mailmap: only look for .mailmap in work tree

mailmap: only look for .mailmap in work tree

When trying to find a .mailmap file, we will always look for it in the
current directory. This makes sense in a repository with a working tree,
since we'd always go to the toplevel directory at startup. But for a
bare repository, it can be confusing. With an option like --git-dir (or
$GIT_DIR in the environment), we don't chdir at all, and we'd read
.mailmap from whatever directory you happened to be in before starting
Git.

(Note that --git-dir without specifying a working tree historically
means "the current directory is the root of the working tree", but most
bare repositories will have core.bare set these days, meaning they will
realize there is no working tree at all).

The documentation for gitmailmap(5) says:

  If the file `.mailmap` exists at the toplevel of the repository[...]

which likewise reinforces the notion that we are looking in the working
tree.

This patch prevents us from looking for such a file when we're in a bare
repository. This does break something that used to work:

  cd bare.git
  git cat-file blob HEAD:.mailmap >.mailmap
  git shortlog

But that was never advertised in the documentation. And these days we
have mailmap.blob (which defaults to HEAD:.mailmap) to do the same thing
in a much cleaner way.

However, there's one more interesting case: we might not have a
repository at all! The git-shortlog command can be run with git-log
output fed on its stdin, and it will apply the mailmap. In that case, it
probably does make sense to read .mailmap from the current directory.
This patch will continue to do so.

That leads to one even weirder case: if you run git-shortlog to process
stdin, the input _could_ be from a different repository entirely. Should
we respect the in-tree .mailmap then? Probably yes. Whatever the source
of the input, if shortlog is running in a repository, the documentation
claims that we'd read the .mailmap from its top-level (and of course
it's reasonably likely that it _is_ from the same repo, and the user
just preferred to run git-log and git-shortlog separately for whatever
reason).

The included test covers these cases, and we now document the "no repo"
case explicitly.

We also add a test that confirms we find a top-level ".mailmap" even
when we start in a subdirectory of the working tree. This worked both
before and after this commit, but we never tested it explicitly (it
works because we always chdir to the top-level of the working tree if
there is one).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>