We write a hacky, one-off script to do that. We need this script only once, so
we don't bother polishing it much. We'll re-purpose it later in the branch to
convert the English translation set to JSON; that is an operation that we need
to do regularly in the future.
An inactive selection is one where the view is part of the context stack, but
not the active view. For example, the files view when you enter the staging
panel, or any view when you open a panel.
Remove the old mechanism of clearing the highlight in Layout.
This fixes a problem with a wrong highlight showing up in the staging panel when
entering a file with only staged changes.
Reproduction recipe:
1. stage all changes in a file by pressing space on it in the files panel
2. enter the staged changes panel by pressing enter
3. unstage one of the changes
This makes the unstaged changes panel visible, but keeps the focus in the staged
changes panel. However, the highlight in the unstaged changes view becomes
visible, as if it were focused.
To explain why this happens, you need to know how the selection highlighting of
a view is turned on or off. It is turned on when it gains the focus, i.e. when
ActivateFocus is called on it, which in turn happens when PushContext is called.
It is turned off in Layout when gocui sees that the current view is no longer
the same as last time, in which case it calls onViewFocusLost on the previous
current view.
This mechanism only works reliably when there is at most one PushContext call
per event handler. If there is more than one, then the first one gets its
highlight turned on, then the second one, but since gocui has never seen the
first one as the active view in Layout, it doesn't get the highlight turned off
again even though it should.
And this happens in the above scenario. When pressing enter on a file with only
staged changes, we first push the staging context (in
FilesController.EnterFile), and then later we push the stagingSecondary context
when we realize we only have staged changes. This leaves the highlight of the
staging context on.
runewidth.StringWidth is an expensive call, even if the input string is pure
ASCII. Improve this by providing a wrapper that short-circuits the call to len
if the input is ASCII.
Benchmark results show that for non-ASCII strings it makes no noticable
difference, but for ASCII strings it provides a more than 200x speedup.
BenchmarkStringWidthAsciiOriginal-10 718135 1637 ns/op
BenchmarkStringWidthAsciiOptimized-10 159197538 7.545 ns/op
BenchmarkStringWidthNonAsciiOriginal-10 486290 2391 ns/op
BenchmarkStringWidthNonAsciiOptimized-10 502286 2383 ns/op
In d5b4f7bb3e and 58a83b0862 we introduced a combined mechanism for rerendering
views when either their width changes (needed for the branches view which
truncates long branch names), or the screen mode (needed for those views that
display more information in half or full screen mode, e.g. the commits view).
This was a bad idea, because it unnecessarily rerenders too many views when just
their width changes, which causes a noticable lag. This is a problem, for
example, when selecting a file in the files panel that has only unstaged
changes, and then going to one that has both staged and unstaged changes; this
splits the main view, causing the side panels to become a bit narrower, and
rerendering all those views took almost 500ms on my machine. Another similar
example is entering or leaving staging mode.
Fix this by being more specific about which views need rerendering under what
conditions; this improves the time it takes to rerender in the above scenarios
from 450-500s down to about 20ms.
This reintroduces the code that was removed in 58a83b0862, but in a slightly
different way.
The rendering of remote branches is in no way dependent on the width of the view
(or the screen mode). Unlike in the local branches view, we don't truncate long
branch names here (because there's no more information after them).
This is an error introduced in d5b4f7bb3e.
For checkboxes it probably doesn't really make sense to use them yet, because
we'd have to find a way how you can toggle them without closing the dialog; but
we already provide rendering for them to lay the ground.
But radio buttons can be used already, because for those it is ok to close the
dialog when choosing a different option (as long as there is only one grounp of
radio buttons in the panel, that is).
Strike it through if not applicable. This will hopefully help with confusion
about the meaning of "all" in the "Discard all changes" entry; some people
misunderstand this to mean all changes in the working copy. Seeing the "Discard
unstaged changes" item next to it hopefully makes it clearer that "all" is meant
in contrast to that.
Several custom patch commands on parts of an added file would fail with the
confusing error message "error: new file XXX depends on old contents". These
were dropping the custom patch from the original commit, moving the patch to a
new commit, moving it to a later commit, or moving it to the index.
We fix this by converting the patch header from an added file to a diff against
an empty file. We do this not just for the purpose of applying the patch, but
also for rendering it and copying it to the clip board. I'm not sure it matters
much in these cases, but it does feel more correct for a filtered patch to be
presented this way.
We're going to add another argument in the next commit, and that's getting a bit
much, especially when most of the arguments are bool and you only see true and
false at the call sites without knowing what they mean.
This currently works (albeit with a bit of manual work, as the user needs to
resolve conflicts), and we add this test just to make sure that we don't break
it with the following change.
This currently works, we add it as a regression test to make sure we don't break
it. It is an interesting test because it turns the deletion of the file in the
moved-from commit into a modification.
This is important when using a pager that draws a horizontal line across the
entire width of the view; when changing from a file or directory that has only
unstaged (or only staged) changes to one that has both, the main view is split
in half, but the PTY task would be run on the view in its old state, so the
horizonal line would be too long and wrap around.
All PTYs were created with the size of the main view, on the assumption that
main and secondary always have the same size. That's not true though; in
horizontal split mode, the width of the two views can differ by one because of
rounding, and when using a pager that draws a horizontal line across the width
of the view, this is visible and looks very ugly.
When switching to a repo that was open before, the context tree is reused, so
before adding keybinding functions to those contexts again, we need to clear the
old ones.
When refreshViewportOnChange is true, we would refresh the viewport once at the
end of FocusLine, and then we would check at the end of AfterLayout if the
origin has changed, and refresh again if so. That's unnecessarily complicated,
let's just unconditionally refresh at the end of AfterLayout only.
We want to add an additional method to ISearchableContext later in this branch,
and this will make sure that we don't forget to implement it in any concrete
context.
Searching in the "Divergence from upstream" view would select the wrong lines.
The OnSearchSelect function gets passed a view index, and uses it to select a
model line. In most views these are the same, but not in the divergence view
(because of the Remote/Local section headers).
ListContextTrait.OnSearchSelect was introduced in 138be04e65, but it was never
called. I can only guess that a planned refactoring wasn't finished here.
Unfortunately it isn't possible to delete them. This would often be useful, but
our todo rewriting mechanisms rely on being able to find todos by some
identifier (hash for pick, ref for update-ref), and exec todos don't have a
unique identifier.
If exactly one candidate from inside the current branch is found, we return that
one even if there are also hunks belonging to master commits; we disregard those
in this case.
Copy the slice into a variable and use that throughout the whole operation; this
makes us a little more robust against the model refreshing concurrently.
It has two modified hunks, one for a master commit and one for a branch commit.
Currently we get an error mentioning those two commits, but we would like to
silently select the branch commit.
Put it into the individual menu items instead.
Again, this is necessary because we are going to add another entry to the menu
that is independent of the selected branch.
Instead, disable the individual entries in the menu.
This is necessary because we are going to add another entry to the menu that is
independent of the selected branch.
Previously the entire status was colored in a single color, so the API made
sense. This is going to change in the next commit, so now we must include the
color in the string returned from BranchStatus(), which means that callers who
need to do hit detection or measure the length need to decolorize it.
While we're at it, switch the order of ↑3↓7 to ↓7↑3. For some reason that I
can't really explain I find it more logical this way. The software out there is
pretty undecided about it, it seems: VS Code puts ↓7 first, and so does the
shell prompt that comes with git; git status and git branch -v put "ahead" first
though. Shrug.
It is a valid case for a branch to share no history with any of the main
branches, in which case git merge-base returns an error (and an empty string).
Since we can't distinguish this from one of the main branches having been
deleted, we shouldn't invalidate the cache in that case.
Use Equals instead of Contains for asserting the status view content. This
solves the problem that we might assert Contains("↓2 repo"), but what it really
shows is "↑1↓2 repo", and the test still succeeds. At best this is confusing.
Also, this way we don't have to use the awkward DoesNotContain to check that it
really doesn't show a checkmark.
To do this, we need to fix two whitespace problems:
- there was always a space at the end for no reason. Simply remove it. It was
added in efb51eee96, but from looking at that diff it seems it was added
accidentally.
- there was a space at the beginning if the branch status was empty. This is
actually a cosmetic problem, for branches without a status the text was
indented by once space. Change this so that the space is added conditionally.
It's a bit awkward that we have to use Decolorise here, but this will go away
again later in this branch.
We aren't using them, yet, except for deciding whether to show the warning about
hunks with only added lines.
Add a bit of test coverage for parseDiff while we're at it.
This broke with 81b497d186 (#3387). In that PR I claimed that we never want to
ask for force-pushing if the server rejected the update, on the assumption that
this can only happen because the remote tracking branch is not up to date, and
users should just fetch in this case. However, I didn't realize it's even
possible to have a branch whose upstream branch is not stored locally; in this
case we can't tell ahead of time whether a force push is going to be necessary,
so we _have_ to rely on the server response to find out. But we only want to do
that in this specific case, so this is not quite an exact revert of 81b497d186.
This broke with #3528.
If the remote branch is not stored locally, we only see question marks in the
branch status. In this case we can't tell whether we need to force-push, so it's
best to assume that we don't, and see if the server rejects the push, and react
to that by asking to force push. This second part is also broken right now,
we'll fix this in the next commit.
When branches are sorted by recency we have this logic that first loads the
branches so that they can be rendered quickly; in parallel, it starts loading
the reflog in the background, and when that's done, it loads the branches again
so that they get their recency values. This means that branches are loaded twice
at startup.
We don't need this logic when branches are not sorted by recency, so we can
simply load branches and reflog in parallel like everything else.
This shouldn't change any user observable behavior, it just avoids doing
unnecessary work at startup.
For tooltips that are just one or two characters longer than the available
width, the last word would be cut off. On my screen this happened for the
tooltip for the fixup command.
It can optionally be used to set the title of the panel that shows the output of
a command (when showOutput is true). If left unset, the command string is used
as the title.
To determine whether we need to ask for force pushing, we need to query the push
branch rather than the upstream branch, in case they are not the same.
In a triangular workflow the branch that you're pulling from is not the same as
the one that you are pushing to. For example, some people find it useful to set
the upstream branch to origin/master so that pulling effectively rebases onto
master, and set the push.default git config to "current" so that "feature"
pushes to origin/feature.
Another example is a fork-based workflow where "feature" has upstream set to
upstream/main, and the repo has remote.pushDefault set to "origin", so pushing
on "feature" pushes to origin/feature.
This commit adds new fields to models.Branch that store the ahead/behind
information against the push branch; for the "normal" workflow where you pull
and push from/to the upstream branch, AheadForPush/BehindForPush will be the
same as AheadForPull/BehindForPull.
Our code doesn't realize that we need to prompt the user to force push, when the
branch is up-to-date with its upstream but not with the branch that we're
pushing to.
It is unexpected that a function called PushBranch also sets the upstream
branch; also, we want to add a PushBranch function in the next commit that
doesn't.
This guards against accidentally renaming a model field and thereby breaking
user's custom commands. With this change we'll get a build failure when we do
that.
In go 1.22, loop variables are redeclared with each iteration of the
loop, rather than simple updated on each iteration. This means that we
no longer need to manually redeclare variables when they're closed over
by a function.
For custom commands it is useful to select an earlier command and have it copied
to the prompt for further editing. This can be done by hitting 'e' now.
For other types of suggestion panels we don't enable this behavior, as you can't
create arbitrary new items there that don't already exist as a suggestion.
In the custom commands panel you can now tab to the suggestions and hit 'd' to
delete items from there. Useful if you mistyped a command and don't want it to
appear in your history any more.
- If _not_ inside a neovim session, treat as
a normal nvim session and suspend lazygit.
- If inside a neovim session:
- Do not try to suspend lazygit.
- Send `q` keystroke to neovim session to quit lazygit.
- Send filename/line/etc. to neovim session.
We are going to truncate overly long lines returned from git log, and the most
likely field that is going to make the line too long is the subject; so we must
put it last, otherwise we'd end up with not enough fields to split when it's too
long.
It might not be obvious from the diff what's happening to the mock command
output in the test: it didn't have the divergence field (">") at all, which was
kind of a bug. It didn't matter for these tests though, because we are not
testing the divergence here, and our production code happens to be resilient
against it missing. But now we must add the ">" field before the subject.
Scanners can return errors (e.g. ErrTooLong), and if we don't handle it, the
cmd.Wait() call below will block forever because nobody drains the command's
output.
This happens for CommitLoader.GetCommits when there's a commit whose subject
line is longer than approx. 65500 characters; in that case, lazygit would lock
up completely. With this fix it remains usable, but the commit list is truncated
before the bad commit, which is not good enough. We'll improve that in the
remaining commits of this branch.
Deadlock reporting broke in e1ceb6892a; since then, it was *off* when running
debug builds normally, but *on* when debugging an integration test. Both of
which are exactly opposite of what we want.
Change `func displayCommit()` so all the individual strings are built first,
then the whole thing `cols` is put together. Before, most strings were built
prior to constructing `cols`, but a few were built inside the `cols`
construction.
The rebase.updateRefs feature of git is very useful to rebase a stack of
branches and keep everything nicely stacked; however, it is usually in the way
when you make a copy of a branch and want to rebase it "away" from the original
branch in some way or other. For example, the original branch might sit on main,
and you want to rebase the copy onto devel to see if things still compile there.
Or you want to do some heavy history rewriting experiments on the copy, but keep
the original branch in case the experiments fail. Or you want to split a branch
in two because it contains two unrelated sets of changes; so you make a copy,
and drop half of the commits from the copy, then check out the original branch
and drop the other half of the commits from it.
In all these cases, git's updateRefs feature insists on moving the original
branch along with the copy in the first rebase that you make on the copy. I
think this is a bug in git, it should create update-ref todos only for branches
that point into the middle of your branch (because only then do they form a
stack), not when they point at the head (because then it's a copy). I had a long
discussion about this on the git mailing list [1], but people either don't agree
or don't care enough.
So we fix this on our side: whenever we start a rebase for whatever reason, be
it interactive, non-interactive, or behind-the-scenes, we drop any update-ref
todos that are at the very top of the todo list, which fixes all the
above-mentioned scenarios nicely.
I will admit that there's one scenario where git's behavior is the desired one,
and the fix in this PR makes it worse: when you create a new branch off of an
existing one, with the intention of creating a stack of branches, but before you
make the first commit on the new branch you realize some problem with the first
branch (e.g. a commit that needs to be reworded or dropped). It this case you do
want both branches to be affected by the change. In my experience this scenario
is much rarer than the other ones that I described above, and it's also much
easier to recover from: just check out the other branch again and hard-reset it
to the rebased one.
[1]
https://public-inbox.org/git/354f9fed-567f-42c8-9da9-148a5e223022@haller-berlin.de/
Sometimes it takes a while to get PRs accepted upstream, and this blocks our
progress. Since I'm pretty much the only one making changes there anyway, it
makes sense to point to my fork directly.
It is now only used as the error handler that is passed to gocui.Gui on
construction; it's not a client-facing API any more. Also, it doesn't have to
handle gocui.ErrQuit, as gocui takes care of that.
This lets us get rid of a few more calls to Error(), and it simplifies things
for clients of OnWorker: they can simply return an error from their callback
like we do everywhere else.
In Gui.onWorker we only make the minimum possible change to get things to
compile after the API-breaking change of the gocui update; we'll make this
cleaner later in this branch.
We upgraded our minimum Go version to 1.21 in commit
57ac9c2189. We can now replace our
`utils.Min` and `utils.Max` functions with the built-in `min` and `max`.
Reference: https://go.dev/ref/spec#Min_and_max
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
Unfortunately the migration code requires yaml v3, but our yaml fork is based on
v2, so we need to import both in app_config.go in this commit, which is ugly. We
can clean this up in the next commit.
Make it recognize URLs wrapped in angle brackets, and followed by punktuation.
We don't need this for the status panel, but we will need it for confirmation
panels.
Changing globals in the init() function of a test file is a bad idea, as it
affects all other tests that run after it. Do it explicitly in each test
function that needs it, and take care of restoring the previous value
afterwards.
For people who have the log.showsignature git config set to true, trying to
reword a signed commit would put the signature verification into the subject
field and the commit subject into the description field of the commit message
panel. Amending commits, adding co-authors to a commit, and copying a commit
message to the clipboard would all be broken in a similar way.
Using the "Add to .git/info/exclude" in a worktree results in an error message,
as the test shows. The same would happen in a submodule, but I'm not adding an
extra test for that, as the circumstances are the same.
When exiting filtering mode, we currently keep the selection index the same in
the commits panel. This doesn't make sense at all, since the index in the
filtered view has no relation to the index in the unfiltered view.
I often use filtering mode (either by path or by author) to find a given commit
faster than I would otherwise be able to. When exiting filtering mode, it's
useful to keep the same commit selected, so that I can look at the surrounding
commits, see which branch it was a part of, etc. So reselect the commit again
after exiting filtering mode.
Sometimes this is not possible, most likely when the commit is so long ago that
it's outside of the initial 300 range. In that case, at least select the commit
again that was selected before I entered filtering; this is still better than
arbitrarily keeping the same selection index.
It would crash when some keybindings are set to null, and the filter string is
such that only those keybindings remain visible.
The reason for the crash is that when inserting non-model items (menu section
headers in this case) you specify a column to align them to. This works on the
assumption that the number of columns is always the same. It can cope with the
case that columns are removed because they are empty for all items; but it can't
cope with the case that the getDisplayStrings function returns a lower number of
columns.
And this is what happened here: MenuViewModel.GetDisplayStrings would omit the
keybinding column when none of the entries have a keybinding. This logic is
unnecessary, the generic list rendering mechanism takes care of this, so
removing this logic fixes the crash.
We do have to make sure though that the column is really empty when there's no
keybinding, so change the logic to use FgCyan only when there's a keybinding.
It is a bad idea to read a git-rebase-todo file, remove some update-ref todos,
and write it back out behind git's back. This will cause git to actually remove
the branches referenced by those update-ref todos when the rebase is continued.
The reason is that git remembers the refs affected by update-ref todos at the
beginning of the rebase, and remembers information about them in the file
.git/rebase-merge/update-refs. Then, whenever the user performs a "git rebase
--edit-todo" command, it updates that file based on whether update-ref todos
were added or removed by that edit. If we rewrite the git-rebase-todo file
behind git's back, this updating doesn't happen.
Fix this by not updating the git-rebase-todo file directly in this case, but
performing a "git rebase --edit-todo" command where we set ourselves as the
editor and change the file in there. This makes git update the bookkeeping
information properly.
Ideally we would use this method for all cases where we change the
git-rebase-todo file (e.g. moving todos up/down, or changing the type of a
todo); this would be cleaner because we wouldn't mess with git's private
implementation details. I tried this, but unfortunately it isn't fast enough.
Right now, moving a todo up or down takes between 1 and 2ms on my machine;
changing it to do a "git rebase --edit-todo" slows it down to over 100ms, which
is unacceptable.
In the test we simply removed the update-ref todo but didn't make any other
changes to the todos. This should really have kept everything the way it was,
including the other branch head. The fact that the star was gone was really
because of the bug that we are going to fix later in the branch.
Change the test so that it also makes a change before the update-ref todo; this
way we test that the star is gone because we deleted the update-ref, not because
of the bug.
To guard against the bug, we add another assertion for the branches view to test
that both branches are still there. This currently fails.
Running WSL without a container would be treated as native linux, causing problems at it would then attempt to use `xdg-open`.
This was caused by `isContainer()` always returning true due to some dubious conditionals. These have been removed.
The env-var check seems to not be used by lazygit, nor any common containers, and therefore appears to only exist to manually tell lazygit to behave as if it were inside of a container.
This functionality has been kept, but the env-var has been changed to be all uppercaps as to comply with the POSIX standard.
Fixes#2757
Bug introduced in 4d78d76
I often copy hashes in the commits panel in order to paste them into Github
comments (or other places), and I can't stand it when they have the full length.
I picked a default of 12 for this; I find this to be a good middle ground
between being reliable in large repos (12 still works in the linux kernel repo
today, but it might not be enough in really huge repos) and not being too ugly
(many smaller repos can probably get away with less).
We deliberately don't change this for the "Copy to clipboard" menu, since this
gives users a way to copy the unabbreviated sha if they need this occasionally.
To support this, we turn the confirmation prompt of the "Create fixup commit"
command into a menu; creating a fixup commit is the first entry, so that
"shift-F, enter" behaves the same as before. But there are additional entries
for creating "amend!" commits, either with or without file changes. These make
it easy to reword commit messages of existing commits.
We pass all of them to a single editor command, hoping that the editor will be
able to handle multiple files (VS Code and vim do).
We ignore directories that happen to be in the selection range; this makes it
easier to edit multiple files in different folders in tree view. We show an
error if only directories are selected, though.
We've seen a lot of issues recently where people complain that lazygit doesn't
behave as documented, but that was only because they were running the latest
release but were looking at the documentation of master. Make the documentation
links in the status panel point to the release that they are using in the hope
that this will help a little bit with this problem.
Some operations don't support switching to the editor from the commit message
panel; an example is the commit message panel that appears when moving a custom
patch into a new commit. Disable the "open in editor" menu entry in this case,
instead of silently doing nothing.
Previously we would hide it if no onSwitchToEditor function was set; that was
from a time when <c-o> was bound directly to the switch-to-editor command. Now
it is bound to showing a menu, and that menu is always available even if no
onSwitchToEditor function is set. (We rather need to disable the switch to
editor item _within_ that menu, see next commit.)
For die-hard fuzzy-searching fans it's probably in the way, so taking it out
makes fuzzy filtering work better. For substring filtering it always retains the
sort order anyway.
By default we now search for substrings; you can search for multiple substrings
by separating them with spaces. Add a config option gui.filterMode that can be
set to 'fuzzy' to switch back to the previous behavior.
It sorts them already, so it's unnecessary. In the next commit we use this same
code for substring searching too, and in that case we don't want to sort because
sorting is by Score, but we don't even fill in the score for substring
searching.
Lazygit has two ways to decide whether it needs to ask the user to force-push:
1. if it knows ahead of time that the push will fail because the branch has
diverged, by looking at the incoming/outgoing information that it shows as ↑3↓7.
2. by examining the error that comes back when the push has failed.
The second situation should happen only rarely, because lazygit fetches every
minute by default, so the ↑3↓7 information is usually up to date. It might not
be if the user turned off auto-fetch (or increased the auto-fetch interval).
However, in this case it's almost always harmful to prompt the user to
force-push, because we know that the reason for diverging is that something was
pushed to the remote, and we would wipe it out by force-pushing. In such a
situation, the more likely user action is to pull the remote changes and then
push normally again.
So just remove the second prompt, and replace it by a better error message when
we detect that updates were rejected remotely.
A little bit of history archeology reveals that the second prompt was added at a
time where we didn't have the first one yet, so at that time it made sense to
have it; but when the first prompt was added, we should have removed the second.
So far, the only situation where we called SetSelectionRangeAndMode was one
where the range could only get larger (in startInteractiveRebaseWithEdit, in
which case update-ref todos can be inserted by the rebase). However, in the last
commit we introduced a new call site where the range can get smaller, including
being reduced to a single item. Since this is indistinguishable from a single
selection, set the mode to none in this case; without this, hitting escape would
seemingly do nothing because it collapses the empty range selection.
... and when recalling a commit message from an old commit by pressing up-arrow.
This is necessary because committing turns our soft line breaks into real ones,
but when rewording we want to turn them back into soft ones again, so that it's
possible to insert words at the beginning of a paragraph and have everything
rewrap nicely.
This is only a best effort; the algorithm only removes those hard line breaks
that can be removed without changing the way the message looks. This works well
when the previous commit message was wrapped at the same width, which for most
users should be the most common case; but if it wasn't, the result is not great.
Specifically, if the old wrap width was smaller, some hard line breaks just
won't be removed; if it was wider though, you'll get an unpleasant comb effect
with alternating long and short lines. In such a case it's best to switch to the
editor and use whatever wrapping features you have there (e.g. alt-Q).
I have seen some commit messages that contain CRLF instead of just LF; I'm not
sure if these were created by a broken git client, but they exist, so we need to
deal with them. Editing them when rewording a commit sort of works, but is a
little strange; the \r characters are invisble, so you need an extra arrow key
press to skip over them.
In the next commit we are going to add more logic related to line breaks, and it
is getting confused by the \r, so it is becoming more important to fix this. The
easiest fix is to normalize the line endings right after loading.
This should arguably have been done in b133318b40 already; it's becoming more
important now because we're going to extend the common code with more logic in
the next commit.
We have such a test already (squash_fixups_above_first_commit.go), but it can't
be used for what we want to check here, because it uses the first commit, and we
can't move down from there. So create a new one that basically does the same
thing, but for a commit in the middle. The focus of this new test is to check
how the selection behaves; as you can see, there is a problem both when creating
a fixup and when squashing fixups. We'll address these separately in the next
commits.
It starts a rebase on the bottom-most commit of the range, and sets all the
selected commits to "edit" (skipping merge commits, because they can't be
edited).
In most real-world scenarios, name and path are usually the same. They don't
have to be though, and it's important to make sure we use the right one when
passing arguments to git commands, so change the tests to have different name
and path.
When doing a non-interactive rebase using a version of git earlier than 2.26, or
by explicitly calling `git -c rebase.backend=apply rebase`, lazygit can display
the pending todos by parsing the numbered patch files in `.git/rebase-apply/`.
Unfortunately, support for this has been broken for more than three years
because of the change in 682db77401 (the string literal "normal" should have
been changed to REBASE_MODE_NORMAL instead of REBASE_MODE_MERGING).
It's not an important bug since you can only get into this situation by doing a
rebase outside of lazygit, and then only with a pretty old git version or by
using very uncommon git options. So instead of fixing the bug, just remove the
code.
This makes it possible again to pass multiple arguments, for example
"--ff-only --autostash". This won't work correctly if you want to use
an argument that contains a space, but it's very unlikely that people
will want to do that, so I think this is good enough.
The additional branch head icon is more confusing than useful in this situation.
The update-ref entries show very clearly where the branch heads will go when
continuing the rebase; the information where the branch heads used to be before
the rebase is not really needed here, and just makes the display more confusing.
I'm not adding more tests here because the changes to the existing tests
demonstrate the change clearly enough.
The helix binary seems to be called "helix" on some distributions (e.g. Arch),
but "hx" on others (e.g. Fedora). Provide presets for both, so that
auto-detection from $EDITOR works.
A common workflow for me is to create a fixup commit from only some of my
current changes; to do that, I enter a file, stage a few hunks, and then want to
invoke ctrl-f to find the base commit for these changes. Currently I need to esc
back to the files panel in order to do that; it's more convenient to be able to
do this right from the staging panel.
This commit introduces a new feature to the commit view, allowing users
to filter commits based on the author's name or email address. Similar
to the existing path filtering functionality, accessible through <c-s>,
this feature allows users to filter the commit history by the currently
selected commit's author if the commit view is focused, or by typing in
the author's name or email address.
This feature adds an entry to the filtering menu, to provide users with
a familiar and intuitive experience
Calling "git reset" on the command line (without further arguments) defaults to
--mixed, which is reason enough to make it the default for us, too.
But I also find myself using --mixed more often than --soft. The main use case
for me is that I made a bunch of WIP commits, and want to turn them into real
commits when I'm done hacking. I select the last commit before the WIP commits
and reset to it, leaving all changes of all those commits in the working
directory. Since I want to start staging things from there, I prefer those
modifications to be unstaged at that point, which is what --mixed does.
For some lists it is useful to keep the same sort order when filtering (rather
than sorting by best match like we usually do). Add an optional function to
FilteredList to make this possible, and use it whenever we show lists of things
sorted by date (branches, stashes, reflog entries), as well as menu items
because this allows us to keep the section headers in the keybindings menu,
which is useful for understanding what you are looking at when filtering.
After #3283 we need to read more lines initially so that the scrollbar goes to
its minimal height of 1 for long diffs. Without this, it would start with a
height of 2 and then become smaller after you scroll down half the window
height.
This fixes two problems:
- each time the custom commands panel was opened, the history of commands would
be shown in reversed order compared to last time. (The reason is that
lo.Reverse modifies the slice in place rather than just returning a new,
reversed slice.)
- when executing a previous command again (either by typing it in again, or by
picking it from the history), it should move to the beginning of the history,
but didn't.
We fix this by storing the history in reversed order (as the user sees it in
the panel), this makes the logic simpler. We just have to prepend rather
than append newly added commands now.
While this is theoretically a breaking change, it's not worth bothering because
the order was wrong for existing users in 50% of the cases anyway.
Scenario:
- show the files of a commit, escape out of it again
- start an interactive rebase of a stack of branches, with the rebase.updateRefs
git config set to true
- select one of the update-ref todos
- trigger a refresh (either manually or by bringing lazygit's terminal window to
the front)
This results in an error message "fatal: ambiguous argument '': unknown revision
or path not in the working tree."
Fix this by putting another band-aid on the check for the commit files refresh.
This is the easiest way to fix the problem, but I don't think it's the best one.
We shouldn't be refreshing the commit files context at all if it isn't visible,
because it's pointless; there's no way to switch to it again except by calling
viewFiles again with a specific ref. But I'm too lazy too figure out how to do
that right now.
After discarding file changes from the commit, the was still referencing
these indexes as being part of the range select. The consequence was
needing to hit escape twice to exit commit files in some situations.
Canceling the range select after discarding changes fixes that.
I'm combining the delete single file case from `discard_old_file_change`
with the content of `discard_range_select` and calling that
`discard_old_file_changes`. Hopefully that cleans things up a little
bit.
This also adds a check that the custom patch is getting reset properly.
The waiting status shouldn't happen until after the user has responded
to the popup.
Since we're not giving a standalone prompt about clearing the patch, all
of the business in `discard` doesn't need to be in a function any more
It can be tedious after each cherry-pick opearation to clear the
selection by pressing escape in order for lazygit to stop displaying
info about copied commits. Also, it seems to be a rare case to
cherry-pick commits to more than one destination.
The simplest solution to address this issue is to clear the selection
upon paste.
The only exception is a merge conflict. Initially, I wanted to clear
selected commits in this scenario too. During a discussion we found out
that it may be convenient to have the copied commits still around.
Aborting the rebase and pasting the commits in the middle of a branch
can be a valid use case.
To do that, change the "Apply fixup commits" command to show a menu with the two
choices "in current branch" and "above the selected commit"; we make "in current
branch" the default, as it's the more useful one most of the time, even though
it is a breaking change for those who are used to "shift-S enter" meaning
"squash above selected".
Before this commit, we had pkg/integration/tests/submodule/add.go
failing with a panic. I'm pretty sure the issue is this: we're now
calling quite a few GetDisabledReason calls on each layout() call,
and if a background thread happens to update a model slice while
we're doing this, we can end up with a selection index that's now
out of bounds because it hasn't been clamped to match the new list
length.
Specifically, here we had the selected index being -1 (the list starts
empty and somehow the value is -1 in this case) and then the list
gets a new submodule so the length is now 1, but the list cursor
doesn't know about this so remains on the old value. Then we confirm
the length is greater than zero and try to get the selected submodule
and get an out of bounds error.
This commit fixes the issue by clamping the selected index whenever
we get the length of the list so that it stays in-sync. This is not
a perfect solution because the length can change at any time, but
it seems to reliably fix the test, and using mutexes didn't seem to
make a difference.
Note that we're swapping the order of IFileTree and IListCursor in
the file tree view model to ensure that the list cursor's Len()
method is called (which performs the clamping).
Also, comment from the PR:
This 'trait' pattern we're using is convenient but can lead to awkward
situations. In this case we have both the list view model and the
(embedded) list cursor with a Len() method. The list cursor Len()
method just calls the list view model Len() method. But I wanted
to make it that the list view model now calls ClampSelection() on the
list cursor whenever it obtains the length. This will cause an
infinite loop because ClampSelection() internally calls Len()
(which calls the list view model's Len() method which in turn
calls ClampSelection() again, etc).
The only reason we were passing the list view model into the list
cursor was to supply the length method, so now we're just doing
that directly, and letting the list view model delegate the Len()
call to the list cursor, which now itself calls ClampSelection.
As part of making lazygit more discoverable, there are certain keys which you almost certainly
need to press when you're in a given mode e.g. 'v' to paste commits when cherry-picking. This
commit prominently shows these keybinding suggestions alongside the others in the option view.
I'm using the same colours for these keybindings as is associated with the mode elsewhere e.g.
yellow for rebasing and cyan for cherry-picking. The cherry-picking one is a bit weird because
we also use cyan text to show loaders and app status at the bottom left so it may be confusing,
but I haven't personally found it awkward from having tested it out myself.
Previously we would render these options whenever a new context was activated, but now that we
need to re-render options whenever a mode changes, I'm instead rendering them on each screen
re-render (i.e. in the layout function). Given how cheap it is to render this text, I think
it's fine performance-wise.
This adds a bunch of tooltips to keybindings and updates some keybinding descriptions (i.e. labels).
It's in preparation for displaying more keybindings on-screen (in the bottom right of the screen),
and so due in part to laziness it shortens some descriptions so that we don't need to manage both
a short and long description (for on-screen vs in-menu). Nonetheless I've added a ShortDescription
field for when we do want to have both a short and long description.
You'll notice that some keybindings I deemed unworthy of the options view have longer descriptions,
because I could get away with it.
We're going to be adding tooltips to the cheatsheet to better explain what each actions
does. As such, we're switching to a table format rather than a list.
I'm also changing how the keys are represented, using a markdown approach rather than
an html approach
This is useful if you want to move a range of commits, so you select them, and
then realize it's better to do it in an interactive rebase. Pressing 'i'
preserves the range now.
This helps work around bugs in editors that may get confused about relative
paths (like nvim-remote, see https://github.com/neovim/neovim/issues/18519), and
shouldn't have any negative effect on others.
As part of this, you must now press enter on a merge conflict file
to focus the merge view; you can no longer press space and if you do
it will raise an error.
This changes GetRepoPaths() to pull information from `git rev-parse`
instead of effectively reimplementing git's logic for pathfinding. This
change fixes issues with bare repos, esp. versioned homedir use cases,
by aligning lazygit's path handling to what git itself does.
This change also enables lazygit to run from arbitrary subdirectories of
a repository, including correct handling of symlinks, including "deep"
symlinks into a repo, worktree, a repo's submodules, etc.
Integration tests are now resilient against unintended side effects from
the host's environment variables. Of necessity, $PATH and $TERM are the
only env vars allowed through now.
I prefer this because I almost never use sticky range selections. Also, this
fixes the issue that just clicking a line in a diff (without dragging) already
creates a range selection. It still does, technically, but it's no longer a
problem because a non-sticky one-line range selection behaves the same as a
non-range selection.
Previously if we marked a line with IsSelected() we would check if it was selected, but
we would not check if other lines were unexpectedly selected. Now, if you use IsSelected(),
we ensure that _only_ the lines you marked as such are the selected lines.
Something dumb that we're currently doing is expecting list items
to define an ID method which returns a string. We use that when copying
items to clipboard with ctrl+o and when getting a ref name for diffing.
This commit gets us a little deeper into that hole by explicitly requiring
list items to implement that method so that we can easily use the new
helper functions in list_controller_trait.go.
In future we need to just remove the whole ID thing entirely but I'm too
lazy to do that right now.
Fixes https://github.com/jesseduffield/lazygit/issues/3077
Show unstaged file names in default colour
Previously, we had the following rules:
* file names were in red when unstaged or partially staged
* directory names were in red if unstaged, yellow if partially staged,
and
green if fully staged
Red text on a black background can be hard to read, so instead I'm
changing it
so that unstaged files have their names in the default text colour.
I'm also making it so that partially staged files are in yellow, just
like how
partially staged directories are yellow (same deal with the commit files
view
when adding to a custom patch).
So the new rules are:
* unstaged files/directories use the default colour
* partially staged files/directories are in yellow
* fully staged files/directories are in green
I've also done a refactor on the code clean up some dead code from when
the file tree
outline was drawn with box characters, and I've made it so that the
indentation in
each line is handled inside the function that draws the line rather than
in the recursive
parent function. This makes it easier to experiment with things like
showing the file
status characters on the left edge of the view (admittedly after
experimenting with it,
I decided I didn't like it). Apologies for having a refactor and a
functional change
in the one commit but by the time I was done, I couldn't be bothered
going back and
retroactively splitting it into two halves.
Previously, we had the following rules:
* file names were in red when unstaged or partially staged
* directory names were in red if unstaged, yellow if partially staged, and
green if fully staged
Red text on a black background can be hard to read, so instead I'm changing it
so that unstaged files have their names in the default text colour.
I'm also making it so that partially staged files are in yellow, just like how
partially staged directories are yellow (same deal with the commit files view
when adding to a custom patch).
So the new rules are:
* unstaged files/directories use the default colour
* partially staged files/directories are in yellow
* fully staged files/directories are in green
I've also done a refactor on the code clean up some dead code from when the file tree
outline was drawn with box characters, and I've made it so that the indentation in
each line is handled inside the function that draws the line rather than in the recursive
parent function. This makes it easier to experiment with things like showing the file
status characters on the left edge of the view (admittedly after experimenting with it,
I decided I didn't like it). Apologies for having a refactor and a functional change
in the one commit but by the time I was done, I couldn't be bothered going back and
retroactively splitting it into two halves.
This wasn't necessary before, because the only available branch sorting option
was by recency, so the sort order couldn't change except by checking out
branches. Now, you can sort by committer date, so the branch order can change by
fetching; in this case it's important to keep the same branch selected. One
important use case is to rebase the checked-out branch onto master; you select
master, press "f" to fetch it (this can now change its position in the list),
and then press "r" to rebase. To make this work smoothly it's important to keep
master selected after pressing "f".
Often if a test fails and there's an unaknowledged toast message, that message will
explain why the test failed. Given that we don't display toast messages in
integration tests when they run (for reasons I can't recall right now), we need to
log it as part of the error message.
We don't need it there so no need to enable it.
I'm leaving the disabled reason checks there, even though they're now redundant,
because they're only one-liners and they communicate intent.
We want to show an error when the user tries to invoke an action that expects only
a single item to be selected.
We're using the GetDisabledReason field to enforce this (as well as DisabledReason
on menu items).
I've created a ListControllerTrait to store some shared convenience functions for this.
This requires us to change the 'v' keybinding for paste to something else,
now that 'v' is used globally for toggling range select. So I'm using
'shift+v' and I'm likewise changing 'c' to 'shift+c' for copying, so
that they're consistent.
We will need to clearly communicate this change in keybindings.
The only time we should call SetSelectedLineIdx is when we are happy for a
select range to be retained which means things like moving the selected line
index to top top/bottom or up/down a page as the user navigates.
But in every other case we should now call SetSelection because that will
set the selected index and cancel the range which is almost always what we
want.
This is the highest priority of the escape actions because it's the thing you're
most likely to want to do upon hitting escape if you have a range selected.
Applying this to the staging/patch-building views is tricky: if we want this logic
for when a range of lines is selected, we'll also need to apply it when a hunk
is selected too. I still think it's worth it though: I've often accidentally
escaped from the staging view when trying to cancel a range selection.
We're not fully standardising here: different contexts can store their range state however
they like. What we are standardising on is that now the view is always responsible for
highlighting the selected lines, meaning the context/controller needs to tell the view
where the range start is.
Two convenient benefits from this change:
1) we no longer need bespoke code in integration tests for asserting on selected lines because
we can just ask the view
2) line selection in staging/patch-building/merge-conflicts views now look the same as in
list views i.e. the highlight applies to the whole line (including trailing space)
I also noticed a bug with merge conflicts not rendering the selection on focus though I suspect
it wasn't a bug with any real consequences when the view wasn't displaying the selection.
I'm going to scrap the selectedRangeBgColor config and just let it use the single line
background color. Hopefully nobody cares, but there's really no need for an extra config.
This adds range select ability in two ways:
1) Sticky: like what we already have with the staging view i.e. press v then use arrow keys
2) Non-sticky: where you just use shift+up/down to expand the range
The state machine works like this:
(no range, press 'v') -> sticky range
(no range, press arrow) -> no range
(no range, press shift+arrow) -> nonsticky range
(sticky range, press 'v') -> no range
(sticky range, press arrow) -> sticky range
(sticky range, press shift+arrow) -> nonsticky range
(nonsticky range, press 'v') -> no range
(nonsticky range, press arrow) -> no range
(nonsticky range, press shift+arrow) -> nonsticky range
Previously we included all navigation keybindings from all views in the keybindings menu, meaning
if you pressed enter on 'next page' in the commits view, you'd end up triggering the action
in the sub-commits view.
Because we obtain disabled reasons after every action, we need to keep the code for doing so
super fast. As such, we should not be hitting the filesystem to get rebase state, instead
we should just get the cached state.
I feel like we should actually be using the cached state everywhere like we do with all
our other models if only for the sake of consistency.
A common issue I have is that I want to move a commit from the top of my branch
all the way down to the first commit on the branch. To do that, I need to navigate
down to the first commit on my branch, press 'e' to start an interactive rebase,
then navigate back up to the top of the branch, then move my commit back down to
the base. This is annoying.
Similarly annoying is moving the commit one-by-one without explicitly starting
an interactive rebase, because then each individual step is its own rebase which
takes a while in aggregate.
This PR allows you to press 'i' from the commits view to start an interactive
rebase from an 'appropriate' base. By appropriate, we mean that we want to start
from the HEAD and stop when we reach the first merge commit or commit on the main
branch. This may end up including more commits than you need, but it doesn't make
a difference.
For some bizarre reason `pkg/integration/tests/filter_by_path/cli_arg.go` is failing as of 8c716184 like so:
```
test_lazygit
Usage:
test_lazygit [git-arg]
Positional Variables:
git-arg Panel to focus upon opening lazygit. Accepted values (based on git terminology): status, branch, log, stash. Ignored if --filter arg is passed.
Flags:
-h --help Displays help with available flag, subcommand, and positional value parameters.
-p --path Path of git repo. (equivalent to --work-tree=<path> --git-dir=<path>/.git/)
-f --filter Path to filter on in `git log -- <path>`. When in filter mode, the commits, reflog, and stash are filtered based on the given path, and some operations are restricted
-v --version Print the current version
-d --debug Run in debug mode with logging (see --logs flag below). Use the LOG_LEVEL env var to set the log level (debug/info/warn/error) (default: false)
-l --logs Tail lazygit logs (intended to be used when `lazygit --debug` is called in a separate terminal tab)
-c --config Print the default config
-cd --print-config-dir Print the config directory
-ucd --use-config-dir override default config directory with provided directory
-w --work-tree equivalent of the --work-tree git argument
-g --git-dir equivalent of the --git-dir git argument
-ucf --use-config-file Comma separated list to custom config file(s)
Unknown arguments supplied: filterFile
```
where the CLI args are:
```
([]string) (len=5 cap=5) {
(string) (len=25) "/tmp/lazygit/test_lazygit",
(string) (len=6) "-debug",
(string) (len=108) "--use-config-dir=/Users/jesseduffieldduffield/repos/lazygit/test/_results/filter_by_path/cli_arg/used_config",
(string) (len=2) "-f",
(string) (len=10) "filterFile"
}
```
This appears to be a bug in flaggy itself. I've updated to the latest version but it still breaks. Bizarrely it works fine on CI and
only fails locally. Running lazygit locally with `lg -f pkg/gui/controllers/helpers/refresh_helper.go` it works fine. So I don't
know what's going on there. At any rate, I'm just going to get the test passing by passing `-f=filterFile` as a single argument.
We need to fetch our list of tests both outside of our test binary and within. We need
to get the list from within so that we can run the code that drives the test and runs
assertions. To get the list of tests we need to know where the root of the lazygit repo
is, given that the tests live in files under that root.
So far, we've used this GetLazyRootDirectory() function for that, but it assumes that
we're not in a test directory (it just looks for the first .git dir it can find). Because
we didn't want to properly fix this before, we've been setting the working directory of
the test command to the lazygit root, and using the --path CLI arg to override it when
the test itself ran. This was a terrible hack.
Now, we're passing the lazygit root directory as an env var to the integration test, so
that we can set the working directory to the actual path of the test repo; removing the
need to use the --path arg.
Git has a bug [1] whereby running multiple fetch commands at the same time
causes all of them to append their information to the .git/FETCH_HEAD file,
causing the next git pull that wants to use the information to become confused,
and show an error like "Cannot rebase onto multiple branches". This error would
occur when pressing "f" and "p" in quick succession in the files panel, but also
when pressing "p" while a background fetch happens to be running. One likely
situation for this is pressing "p" right after startup.
Since lazygit never uses the information written to .git/FETCH_HEAD, it's best
to avoid writing to it, which fixes the scenarios described above.
However, it doesn't fix the problem of repeatedly pressing "f" quickly on the
checked-out branch; since we call "git pull" in that case, the above fix doesn't
help there. We'll address this separately in another PR.
[1] See https://public-inbox.org/git/xmqqy1daffk8.fsf@gitster.g/ for more
information.
The algorithm works by blaming the deleted lines, so if a hunk contains only
added lines, we can only hope that it also belongs in the same commit. Warn the
user about this.
Note: the warning might be overly agressive, we'll have to see if this is
annoying. The reason is that it depends on the diff context size whether added
lines go into their own hunk or are grouped together with other added or deleted
lines into one hunk. However, our algorithm uses a diff context size of 0,
because that makes it easiest to parse the diff; this results in hunks having
only added lines more often than what the user sees. For example, moving a line
of code down by two lines will likely result in a single hunk for the user, but
in two hunks for our algorithm. On the other hand, being this strict makes the
warning consistent. We could consider using the user's diff context size in the
algorithm, but then it would depend on the current context size whether the
warning appears, which could be confusing. Plus, it would make the algorithm
quite a bit more complicated.
There are two possible fixes for this bug, and they differ in behavior when
rewording a commit. The one I chose here always splits at the first line feed,
which means that for an improperly formatted commit message such as this one:
This is a very long multi-line subject,
which you shouldn't really use in git.
And this is the body (we call it "description" in lazygit).
we split after the first line instead of after the first paragraph. This is
arguably not what the original author meant, but splitting after the first
paragraph doesn't really work well in lazygit, because we would try to put both
lines into the one-line subject field of the message panel, and you'd only see
the second and not even know that there are more.
The other potential fix would have been to join subject and description with two
line feeds instead of one in JoinCommitMessageAndDescription; this would have
fixed our bug in the same way, but would result in splitting the above message
after the second line instead of the first. I think that's worse, so I decided
for the first fix.
While we're at it, simplify the code a little bit; strings.Cut is documented to
return (s, "") when the separator is not found, so there's no need to do this on
our side.
We do have to trim spaces on the description now, to support the regular reword
case where subject and body are separated by a blank line.
SplitCommitMessageAndDescription splits at the first '\n\n' that it finds (if
there is one), which in this case is between the two paragraphs of the
description. This is wrong.
Use git log instead of git rev-list, this way we don't get a line "commit <sha>"
at the beginning that we then have to discard again.
The test TestGetCommitMsg is becoming a bit pointless now, since it just
compares that input and output are identical.
Without this it's not reliably possible to ask whether a given view is visible
by asking
windowHelper.TopViewInWindow(context.GetWindowName()) == context.GetView()
because there could be transient, invisible contexts after it in the Z order.
I guess it's a bit of a coincidence that this has never been a problem so far.
The output of the GetWindowDimensions function is hard to understand just by looking at it,
so I've added a helper function in the tests to render the window layout as text, so that
in order to create a new test you just come up with some args and paste the output as the
expected output.
This has the same downsides that any snapshot-based testing has: it's more brittle than
targeted assertions. But it is much easier to make sense of these snapshots than it is
to make sense of more fine-grained assertions, and I like the fact that these tests can
serve as documentation.
We are also removing the single-character padding on the left/right edges of the bottom
line because it's unnecessary
Unfortunately we need to create views for each spacer: it's not enough to just
layout the existing views with padding inbetween because gocui only renders
views meaning if there is no view in a given position, that position will just
render whatever was there previously (at least that's what I recall from talking
this through with Stefan: I could be way off).
Co-authored-by: Stefan Haller <stefan@haller-berlin.de>
It sounds like at some point we only showed a slash as the search prompt, but I
dug a bit through the history and couldn't find a state of the code where that
was the case. (shrug)
This PR captures the code coverage from our unit and integration tests. At the
moment it simply pushes the result to Codacy, a platform that assists with
improving code health. Right now the focus is just getting visibility but I want
to experiment with alerts on PRs when a PR causes a drop in code coverage.
To be clear: I'm not a dogmatist about this: I have no aspirations to get to
100% code coverage, and I don't consider lines-of-code-covered to be a perfect
metric, but it is a pretty good heuristic for how extensive your tests are.
The good news is that our coverage is actually pretty good which was a surprise
to me!
As a conflict of interest statement: I'm in Codacy's 'Pioneers' program which
provides funding and mentorship, and part of the arrangement is to use Codacy's
tooling on lazygit. This is something I'd have been happy to explore even
without being part of the program, and just like with any other static analysis
tool, we can tweak it to fit our use case and values.
## How we're capturing code coverage
This deserves its own section. Basically when you build the lazygit binary you
can specify that you want the binary to capture coverage information when it
runs. Then, if you run the binary with a GOCOVERDIR env var, it will write
coverage information to that directory before exiting.
It's a similar story with unit tests except with those you just specify the
directory inline via `-test.gocoverdir`.
We run both unit tests and integration tests separately in CI, _and_ we run them
parallel with different OS's and git versions. So I've got each step uploading
the coverage files as an artefact, and then in a separate step we combine all
the artefacts together and generate a combined coverage file, which we then
upload to codacy (but in future we can do other things with it like warn in a PR
if code coverage decreases too much).
Another caveat is that when running integration tests, not only do we want to
obtain code coverage from code executed by the test binary, we also want to
obtain code coverage from code executed by the test runner. Otherwise, for each
integration test you add, the setup code (which is run by the test runner, not
the test binary) will be considered un-covered and for a large setup step it may
appear that your PR _decreases_ coverage on net. Go doesn't easily let you
exclude directories from coverage reports so it's better to just track the
coverage from both the runner and the binary.
The binary expects a GOCOVERDIR env var but the test runner expects a
test.gocoverdir positional arg and if you pass the positional arg it will
internally overwrite GOCOVERDIR to some random temp directory and if you then
pass that to the test binary, it doesn't seem to actually write to it by the
time the test finishes. So to get around that we're using LAZYGIT_GOCOVERDIR and
then within the test runner we're mapping that to GOCOVERDIR before running the
test binary. So they both end up writing to the same directory. Coverage data
files are named to avoid conflicts, including something unique to the process,
so we don't need to worry about name collisions between the test runner and the
test binary's coverage files. We then merge the files together purely for the
sake of having fewer artefacts to upload.
## Misc
Initially I was able to have all the instances of '/tmp/code_coverage' confined
to the ci.yml which was good because it was all in one place but now it's spread
across ci.yml and scripts/run_integration_tests.sh and I don't feel great about
that but can't think of a way to make it cleaner.
I believe there's a use case for running scripts/run_integration_tests.sh
outside of CI (so that you can run tests against older git versions locally) so
I've made it that unless you pass the LAZYGIT_GOCOVERDIR env var to that script,
it skips all the code coverage stuff.
On a separate note: it seems that Go's coverage report is based on percentage of
statements executed, whereas codacy cares more about lines of code executed, so
codacy reports a higher percentage (e.g. 82%) than Go's own coverage report
(74%).
Situations where a view's width changes:
- changing screen modes
- enter staging or patch building
- resizing the terminal window
For the first of these we currently have special code to force a rerender, since
some views render different content depending on whether they are in full-screen
mode. We'll be able to remove that code now, since this new generic mechanism
takes care of that too.
But we will need this more general mechanism for cases where views truncate
their content to the view width; we'll add one example for that later in this
branch.
refreshWorktrees re-renders the branches view, because the branches view shows
worktrees against branches. This means that when both BRANCHES and WORKTREES are
requested to be refreshed, the branches view would be rendered twice in short
succession. This causes an ugly visual glitch when force-pushing a branch,
because when pushing is done, we would see the ↑4↓9 status come back from under
the Pushing status for a brief moment, to be replaced with a green checkmark a
moment later.
Fix this by including the worktree refresh in the branches refresh when both are
requested. This means that the two are no longer running in parallel for an
async refresh, but hopefully that's not so bad.
When pulling/pushing/fast-forwarding a branch, show this state in the branches
list for that branch for as long as the operation takes, to make it easier to
see when it's done (without having to stare at the status bar in the lower
left).
This will hopefully help with making these operations feel more predictable, now
that we no longer show a loader panel for them.
