A focused diff must replace the patch builder's whole-file toggle
without losing file creation or deletion metadata. Ordinary modifications
and renames must remain line-level selections even when every changed
content line is selected.
Ask the patch builder's canonical raw diff whether it consists of one
hunk containing only additions or only deletions and no context. Use the
whole-file operation only when the selection also covers every change.
Co-Authored-By: GitHub Copilot <copilot@github.com>
The patch-building panel could move an added file as a file, while
selecting every visible change through the focused diff currently moves
only its content and leaves the empty file behind. Keep the intended
index state beside the current one so the replacement flow demonstrates
that gap before it is fixed.
Co-Authored-By: GitHub Copilot <copilot@github.com>
The commit rewrite triggered by moving a custom patch must keep the
user's place without relying on the patch-building panel. Build the
patch and leave the affected range selected in the focused diff so the
regression test follows the UI that survives this branch.
Co-Authored-By: GitHub Copilot <copilot@github.com>
Removing the staging panel must not lose the handoff that happens when a
file's final change disappears. Drive that behavior through the focused
main view so the test continues to require focus and selection to follow
the files panel onto the next diff.
Co-Authored-By: GitHub Copilot <copilot@github.com>
Keeping the same staging interactions covered through the explorer would
pin the suite to the panel that this branch removes. The focused main
view now owns line, range, hunk, navigation, search, context-size, and
rapid-input coverage.
Co-Authored-By: GitHub Copilot <copilot@github.com>
The hint was raised on entering the staging panel, once, to explain that
the selection mode there had changed and how to get the old one back. That
panel is gone, and with it the moment the hint was tied to; what is left is
a string nothing prints and a flag nothing reads.
The advice it carried is not lost: the option it names is documented, and
the key that switches modes is on screen while a diff is focused.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Space in the pane previewing the patch had nothing to act on: everything
there is in the patch already, so it can only mean taking those lines back
out, in the way that space in the staged side of the working tree's diff
only unstages.
A line of the patch can't be found by its number: the patch renumbers
whatever follows a change it leaves out, so a line of it names a line of
the commit's diff that isn't the one shown. A line is identified instead
by its position among its file's changes, counted in the diff the patch is
shown as. That is the same position it has among the changes the patch
holds for the file, since both are in the order the file has them. This
holds however the rendering came out, so a renderer that groups a hunk's
deletions before its additions, or leaves a change out altogether, no
longer matters.
The pane's rows are stated in terms of the trees the patch was
materialized into, so the paths come back to the repo's own before
anything is looked up in them. This also lets the lines there be copied
and edited, as the lines of a commit's diff can be.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The pane beside a commit's diff showed the patch being built from it by
assembling the text itself. That text could not be handed to a diff
renderer the way a diff can: a stdin filter might have coped, but a tool
that diffs two files could not. Its idea of how much context to show
around a hunk was also its own rather than git's.
Materialize the patch instead: write each of its files as it is before the
patch into one tree and as it is after into another, and let git diff the
two trees. The patch becomes a diff of real files, rendered by whatever
renders the rest of them, with git's own context around it. Its lines can
then be pointed at; taking them back out of the patch will need that.
The trees are named a and b, so that with git's own prefixes suppressed
the paths read like an ordinary diff's over the repo's own paths. They are
written when the patch changes rather than when it is shown, the patch
builder counting its own versions for that, and they go away with the
patch.
A renamed file is materialized under the name the patch expects to find it
under. Where the patch carries the rename, that is the name the file had
before, so the rename comes out as a rename. git names the trees
themselves in the two rename lines, having only the two paths to go by.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A patch built from what is on screen has to show what is in it, over the
diff those lines were taken from — which may be a diff renderer's rendering
of it, whose bytes are none of ours to touch.
So the marks are drawn in a gutter over the content, from the lines the
patch holds rather than from where they were drawn last: they are worked out
again whenever a pane's content settles, which is what keeps them right
across a renderer switch, a change of context size, or a walk through the
commits, and whenever the patch itself changes.
They come and go with the focus, being what the space key in the focused
view would act on, and stay while the focus moves between the diff and the
patch previewed beside it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The five panels that show a commit's diff now offer on it what the patch
explorer offers on the one file it can show at a time: space takes the
selected lines into the custom patch being built, or back out of it again,
and d takes them out of the commit itself.
So a patch can be built straight from the diff already on screen, over
several files at once and without entering the commit's files first — and
from a stash entry or a reflog entry too, whose diffs the explorer could
reach but which had no patch preview of their own to show for it.
The two arrive together because they are one answer to what a commit's diff
offers, and because d needs exactly what space builds: a patch of the
selection, which the rebase then removes from the commit. Which of the two
space is doing is decided by the first selected line, as toggling a file in
the commit files panel is, and the selection then moves past the lines
dealt with — they are still in the diff, a patch leaving the commit alone,
so a second press would otherwise take them straight back out.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Which line of a file a diff row is, in the terms a patch is built in, is
asked in two places already — staging a selection and editing a hunk — and
is about to be asked in a third, of a commit's diff rather than the working
tree's. The rule is the same one everywhere: an addition is where it sits in
the new version of the file, a deletion where it sat in the old one.
So the row is asked for its identity and the patch package for the lines
with those identities, instead of walking the diff working the identities
out again.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Removing lines from a commit, moving a custom patch out of one, undoing
either — none of it goes through the focused main view, so a selection over
that commit's diff was left where it was, painted over a rendering of a diff
that no longer exists.
Every render now asks whether it is showing a different diff than the one on
screen, which is what a rewritten commit looks like, and puts the selection
back on the change that has taken its place — the same reveal an action in
the view does for itself. It stands down for a render of the same diff,
where a selection, perhaps a range still being made, is exactly right as it
is, and for one that something more precise is already waiting to place.
The key a render is remembered under moves ahead of anything that might
change the command's arguments, so that it says which diff is being
rendered and nothing else.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
It is built entirely on that helper, and the render chokepoint that is about
to want it lives in the gui package, which cannot reach a controller.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Five panels show a commit's diff in the focused main view — one file of it
from the commit files panel, the whole of it from the commits, sub-commits,
stash and reflog panels — and each answered for that diff itself, in the
same words.
Have them share one object instead, told only which diff it is, so that what
follows has a single place to say what can be done to a commit's diff and
needs saying once rather than five times.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A patch is built in terms of where a line sits in the file's diff, which is
what the patch explorer has to hand. The main view doesn't: a row of it
resolves to a line of a file, and what index that line has depends on how
much of the diff has been read and how a diff renderer chose to lay it out.
So let the two meet at the patch builder's edge, in the identity of a change
line — its number on the side it belongs to — leaving the main view to speak
only of lines it can see.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The lines of a commit's diff that are in the custom patch being built have
to be shown as such over whatever a diff renderer made of that diff, whose
bytes we can't touch — and mustn't, since everything we know about a row is
keyed by where it sits in the content.
So the marks are a decoration drawn over the content instead: a column
reserved at the left of every line, blank but for the lines that are in,
with the content shifted along past it. Nothing about the content changes
but the width it has to wrap in.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The staging view can hand the hunk you are on to an editor and apply what
comes back, which is how you stage something the diff cannot express: half
of a changed line, or a change written differently from either side. The
focused main view has to be able to do the same before that view can go.
The hunk is git's own, context and all, rather than lazygit's block of
adjacent changes: an editable patch is one that still applies, and the
context lines are what let git place it. What the editor leaves behind is
applied whole rather than matched against the file's diff again, the point
being that it says something the diff didn't.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Editing a hunk points an editor at a patch we wrote to a temp file and
waits. Nothing about the repo has changed when the editor returns: the
change lands when the caller applies what came back. But the suspend path
this went through refreshes as soon as the subprocess exits, so it reads
the state from before the patch is applied and then races the caller's own
refresh to publish it. Whichever lands last wins, and when it is the stale
one the files panel goes on showing the file as it was before the edit
until something else refreshes.
The caller is the one that knows when there is something new to see, and
both callers already refresh once they have applied the patch, so the
refresh in the middle only ever had a wrong answer to give.
Being a publish-order race, it doesn't reproduce reliably enough for a
test; holding a background refresh between its read and its publish shows
it every time.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Staging a selection walks the file's diff to find the lines the user
pointed at, by where each of them sits in the file. The next commit needs
the same answer for a single row, so pull the walk out of the staging path
before there are two copies of it.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two space presses in quick succession only staged one hunk. Input is
withheld until the refresh has landed, which is enough where the diff is
rebuilt on the spot, but the main view re-renders asynchronously: the
selection only moves to the next change once that render is on screen, so
the second press acted on lines that were no longer in the diff, and
staged nothing.
So the wait is now for the selection to be where the work carries on from,
rather than for the model. A restore therefore has to say when it is done —
which it can be either way, since a view given a message rather than a
re-render now gives up the restore it was holding instead of leaving it to
claim some later render.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
EndBlockingEvents dispatches the keys buffered during a block from inside its
own call, so a keybinding handler runs in the middle of whatever the caller was
doing. If a caller ends the block partway through updating the screen, that
handler acts on state the caller has yet to finish writing.
No caller does that today; both of them end their block from a UI-thread
callback of their own, with nothing left to do afterwards. The commit after this
one adds a caller that can. It ends the block from a render restore, and a
render of the two main panes resolves that restore halfway through laying the
panes out.
Queue the replay through Update instead, so the buffered keys arrive on a later
pass of the loop, as they would have if the user had pressed them then. Input
stays withheld until that pass runs. Gui events are dispatched in preference to
queued work, so a key pressed in between would otherwise be handled ahead of the
keys buffered before it.
The replay's error now reaches the error handler along with every other
handler's, and EndBlockingEvents has nothing left to return.
Three cases where landing on the change that took the place of the one acted
on is not the same as landing on the next line, or on the same line number:
staging an inserted line moves every later line of the file, so the hunk
below it is somewhere else afterwards; consecutive deletions all sit at the
same place in the new file, so nothing but their order tells them apart; and
unstaging half of a modification carries on in the pane the staged side has,
which is where the work was already.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Some merge conflicts can only be resolved by picking a side. The files
panel explains those rather than diffing them, and where one side deleted
the file and the other modified it, git's diff of that modification is
shown below the explanation. The focused main view took those change lines
for a diff of its own. It drew a selection over them and offered to stage
hunks of a file whose conflict staging can't resolve.
So have a render say whether it holds the diff the panel offers in the
main view, and put a selection only on one that does. Every diff render
already goes through NewMainViewDiffTask, so it says so for itself; the
custom patch preview, assembled as text rather than run as a command, says
so through NewMainViewDiffStringTask. Establishing a selection asks the
pane the same question rather than looking for change lines itself.
The selection has been wrong over this content since "Show a selection in
the focused main view" introduced it, and the fix belongs there. It lands
here instead because a render had no way to say what it holds until "Show
git's own diff when the renderer's can't be acted on" gave every diff
render one constructor to go through.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A diff renderer may lay a diff out however it likes: line numbers in a
gutter, the +/- column replaced by colour, the two sides in columns. Once
it has, we can only tell which line of which file a row shows if the
renderer says so. Under a renderer that doesn't say, the main view holds
a diff that can be read but not staged, edited or copied from. That is no
good now that the main view is where you stage.
So focusing it brings git's own diff instead, and every re-render while it
stays focused keeps to that, so staging a hunk doesn't flip back. Browsing
is untouched: you see what the renderer produced until you focus the view
to act on it.
Whether the renderer says anything is settled by asking it rather than by
watching it work: run it on empty input and see whether it announces the
protocol. Announcing is a property of the renderer, so the answer is known
before we render anything, and a diff with no lines to describe can't fool
it. Watching would have to see a diff go by first, and a binary file's diff
holds nothing that would tell the two cases apart. The answer is remembered
until the renderer changes. git itself is asked the same question, with the
renderer's own arguments, since it announces itself for exactly the formats
whose output can't be read back as a diff.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two things about a diff command follow from what its output is for: whether
the diff renderer produces it, and whether it is coloured. That was a
`plain` flag, which covers two of the three cases — the diff as configured,
and git's own uncoloured diff for building patches out of — and leaves no
room for the third, which is about to be needed: git's own diff, coloured,
for showing where the renderer's version of it can't be acted on.
So the flag becomes a mode. It also takes over deciding the colour, which
each command spelled out for itself, and it settles a question the flag
couldn't put: whether ignoring whitespace applies. It is about what the user
wants to see, so it holds for anything shown, and not for a diff a patch is
built from, which has to describe every change.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Staging in the main view leaves you looking at the working tree's diff with
everything you meant to stage in the index — and, until now, having to go
back to the files panel to commit it. The commit keys, and the one that
finds the commit a fixup belongs to, are offered there too, so that the
whole staging-to-committing round happens in one place.
They act on the working tree, which the main view only sometimes shows, so
they do nothing over a commit's or a stash's diff — browsing history can't
commit by accident — and are listed only where they do something. The check
happens per press: what the main view is showing changes as the user moves
about, while the bindings are registered once.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The remove key now works on a diff selection the way it does in the staging
view: on the unstaged side it throws the selected lines away, which it asks
about first, and on the staged side it takes them out of the index, which is
unstaging and needs no warning.
It goes through the same path as staging, so that everything around the
action behaves identically — the selection lands on the change that took the
place of the discarded one, and the focus follows the side it acted on.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Each side of a file's diff has a pane of its own, but a pane is only shown
while its side has something in it: staging the last unstaged change takes
the upper pane away, and unstaging the last staged one takes the lower one
away. The focus follows into the pane that is left, and the selection has to
be waiting there when it arrives — on the lines just acted on, which is where
they are now, unless they were discarded rather than moved, in which case on
what is left of the file. The pane being moved to shows no selection until
the restore places one, so that the selection it was left with the last time
it was used doesn't appear for a frame.
Whether the acted-on side still holds anything is a question the model can't
answer: a refresh only queues its update, and by the time it lands the
re-render this has to ride is already under way. So the answer is worked out
from what we just did — the files we changed report whether the selection
covered all of their changes, and the ones it didn't touch are as the model
describes them.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Configured to always split the diff, a pane is shown whether or not its side
of the file holds anything, so emptying the side the focus is on no longer
takes the pane away — but it does take away everything there was to do there,
which is the question the focus is really asking.
So the render says which of the panes it is giving something to act on,
rather than the focus reading that off which panes are shown.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Nothing has read that since the paint started settling the scroll position
before consulting the restore: what the answer was for was deciding whether
the reset the new content was owed still had to happen, and by then it has.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Each side of a file's diff has a pane of its own, and a pane is only shown
while its side has something in it. So anything that empties the side the
focus is on takes that pane away with it: committing what was staged, or a
commit or a discard happening outside lazygit and arriving with a refresh.
The focus was left on a pane that isn't there any more, where the next
keypress acted on nothing.
The render is what decides which panes are shown, so the question is asked
there, of every render rather than of the handful of actions that could think
to ask it themselves. The pane the focus moves into gets its selection once
that render has finished and there is something to put one on, the way
focusing it by hand would — and shows none until then, so that the selection
it was left with the last time it was used doesn't appear for a frame.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Where the selection starts out, and how it widens to a whole change block,
are questions about what the view is showing — the same rendered diff the
queries next door read. Nothing about them belongs to a keybinding, and the
render funnel is about to need them too, from a layer that can reach a helper
but not a controller.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Staging takes the lines it acted on out of the diff, so the selection has
nothing to sit on afterwards and would be left wherever those lines used to
be. What the user wants is the change that moved up into their place, so
that pressing the key again goes on to the next one — which is how staging
line by line through a file works.
The line acted on is gone, so it can't be remembered by identity the way a
re-render of the same diff remembers one; what is remembered instead is its
place in the sequence of the diff's changes, which the change after it
inherits. Staging the last change is the one case with nothing to inherit
it, and there the selection stays on the last change there is.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Putting a view back on a remembered line as it re-renders is two things: the
plumbing that watches the content arrive, reveals it at the right moment and
places the view, and the search that says which row to land on. Only the
second is specific to what is being remembered, and a second kind of it is
about to arrive — the change line an action leaves the selection on, which
is found by counting rather than by identity.
Behaviour-preserving.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The diff of a deleted file is its content going away, so putting every
line of it into the index leaves an empty file there (modified in the
index, deleted in the working tree) rather than the deletion the user
selected. The reverse case matches: the diff of an added file is its
whole content, and taking all of it back out of the index leaves the file
tracked and empty rather than untracked again. In the staging view you
had to enter such a file deliberately to reach these cases, but stepping
through a directory's diff hunk by hunk runs into them routinely.
Selecting every change of a file says "this file", so stage or unstage
the file itself. This applies to any file, since applying a file's whole
diff amounts to the same thing everywhere else.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Space in the focused main view now acts on the selected lines the way it
does in the staging view: over the working tree's unstaged changes it puts
them into the index, and over the staged ones it takes them back out. Which
of the two it does is a property of the pane, each side of the diff having
one of its own. Nothing has to be entered first, and a selection reaching
across several files of a directory's diff is applied as one patch per file.
The rows on screen are only a picture of the diff, so the patch is built
from the diff itself: each selected row's identity — which file, which line
of it, and whether it is a deletion — is looked for in the file's own diff,
and the lines that match are the ones the patch includes. Matching by
position rather than by counting rows is what tells the two halves of a
modified line apart, since the deletion and the addition replacing it sit at
the same place in the new file.
Panels other than the working tree offer no action on their diff yet, so
space says nothing and does nothing there.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
A command that acts on a diff selection has nothing to say over content that
isn't a diff, and says so by describing itself as nothing — but a binding
that is displayed on screen is displayed whatever its description, so it
would show as a key with an empty label. Leave it out instead.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The main view shows a diff renderer's picture of a diff, and that is not
what you want on your clipboard: a renderer may drop the +/- column, move
the line numbers into a gutter, group a hunk's deletions before its
additions, or leave lines out altogether. So copying takes the lines from
the diff itself, locating them by the identity of the selected rows.
Only the panel that produced the diff can produce it again, though: the
working tree's staged or unstaged side, a commit's, a stash entry's. The
new seam is there for that. It asks per file, so that copying three lines
of a commit's diff doesn't fetch the whole of it, and it will grow the
actions on a selection as staging and patch building arrive.
The clipboard gets the run of diff lines from the first selected line of
a file to the last, so that lines the renderer hid come along and the
result still reads as a diff. Headers count as selected lines too. A hunk
header names the first line of its hunk, so it is looked for the way a
line of the file is. A file header names no line at all, and a rendering
may draw it over any number of rows, so a selection touching one of them
takes the whole header.
As in the patch explorers, a selection that is all additions or all
deletions loses its +/- column, ready to be pasted into code. One that
reaches into a header keeps the columns, and what comes out is a patch
fragment rather than lines of code.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The commit files context works out the two ends to diff from the ref (or
range of refs) it was entered for. The panels that hold those refs
themselves — commits, sub-commits, stash, reflog — are about to need the
same two ends, to hand out the diff behind what they render into the
focused main view. Pull the derivation out of the context so they can ask
for it rather than each spelling out the parent-of-from rule again.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Which pane a side of a file's diff appeared in depended on what else the file
had: the staged side had the lower pane while there were unstaged changes
above it, and took over the upper one when there weren't. So the same content
moved about, and which side a pane was showing was something the code had to
work out from the file's status rather than knowing from the pane.
Now each side has a pane of its own — unstaged above, staged below — shown
when there is something on it. A file with nothing unstaged shows its staged
changes in the lower pane alone, which then has the whole space, including
the label of the key that focuses it.
Nothing about this is visible to the user: the same diff appears in the same
place, with the same title and the same label, and the same keys focus and
scroll it. What it is for is the code, which no longer has to ask the file
what a pane is showing — and, once the diff can be staged from, no longer has
to make one key mean opposite things in the same pane.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
The main section shows one pane or two, which was enough while the second
pane only ever accompanied the first. It is about to have to show the second
one by itself: the working tree's staged changes are moving there for good,
and a file with nothing but staged changes has only that side to show — it
should have the whole section rather than sit under an empty pane.
So which panes are shown becomes a three-way answer, derived from which of
them the render has content for.
Behaviour-preserving: nothing renders into the secondary pane alone yet.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
A pane that has been emptied is showing nothing, but it kept the scroll
position it was left at and went on claiming the render it used to show,
so the next render into it — the same command's output, the file it was
showing being selected again — counted as content the view already had,
and was revealed partway down.
So say what the empty pane is: at the top, and showing nothing that a
render can be a re-render of.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
A pane the render has nothing to show is emptied, but keeps the scroll
position it was left at and its claim to the render it was showing. So
when it comes back — the file it was showing before is selected again —
its content is rendered under the same command it already had, which is
taken for the content the view is showing, and the user is left partway
down a diff they have only just been given.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
A range or hunk selection covers a stretch of the diff, and restoring only
the cursor left the other end pointing at whatever line the new rendering
happened to put there — with more context lines above, a selected hunk
would grow a tail of context it never covered.
So the other end is remembered by identity too, and put back before the
cursor. If it is a line the re-render dropped, the selection is left as the
single line we landed on: there is no telling which line inherits a
selection's edge, and a wrong guess acts on lines the user never chose.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Prep: pull the index out of the candidate search, so that looking up a
single diff line doesn't have to phrase itself as a search for the nearest
survivor among one candidate.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Ignoring whitespace is how you ask whether what you are looking at is more
than reindentation, so being dropped at the top of the diff is a poor
answer: you have to find your way back to the change you were asking about
to see what became of it.
Unlike the other ways of re-rendering a diff, this one can take the line
you were on away for good, along with the hunk or even the file it was in.
When it does, the view lands on the nearest line the diff kept, wherever in
it that is — the walk of fallbacks doesn't stop at the file's edge, since a
file of nothing but reindentation leaves nothing nearer to land on. And
when ignoring whitespace empties the diff altogether, there is nothing to
keep and the view simply shows what is left.
A whitespace-only change that shares a hunk with a real one is a happier
case: it is shown as a context line rather than as a change, but it is
still the same line of the same file, so we stay on it.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Cycling through the diff renderers is for comparing how they show the same
change, which is hard to do from the top of the diff each time. The line
you were on is the same line of the same file whichever renderer draws it,
so the restore finds it again — by the records a renderer states, where it
speaks the protocol, and by parsing its output as a diff otherwise.
The restore sits in DiffHelper.RenderToMainAgain. A switch between a
dark and a light terminal background renders the diff again through the
same helper, so it keeps your place as well.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Pressing { or } re-rendered the diff with less or more context around each
change and dropped you back at the top of it, so finding your way back to
what you were reading was on you.
Remember where the view is before triggering the re-render, and have the
restore put it back there. The line to remember is the selected one when a
selection is showing, and the middle visible line otherwise — what you are
looking at, rather than the view's top edge. It is remembered by identity,
because the new rendering puts it on a different line of the view, and it
is found again by the records a diff renderer states for its rows, or by
parsing the rendering as a diff where it still looks like one.
The line may not be there at all afterwards: shrinking the context takes
context lines away. So the lines around it come along as fallbacks, nearest
first, and the view lands on the nearest one that survived, put back on the
screen row it was on — leaving what you were reading where it was, give or
take the line that went. The walk that collects them covers the whole diff
rather than stopping at the nearest change on either side: those always
survive a context-size change, but not everything a re-render can do to a
diff is that considerate.
Under a diff renderer that says nothing about its rows there is nothing to
look for at all, and the view is left at the offset it had instead: often not
the line it was on, but always closer to it than the top of the diff. That is
what the re-render asks for besides the restore, since it is a different
command from the one that produced what is on screen and would otherwise be
taken for a diff the user has never seen.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
A re-render builds into an off-screen buffer and swaps it in when it has
read enough to paint. Deciding where the new content should be shown means
reading it before that swap: afterwards it is on screen already, and
whatever we then scroll to has been seen at the wrong position first.
So expose the off-screen buffer's diff-line contents and line count, the
latter for telling when a line found there has a screenful below it. The
contents come in two forms, the whole buffer and everything from a given
line on, so that a reader following the render as it loads can look at each
line once instead of re-reading the buffer per line.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
A view that re-renders content the user is already looking at, laid out
differently — a different context size, another diff renderer — starts the
new render from the top, losing where they were. Where that is can't be
carried over as a scroll position, because a different layout of the same
content puts the line they were on somewhere else; it can only be found by
looking at what the re-render brings in.
So let a restore be installed on the buffer manager just before the
re-render is triggered. It rides the next command task, which asks it after
each line whether enough has arrived to show what it remembers, and then
hands it the first paint: the restore searches the off-screen buffer, swaps
it in, and places the view, in that order, so that the search happens while
the previous content is still displayed and the new content is never drawn
at the previous render's scroll position. A restore that placed the view
keeps the scroll reset new content would otherwise get; one that couldn't
find what it was looking for leaves the render to do what it would have
done anyway, and the lines-read count still has the last word on when to
paint, so a restore can never hold a render back for ever.
Some renderings can't be searched at all: a diff renderer is free to say
nothing about which line of which file each row shows, and then no line of
the old rendering can be looked for in the new one. There the offset into
the content is all that is left to go on, and it is nearer to where the user
was than the top is, so a re-render can also ask merely to be left where it
is. That request rides the next task the same way, and answers the same
question the scroll reset and the loading placeholder are asking: whether
what is coming is content the user has not seen.
Both outlive the task they were installed for, like the pending scroll reset
does and for the same reason: that task can be stopped and replaced by a
background refresh before it ever paints, leaving the replacement to honour
it. The loading placeholder stays out of the way while either is pending —
blanking the view for a message before putting the user back where they were
is the flicker they exist to avoid.
Nothing installs either of them yet.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
