Re: git diff bug?

From: Junio C Hamano <gitster@pobox.com>
Date: Jun 6, 2013, 22:17 UTC
Message-ID: <7v4ndayjdt.fsf@alter.siamese.dyndns.org>
In-Reply-To: <51B10236.2030009@ensimag.fr>

Célestin Matte <celestin.matte@ensimag.fr> writes:

Show 16 quoted lines

> Le 06/06/2013 23:26, Sarma Tangirala a écrit :
>> Hello All,
>> 
>> If I did 'git diff HEAD^..HEAD -- file' should git not report some
>> kind of warning if it could not match the file? For example, if 'file'
>> were infact 'dir/file' and 'file' were unique, would it not be a good
>> idea to report that in the present working directory 'file' were not
>> found but 'dir/file' were a match?
>
> I don't know any program doing such a thing, and I don't think it is the
> role of the program to predict which file the user actually wanted to
> provide in the command line.
> That would imply looking for files with the same name or a close name in
> the current directory and its subdirectories - and maybe even in the
> superdirectory? It is hard to decide when you have to stop looking for
> the file.

The parameters after "--" are pathspecs, which is a set of patterns the paths discovered by the operation (in this case "diff" that finds paths in HEAD^ and HEAD) are matched against. They are used to filter out uninteresting paths.

If HEAD^ and HEAD does not have anything that match the given pattern (in this case, literal four-letter string "file"), the set of interesting paths may become empty and that is perfectly normal.

So this is working as designed.

Having said that, we do detect typo by noticing when a pathspec did not find _any_ path that matched it in some front-end Porcelain commands, e.g.

	$ git add 'foo*'
        fatal: pathspec 'foo*' did not match any files

It is unreasonable to do the same in "git log old..new -- path" and error out when the pathspec does not match, because it is normal for some revisions to have path while some other revisions to lack it.

But for a two-endpoint diff Porcelain (not the plumbing diff-files, diff-index and diff-tree), I do not think it is particularly a bad idea to add such a "typo-detection" feature.