This is the expected behavior, and also how `db/run-or-hide-shell` works. The
implementation is structured differently, though … because I reinvented the
implementation of `db/run-or-hide-eshell` instead of generalizing the one of
`db/run-or-hide-shell`. Ah, anyway …
When checking whether we are in a git repository or not, remove all symbolic
links in the current path before looking for a .git in the current path;
otherwise, symbolic links might mislead `locate-dominating-file` to find some
.git directory which is only reachable via symbolic links, which subsequently is
ignored by git (rightly so!) and thus yields errors.
Also, catch the case that the current git repository is not initialized yet by
checking the return value of `git rev-parse --abbrev-ref HEAD`.
To determine whether a merge is in progress, do not check in the current
directory for .git/MERGE_HEAD, but do this in the repository directory.
When computing `base-dir`, take the whole non-directory part, not just the
basename; otherwise directories like .emacs.d will get the final `.d` get
stripped off.
This allows to see of which git repository the current working directory is part
of.
This is also included in Howard Abrams configuration, but he instead determines
the name of the current git repository by looking at `remote.origin.url`.
However, my upstream branch is not always called `origin`, and sometimes there
is no upstream repository at all. Using the name of the current directory (not
the whole path, though) instead seems to be a good compromise from my point of
view.