism #1: parse the decolorized view buffer — walk up to the file's
"diff --git" section, reuse patch.Parse on it (splitting multi-file commit diffs
on the "diff --git" boundaries), and read the type and line numbers off the patch
arithmetic. It serves the structure-preserving renderings — no pager, git diff
--color, and delta --color-only without line numbers — with no external
dependency.
To avoid trusting a mis-parse, the parser bails when a hunk's body no longer
matches its header (Patch.IsWellFormed). That's what happens when a pager keeps
the diff/hunk headers but restructures the body: delta's line-number gutters push
the +/- marker off the start of each line, so every body line reads as context.
Such renderings fall through to the next backend rather than yielding a confident
wrong answer. (diff-so-fancy goes further and rewrites the headers too, so it
fails even earlier, on the missing "diff --git".)
GetDiffLineInfo is a seam with swappable backends: the buffer parser first, then
the old hyperlink reader as a fallback for renderings the parser can't handle
(delta's default mode, or delta with line-number gutters). The future #2 OSC
per-cell metadata reader plugs in ahead of both, behind the same record shape.
Wire the consumers to the record per that doc's field mapping:
- dive into staging/patch building lands on the exact patch line, looking a
deletion up by its old-file line number (PatchLineForOldLineNumber) so the
two-deletions case resolves correctly;
- `e` edits at the new-file line;
- `G` anchors the PR link on the left (old) side for a deletion, the right (new)
side otherwise.
The hyperlink fallback can't convey the side, so it reports DiffLineOther, which
the consumers treat as a non-deletion — i.e. exactly today's behavior.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Diving into staging or patch building from a focused main view (by
double-clicking a line, or pressing enter on the selected line) always
landed on a single-line selection. Entering the same views through the
side panel honours the UseHunkModeInStagingView config and selects the
whole hunk by default. The two ways in should agree, so that diving in
from the main view feels like the established flow.
A non-negative line index in NewState was overloaded for two intents:
clicking directly on the patch explorer view (where a single-line range
is the start of a drag) and diving in from the main view (where we want
the default select mode). Distinguish them with SelectLineInDefaultMode
on OnFocusOpts: the main-view entry points set it; the click-to-drag
path does not.
In hunk mode the selection covers the block of changes around the
clicked line. A context line has no surrounding changes, so we snap to
the next change line (as toggling hunk mode does); the clicked context
line itself is then not part of the selection.
This is a separate commit only because the branch is a throwaway
prototype; in a real history it would be folded into the commit that
introduces the focused-main-view enter behavior rather than landing on
top of it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
EscapeFromPatchExplorer re-renders the side panel's content back into the main
view and wants to land at the scroll position and selection the user had before
diving into staging/patch building. The previous version set the origin on the
next UI tick, after the placeholder had already been painted at the wrong
position, so the restore was visible as a jump.
Instead, ask the re-render itself to restore the scroll (via
ScrollToOriginYForNextTask), so the saved position is applied in the first
paint that shows the real content. The selection still needs the diff loaded
down to the selected line, so restore it via ReadToEnd once the content is
fully read; the scroll is no longer touched there.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This was already possible, but only when a file was selected, and it woudln't
always land on the right line when a pager was used. Now it's also possible to
do this for directories, and it jumps to the right line.
At the moment this is a hack that relies on delta's hyperlinks, so it only works
on lines that have hyperlinks (added and context).
The implementation is very hacky for other reasons too (e.g. the addition of the
weirdly named ClickedViewRealLineIdx to OnFocusOpts).
Git doesn't always write conflict markers of seven characters: the
conflict-marker-size gitattribute overrides that per file, and it is set
for good reasons — for file types whose regular content tends to contain
marker-looking lines, such as documentation about merging, or test
scripts. We hard-code seven characters everywhere we look for markers,
so none of that works.
Prepare for honoring the attribute by threading the marker size through
everything that recognizes a marker, carried on the file model. Nothing
fills it in yet, so we still use git's default size of seven everywhere,
and matching is unchanged: a marker consists of exactly that many marker
characters, and all but the "=======" one are followed by a space and a
label.
Entering a submodule clears GIT_DIR and GIT_WORK_TREE, as it must: they
say where the superproject is. But the stack we push the superproject
onto so that escape brings us back only held its path, and for a repo
opened with --git-dir/--work-tree the path leads nowhere — git can't
find a repo there. Escaping out of a submodule of a dotfile repo failed
with "not a git repository", or, if some unrelated repo happened to lie
above the work tree, quietly switched to that one instead.
Push the environment onto the stack along with the path, taken from the
repo paths rather than from the process env, so that it also covers a
repo we worked the location out for ourselves.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The commit list a filtering mode change leaves behind has nothing to do
with the one that was showing, so the scroll position it inherits says
nothing about where the selection ended up, and the selection can land
anywhere off screen. PostRefreshUpdate only moves the cursor within the
existing scroll position, so ask for the scroll separately, the way the
commits refresh does when it moves the selection itself.
Exiting filtering mode looked like it worked, but only by accident: the
commits refresh recognizes the commit that was selected before it ran,
selects it again at its new index, and scrolls because the index moved.
That does nothing for the case where the commit is gone from the list, or
for entering filtering mode, where we select the first commit ourselves.
Entering or leaving filtering mode switched the screen mode and the
focused panel immediately, then reloaded the commit list in the
background. The result was an unfiltered list presented in the layout
that says "you are filtering", with nothing to say that anything was
still happening — and in a big repo that state can last seconds.
Before we stopped blocking the UI thread on refreshes, the reload
happened before any of it, so the two always agreed; the price was a
frozen UI for the duration.
Do neither: reload on a worker, so the UI stays live, and hold back
everything the user can see of the change until the new lists are ready,
so they still land together in one frame. A waiting status says what is
going on in the meantime, and blocking input means the keys pressed while
it runs arrive after the change rather than acting on a list that is
about to be replaced.
Setting a filter and clearing it are the same transition in opposite
directions: mutate the mode, bring the screen mode in line with it,
reload the views that depend on the filter, and put the selection
somewhere sensible in the reloaded commit list. They were implemented
twice, once in the filtering menu and once in ModeHelper, which is how
the two came to repaint the commit list in different ways.
Derive the screen mode and the panel switch from whether a filter is
active after the change, so both directions fall out of the same code,
and give ModeHelper the entry points for both. The filtering menu is
left with nothing but the menu.
Blocking keyboard input and hiding the working tree state mode are two
separate concerns; they were fused into one helper because every caller
so far wanted both. A caller that blocks input for something other than a
rebase would then hide the "Rebasing" indicator for the duration of its
operation, which has nothing to do with it.
Make it an explicit option instead, so blocking input on its own doesn't
imply anything about the modes on display.
The mode it suppresses is active for any working tree state, not just a
rebase: merging, cherry-picking and reverting show through the same
indicator. Name it after what it hides.
Whether a graph can be drawn was read from the filtering mode, while the
graph itself is drawn over the commit list in the model. Those two only
agree once the list has been reloaded for the new mode, and a filtering
mode change reloads the list in the background, so in between we can be
asked to draw a graph over a list the graph makes no sense for.
That is not just cosmetic. Commits in a filtered list are almost never
each other's parents, so no pipe ever terminates: the pipe set grows by
one per row and every continuing pipe rescans it, which is cubic in the
length of the list. Escaping out of filtering mode with a filtered list
of 13000 commits — as you get once the 300 commit limit has been lifted,
which happens for good as soon as the selection passes COMMIT_THRESHOLD
— wedges the UI thread for around twenty minutes.
Record whether the list was loaded with a filter, right where the list
itself is stored, and decide from that. The graph now also stays up while
the pre-change list is still on display, rather than vanishing a moment
before the list it belongs to.
Having the cache in state.yml causes this file to be rewritten every
60s, which is annoying if you have a lazygit running in the background
somewhere without even realizing it, and it keeps overwriting the
foreground lazygit's newer command shell history and recent repos list
with its stale data. State.yml should only contain things that change in
response to user actions, not periodically.
GitHub exposes a combined status for the head commit without requiring
individual check contexts. Include that rollup in the existing request
and startup cache so every consumer sees the same state without making a
second network request.
This fixes the problem described in the previous commit; we no longer
capture the selection at the start of the refresh. There's no reason to
do that (we don't do it for branches either). It is enough to capture
the selection in the final bounce, before we assign the new model slice.
A refresh from the UI thread returns immediately and applies its model
and view updates as queued UI-thread callbacks. A key pressed before
those have run is handled against the stale, pre-refresh state. For most
keys that's harmless, but some handlers turn that state into git
commands: pressing space twice in quick succession in the staging panel
builds the second patch from the already-applied diff and fails with
'patch does not apply', because the refresh after the first press is
what moves the selection to the next stageable hunk.
Notably, this is not just a regression of the recent change that made
UI-thread refreshes non-blocking; the window was merely much narrower
before. A blocking refresh parked the UI thread while the scopes'
bounces were queued, and the event loop drains pending keyboard input
with priority over queued user events, so a key pressed during the
blocked window still beat the queued state updates. The guarantee that
the next keypress sees post-refresh state had already ended when the
scopes' state updates moved from worker-side mutex-guarded writes to
UI-thread bounces.
Fix it with the input-blocking mechanism we already use for commit
surgery, exposed as a new RefreshBlockingInput entry point: it begins
blocking events synchronously in the calling handler, and ends the
block from a callback that the finishing step queues behind the
refresh's own updates. Keys pressed while the refresh is in flight are
buffered and replayed, in order, against the fully refreshed state;
since a replayed key's handler re-enters this same path, a burst of
keypresses applies sequentially, each one seeing the previous one's
refresh. Unlike the old blocking refreshes, this doesn't freeze the UI
thread: rendering, spinners, resizing, and mouse scrolling keep working
while input is withheld.
Blocking input is opt-in per call site rather than the default for all
UI-thread refreshes, because most refreshes (the focus-in and startup
refreshes, say) don't produce state that the next keypress depends on,
and blocking on them would delay typing for no reason. It should also be
limited to quick, narrow-scoped refreshes: a full refresh, or any scope
that pulls in COMMITS, can take very long in large repos and should
usually not hold up input.
The staging panel's stage/discard/edit-hunk refreshes use it now.
A few comments still reasoned in terms of SYNC vs ASYNC refreshes, a
distinction that no longer exists: sync vs async is now derived from the
calling thread. Restate them in terms of the current mechanisms
(RefreshFromWorker blocking its worker, model updates being enqueued on
the UI thread) without changing any behavior.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Checking out a remote branch that has no local counterpart creates the
local branch, refreshes, and then checks it out. The refresh exists so
that CheckoutRef finds the new branch in the model and attaches an inline
status to the branch item instead of showing a global waiting status. But
since UI-thread refreshes stopped blocking, the checkout started before
the refreshed branches had landed in the model, so the lookup failed and
we always got the waiting status. Run the checkout from the refresh's
Then, which is queued behind the model update.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
PostFetchRefresh's refresh is the only background refresh carrying a
Then callback, and Then callbacks are not generation-guarded: when the
background fetch's refresh crossed a repo switch, the callback still
ran — in the new repo — and auto-forwarded the new repo's branches
because the old repo's fetch had completed. That was harmless in
practice (the update-ref call compares against the expected old value,
and it only does what the next fetch's auto-forward would do anyway),
but mutating refs in a repo whose fetch never happened is not an action
the user took. Skip the auto-forward when the repo generation changed
since the fetch started.
The generation is captured by the fetch's callers before the fetch
runs, not by PostFetchRefresh itself: the background fetch doesn't
block repo switching and is a network call, so by the time
PostFetchRefresh runs a switch may already have happened — a capture
there (or the one the refresh itself takes) would compare against the
new repo's generation and let the auto-forward through. For the manual
fetch the capture point makes no difference, since a foreground
operation blocks repo switching for its entire duration.
This deliberately guards only this call site rather than making Then
callbacks generation-guarded in general: a Then is an arbitrary
callback, and whether it is safe to skip on a repo switch is a decision
for the author of the call site.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The refresh workers read a few files at paths relative to the process
working directory: the submodule config read of .gitmodules, the files
refresh's check for conflict markers, and the submodule stash's
existence check. Git commands are pinned to the repo their instance was
created for, but these Go file reads still followed the cwd, so a
background refresh crossing a repo switch would read the new repo's
files while computing data for the old one. Join them with the worktree
root of the instance they belong to. (Most git-state file reads —
working tree state, rebase todos, bisect info — already resolve
against RepoPaths and need no change.)
This also fixes the submodule stash's existence check for nested
submodules: it stat'ed submodule.Path, which is relative to the parent
module, against the repo root — now it uses the submodule's full path,
matching the stash command right below it.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The refreshes on focus-in, right after a repo switch, and after
returning from a subprocess are full foreground refreshes, so their
tasks kept Busy() true for as long as the slowest scope took — and any
switch attempt in that window was refused with the "can't switch"
toast. The focus-in one is particularly annoying: focusing lazygit is
often precisely what the user does in order to switch repos, and right
after regaining focus is when a refresh takes longest.
Blocking the switch bought nothing there. The refusal exists for user
operations, whose follow-up work (e.g. a Then callback reading the
model) isn't covered by the switch-safety guards; but these refreshes
merely reload state, and a refresh by itself is now switch-safe: its
git commands run against the repo it was started for, and the
generation guard drops its updates when the repo changed.
We can't just mark them Background, because that flag also decides
whether the files refresh lets git take optional locks to persist its
refreshed stat cache — worth doing for an attended refresh, and the
focus-in refresh (typically running right after external changes) is
the case that profits most. So split the two meanings: a new
DontBlockRepoSwitch option dispatches the refresh's tasks as background
tasks (excluded from Busy()) while keeping the attended optional-locks
behavior. Combining it with Then panics, since Then is not
generation-guarded.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
An error returned from a gocui worker is shown to the user in an error
popup. For the branch loader's behind-counts worker that used to be the
"no such ref" popup when a background refresh crossed a repo switch:
the old repo's main branch didn't exist in the new repo. The previous
commits fix that scenario properly — the command now runs against the
repo the refresh was started for — but a stale worker can still fail
legitimately, most plausibly because that repo was deleted after
switching away from it (e.g. removing a worktree). Its results are
dropped anyway, so log the error instead of alarming the user about a
repo they already left.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A background refresh's model writes are dropped by the generation guard
when the repo is switched mid-flight, but its git commands kept running
— and because the refresh read the live git instance at each step, any
command issued after the switch ran against the new repo. Now that git
commands are pinned to the directory of the instance they were built
from, capture the instance once when the refresh starts and run every
scope's git work through it, so a switch-crossing refresh keeps
addressing the repo it was started for.
The instance is captured together with the repo generation, on the UI
thread (where repo switches run), so the pair can't straddle a switch:
an old instance paired with the new generation would compute data from
the old repo and write it into the new repo's model unguarded.
This also removes the refresh workers' unsynchronized reads of the live
instance pointer, which raced its reassignment on the UI thread when a
background refresh crossed a repo switch (foreground refreshes can't
cross one: they keep Busy() true, which refuses the switch).
Two reads keyed app-state by the live instance's repo path on a worker
and now use the captured instance, fixing which repo they file under
when crossing a switch: the pull-request cache, and the "user dismissed
the base-remote prompt" flag. The base-remote menu's handlers keep
reading the live instance: a switch dismisses any open popup, so they
can't run against the wrong repo (and the OnPress body runs under a
foreground task, which blocks switching anyway).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
When the user picks a base remote in the "select remote repository"
prompt, we called setGithubPullRequests directly, bypassing the refresh
machinery — which meant hand-rolling the refresh env that call needs
(with a comment explaining why), and fetching against the branches
captured when the prompt was created. Issue a PULL_REQUESTS-scoped
refresh instead: it re-reads branches and remotes (both fast even in
large repos), fetches against those fresh values, and gets the refresh
machinery's guarantees without any special-casing. The config write is
re-read by the refresh from git config, so it is guaranteed to be
picked up.
The waiting status now covers the config write and the branches/remotes
reload, while the GitHub request itself continues as a background task
— which is how every other pull-request fetch behaves.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Each status used to start a spinner render loop of its own, running on
a worker that inherited the foreground/background flavor of the
status's owner, and exiting only once the entire status stack was
empty. That shape had a real bug: a foreground operation's loop could
be kept alive by someone else's status. Finish a quick operation with
a waiting status while a background fetch's "Fetching..." status is
still showing, and the operation's render loop — a foreground worker
task — keeps ticking until the fetch ends. Busy() stays true for that
whole time, so repo switching is refused even though nothing is in
flight anymore; with a fetch hanging on a slow network, that means
minutes. The shape was also wasteful: overlapping statuses were each
drawn by their own loop (plus a duplicate whenever a task was paused
and resumed while another status was showing), all redundantly
redrawing the same top status.
Replace the per-status loops with a single loop owned by the status
stack as a whole: whoever shows the first status starts it, and it
exits after drawing a final empty frame once the last status is
removed. The claim/release methods on StatusManager keep the loop
flag's transitions atomic with the stack under the one mutex, so a
status added while the loop is about to exit starts a fresh loop
instead of going unrendered.
The loop always runs as a background task now: rendering issues no
git commands, so it never needs to block repo switching, and a
foreground operation's busy-ness is already carried by its own worker
task. This is what fixes the bug above, and it retires the need to
thread a foreground/background flag through the waiting-status
helpers altogether.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Every full refresh includes the PULL_REQUESTS scope, and the worker it
spawns inherited the refresh's foreground/background flag. Full
foreground refreshes happen at startup, after switching repos or
worktrees, and when the terminal regains focus, so the GitHub API
request ran as a foreground task there, keeping Busy() true until it
completed. On a healthy network that's a few hundred milliseconds and
nobody notices; on a very slow one the request can stall for minutes,
and every attempt to switch repos in that window was refused with
"Can't switch repositories while an operation is in progress" even
though lazygit looked completely idle. (The request has no visible
status; at most, a background fetch hanging on the same bad network was
showing its "Fetching..." spinner, pointing the blame at the wrong
operation.)
The switch-safety guard only needs to wait for operations whose
remaining git commands would run against the wrong repo after a switch.
The pull-request fetch runs no git commands at all, and its model
writes are dropped when the repo generation has changed in the
meantime, so there is no reason for it to block switching.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
