On Sat, Jul 30, 2022 at 12:53 PM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> And "git symbolic-ref" is perfectly happy to take complete garbage
> input. There seems to be no advantage over using that silly "echo"
> model.

Side note: it looks like that patch may break a test.

And that test most definitely *should* be broken.

t3200-branch.sh does

      git symbolic-ref refs/heads/dangling-symref nowhere

which really depends on that whole "git symbolic-ref does no sanity
checking at all".

In fact, it seems to depend particularly on the fact that for non-HEAD
refs, it does even *less* sanity checking, and doesn't even check that
the ref starts with a "refs/"

There is also t4202-log, which wants to test that "git log" reacts
well to a bad ref. But now that "git symbolic-ref" refuses to create
such a bad ref, that test fails.

Anyway, here's a slightly updated patch that just fixes that test that
depended on not just a dangling symref, but an *invalid* dangling
symref. And it changes t4202-log to use "echo" to create the bad ref
instead. Which is what the previous test did too, to create the bogus
hash.

Again - this is such a low-level plumbing thing that maybe nobody
cares, but it just struck me as a bad idea to have these kinds of
maintenance commands that can be used to just mess up your repository.
And if you have a bare repo, this really does look like the command
that *should* be used to change HEAD, since it's not about "git
checkout"

                  Linus