From: Junio C Hamano <gitster@pobox.com>
Date: Thu, 24 Feb 2011 17:09:48 GMT
Subject: Re: [PATCH] completion: complete "git diff ...branc<TAB>"
Message-ID: <7vfwrd2wbn.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20110224151328.GB14720@neumann>

SZEDER Gábor <szeder@ira.uka.de> writes:

> There is this 'ref:file' notation (as in 'git show master:README', but
> I don't know the proper term for it), which is understood by
> __git_complete_file(), and can be useful for 'git diff', e.g. to
> compare a file in two branches when the file was renamed in between:
>
>   git diff branchA:old branchB:new
>
> However, __git_complete_revlist() doesn't understand this notation,

If that is the case, I wonder if complete-revlist should learn it as well.
The "<tree>:<path>" notation is just one of the "extended SHA-1
expressions" to name an entry in a tree object and should be usable
anywhere an object name is expected to occur.

The users of complete-revlist seems to be "log" and their friends, and
they don't expect arbitrary object names; they are only interested in
commit objects and tags and then paths.  Teaching <tree>:<path> to it may
make "git log HEAD:bar<TAB>" to complete to "git log HEAD:bar/boz" that
would be a nonsense, but anything that the original "git log HEAD:bar"
would complete would be a nonsense to begin with (as soon as you said
"HEAD:<something to follow>", you are naming either a blob or a tree
object, and you cannot start log from them), so I don't see a harm done in
doing so.

IOW, shouldn't there be just one unified complete-rev-then-path that is a
superset of the current complete-revlist and complete-file that all the
current users of complete-revlist and complete-file should use?