On Wed, Jul 29, 2020 at 01:40:17PM -0700, Linus Torvalds wrote:
> > So really. Both from a technical _and_ a terminology angle, that
> > commit was just wrong.
>
> So here's the fix.
>
> The fix is effectively a revert of the patch (and a follow-up
> test-case patch), and instead just replaces the "master" nomenclature
> with "git_default_branch_name()".
>
> This patch actually removes more uses of "master" than it adds to git,
> and unlike the patches it replaces, actually removes it from other
> projects commit logs.
>
> Now whatever default branch name you choose really is immaterial.
I don't think that covers all cases.
If I set init.defaultBranch, then that only sets the default branch name
for new repositories I create with "git init". But I may still interact
with many repositories that use a different name (either "master", or
other names chosen by individual projects).
If you suppressed the output for both "master" and
git_default_branch_name(), that would cover more cases, at the expense
of occasionally doing the wrong thing when some project uses "master"
but doesn't mean for it to be a default branch (but that's presumably
rare). It would still do the wrong thing if you prefer "foo" as the
default branch name for your projects, but sometimes work on a project
that prefers "bar".
Users can set init.defaultBranch in individual repository config files
to get the right behavior, though probably it should be a separate
config variable (the reason it is init.defaultBranch and not
core.defaultBranch is that you may want different defaults for different
operations).
So I think a path forward is more like:
1. Add a new config option to shorten fmt-merge-msg's output when the
destination branch matches it (and this should perhaps not even be
a single name, but a set of globs, which supports more workflows).
Call it merge.suppressDest or something.
2. Optionally a repository created with "git init" could copy its
init.defaultBranch into merge.suppressDest. And likewise a clone
might copy the remote HEAD into that variable. I'm not sure if that
is worth doing or not, but it would restore the original behavior
for the most part.
-Peff
