A recent refactor caused the API for ReadBranchConfig to change, resulting
in changes for its consumers. Additionally, there was a large refactor of
the tests associated with ReadBranchConfig and its consumers to gain
confidence in this refactor. This commit attempts to resolve the
conflicts between the refactor and this effort as well as massage the
changes introduced here to reflect the refactor.
The refactor PR can be found here: https://github.com/cli/cli/pull/10197
I'll note that there are still a few failing tests in status_test.go. I
haven't had a chance to fully grok while they are failing, yet, and
suspect that some insights from the original PR author may be helpful
here.
Full disclaimer: I haven't verified any of this is working locally yet.
My primary motivation is to get these new changes working together in a
manner that unblocks further iteration on this effort.
* trunk: (79 commits)
Enhance help docs on ext upgrade notices
chore: fix some function names in comment
Expand docs on cleaning extension update dir
Simplifying cleanExtensionUpdateDir logic
Separate logic for checking updates
Capture greater detail on updaterEnabled
Restore old error functionality of prSelectorForCurrentBranch
Change error handling on ReadBranchConfig to respect git Exit Codes
fix: add back colon that I removed
fix: actually read how MaxFunc work and simplify the code
fix: padded display
Collapse dryrun checks in ext bin upgrade
Bump github.com/mattn/go-colorable from 0.1.13 to 0.1.14
Rename test user in tests
Change pr number in test
Surface and handle error from ReadBranchConfig in parseCurrentBranch
Directly stub headBranchConfig in Test_tryDetermineTrackingRef
Refactor error handling in ReadBranchConfig to avoid panic
Refine error handling of ReadBranchConfig
Add test for empty BranchConfig in prSelectorForCurrentBranch
...
cmd.Output() will return an error when the git command ran successfully
but had no output. To handle this, we can check Stderr, as we expect it to
be populated for any ExitErrors or otherwise when there is a command
failure.
This allows for propagation of this error handling up the call chain, so
we are now returning errors if the call to git fails instead of just
handing off an empty BranchConfig and suppressing the errors.
Additionally, I've removed some more naked returns that I found in
pkg/cmd/pr/create.go createRun
I think I went too far with my previous refactor and am backing out of it.
Adding a private readBranchConfig method on the client wasn't providing
any real additional value, so I've put it back into ReadBranchConfig.
However, I think there is still value in having parseBranchConfig
(formerly createBranchConfig) as a separate util function, as it both
improves readability of ReadBranchConfig and makes parsing its purpose
easier through the bespoke tests for it.
When using a push.default = current triangular workflow, apart from
using @{push} to determine the remote branch name, we should also follow
the
1. branch.<name>.pushRemote
2. remote.pushDefault
3. branch.<name>.remote
...list to determine which remote Git pushes to.
When using a push.default = current central workflow [1], we should use
@{push} instead to locate the remote branch.
In fact, @{push} covers most cases in push.default = upstream too. The
branch.<name>.merge is probably only needed when using RemoteURL and
different remote / local branch names.
[1] https://github.com/tpope/vim-fugitive/issues/1172#issuecomment-522301607
Theoretically this should be clearer and more robust than the previous
version which had some custom loop logic while trying to parse newlines
and determine whether it had reached a new commit entry by trying to parse
a git sha. This would not have worked correctly if a commit body contained
a sha on a new line.
Since this is guarded by the line starting with a git sha, it must
be a line of the form 'sha,title,body', so there can never be only
two entries since the Split fn will produce an empty string in the
last spot in the case of a missing body.
