An existing "push-*" bookmark should usually be in position. If it wasn't
because of e.g split, I think the user should be aware of that and take an
explicit action.
Follows up on https://github.com/jj-vcs/jj/pull/5698 and
https://github.com/jj-vcs/jj/pull/6236.
The behavior of `--named` in corner cases differs. This wasn't detected
in #5698 because of a bug in the test that was fixed in #6236. This
commit better illustrates the more important case where the behavior is
the same, and the less important case where it differs.
It's deprecated since 1aad25042029 "Shorten the git push branch when possible
using the short change ID hash" (2023-01-12). I don't think we need the
fallback. I also removed the check for ambiguous prefixes as I believe it
wouldn't practically matter. If needed, we can add a check for existing bookmark
pointing to different commit. We can also make it templated with default
"'push-' ++ change_id.shortest(12)".
We've been finding that a lot of bug reports on `jj git push` come from
sub-standard error reporting on the reasons the failure happens.
It can come from a number of places:
- hook failure
- remote branch protection
- git config
This commit forwards the reason as explained by the ouptut of git push
to help users figure out what is happening.
We need to report more complicated errors on push.
Firstly, we can have a mix of unexpected ref locations and remote
rejections. We should report both at the same time.
Second, git gives us a reason for why a push failed.
For this to work, it's relevant to refactor the current error reporting
path to allow us to inject this information.
The git remote sideband adds a dummy suffix of 8 spaces to attempt to clear
any leftover data. This is done to help with cases where the line is
rewritten.
However, a common option in a lot of editors removes trailing whitespace.
This means that anyone with that option that opens this file would make the
following snapshot fail. Using the insta filter here normalizes the
output.
When we push a ref to a git remote, we use --force-with-lease
Our understanding was that this could fail iff the expected location of
the ref on the remote was not that of the local tracking ref
However, if the ref is from a protected branch (e.g., main) it will be
rejected by the remote for a different reason.
This commit solves this, by detecting this difference.
I'm going to add "[EOF]" marker to test that command output is terminated by
newline char. This patch ensures that callers who expect a raw output string
would never be affected by any normalization passes.
Some common normalization functions are extracted as CommandOutputString
methods.
Although this behaviour is accepted by git, it's a degenerate case.
Especially because we implicitely rely on being able to parse out the
remote from the refname (i.e., `refs/remotes/<remote>/<branch>`).
Branches can have forward slashes, but if remotes can also have them,
parsing the refname becomes ambiguous, and a pain. Especially because it
would be totally legal to have a branch "c" on remote "a/b" and a branch
"b" on remote "a".
Fixes#5731
Forgetting remote bookmarks can lead to surprising behavior since it
causes the repo state to become out-of-sync with the remote until the
next `jj git fetch`. Untracking the bookmarks should be a simpler and
more intuitive default behavior. The old behavior is still available
with the `--include-remotes` flag.
I also changed the displayed number of forgotten branches. Previously
when forgetting "bookmark", "bookmark@remote", and "bookmark@git" it
would display `Forgot 1 bookmarks`, but I think this would be confusing
with the new flag since the user might think that `--include-remotes`
didn't work. Now it shows separate `Forgot N local bookmarks` and
`Forgot M remote bookmarks` messages when applicable.
With this change a warning is shown if the user does not explicitly specify the target revision, but the behavior is unchanged (it still defaults to the working copy).
In the future the warning will be turned into an error. In other words, it will be required to specify target revision.
The bulk of the changes here are to prepare tests for the upcoming change, to make the transition easier.
For additional details please see:
* https://github.com/jj-vcs/jj/issues/5374
* https://github.com/jj-vcs/jj/discussions/5363
Given the previously‐stated intention of making this default
for the 0.27 release, prepare for that decision ahead of time by
enabling subprocessing by default on trunk. This will help surface
any regressions and workflow incompatibilities and therefore give
us more information to decide whether to keep or revert this commit,
without inconveniencing any users who haven’t already opted in to
the bleeding edge.
Please feel free to revert without hesitation if any major issues
arise; this is not intended as a strong commitment to enable this
option for the next stable release if it turns out to not be ready. In
that case, it’s better that we learn that early on in the cycle,
rather than having to revert at the last minute or, worse, cutting
a stable release that we later find contains a serious regression.
Currently, the Git subprocess tests only work on Linux due to a default
path being used for the `git` executable when `$PATH` is unset. This can
break if Git isn't installed at the expected path. Also, I believe it is
currently necessary to set the `$TEST_GIT_EXECUTABLE_PATH` environment
variable on Windows for tests to pass. Instead, we should use the user's
`$PATH` to locate the `git` executable, as well as any other executables
that are needed. This also makes `$TEST_GIT_EXECUTABLE_PATH` no longer
necessary, so it can be removed.
This goes against our rule that we shouldn't add config knob that changes the
command behavior, but I don't have any other idea to work around the problem.
Apparently, there are two parties, one who always wants to push new bookmarks,
and the other who mildly prefers to push&track new bookmarks explicitly.
Perhaps, for the former, creation of bookmarks means that the target branches
are marked to be pushed.
The added flag is a simple boolean. "non-tracking-only" behavior #5173 could be
implemented, but I don't want to complicate things. It's a failed attempt to
address the issue without introducing config knob.
Closes#5094Closes#5173
Since source.contains(':') can't be used for non-local path detection on
Windows, we now use gix::url for parsing. It might be stricter, but I assume it
would be more reliable.
Closes#4188
The working-copy revision is usually the latest commit, but it's not always
true. This patch ensures that the wc branch is emitted first so the graph node
order is less dependent on rewrites.
