On 11/05/2017 07:17 AM, Junio C Hamano wrote:
> Junio C Hamano <gits...@pobox.com> writes:
>> Rafael Ascensão <rafa.al...@gmail.com> writes:
>> ...
>>> Because changing the default behavior of that function has
>>> implications on multiple commands which I think shouldn't change. But
>>> at the same time, would be nice to have the logic that deals with
>>> glob-ref patterns all in one place.
>>>
>>> What's the sane way to do this?
>>
>> Learn to type "--decorate-refs="refs/heads/[m]aster", and not twewak
>> the code at all, perhaps.  The users of existing "with no globbing,
>> /* is appended" interface are already used to that way and they do
>> not have to learn a new and inconsistent interface.
>>
>> After all, "I only want to see 'git log' output with 'master'
>> decorated" (i.e. not specifying "this class of refs I can glob by
>> using the naming convention I am using" and instead enumerating the
>> ones you care about) does not sound like a sensible thing people
>> often want to do, so making it follow the other codepath so that
>> people can say "refs/tags" to get "refs/tags/*", while still allowing
>> such a rare but specific and exact one possible, may not sound too
>> bad to me.
> 
> Having said all that, I can imagine another way out might be to
> change the behaviour of this "normalize" thing to add two patterns,
> the original pattern in addition to the original pattern plus "/*",
> when it sees a pattern without any glob.  Many users who relied on
> the current behaviour fed "refs/tags" knowing that it will match
> everything under "refs/tags" i.e. "refs/tags/*", and they cannot
> have a ref that is exactly "refs/tags", so adding the original
> pattern without an extra trailing "/*" would not hurt them.  And
> this will allow you to say "refs/heads/master" when you know you
> want that exact ref, and in such a repository where that original
> pattern without trailing "/*" would be useful, because you cannot
> have "refs/heads/master/one" at the same time, having an extra
> pattern that is the original plus "/*" would not hurt you, either.
> 
> This however needs a bit of thought to see if there are corner cases
> that may result in unexpected and unwanted fallout, and something I
> am reluctant to declare unilaterally that it is a better way to go.

There's some glob-matching code (somewhere? I don't know if it's allowed
everywhere) that allows "**" to mean "zero or one path components. If
"refs/tags" were massaged to be "refs/tags/**", then it would match not only

    refs/tags
    refs/tags/foo

but also

    refs/tags/foo/bar

, which is probably another thing that the user would expect to see.

There's at least some precedent for this kind of expansion: `git
for-each-ref refs/remotes` lists *all* references under that prefix,
even if they have multiple levels.

Michael

Reply via email to