The parameter *is* being declared with the redefinition of the
function, but not in its initial form, which gets you a warning
about the parameter being defined globally. This fixes it.
This is related to 33405. Turns out that not all other calls to
VCS_INFO_set are okay: With recent zsh versions the early call to that
function with the '-preinit-' argument causes a warning like this:
VCS_INFO_set:9: math parameter maxexports created globally
in function VCS_INFO_set
This fixes it.
Mikael informs me on IRC, that in new versions of git (he used 2.6.1)
where the "am" subcommand is now a builtin, a file that is used by the
git backend of vcs_info (namely .git/rebase-apply/msg-clean) is not
available anymore, leading to an annoying error message:
VCS_INFO_get_data_git:232: no such file or directory: .git/rebase-apply/msg-clean
This patch checks for the availabiliy of the file before using it,
and adjusts the value of the dependant values accordingly.
Before this patch, $gitbranch would be set to empty, which caused
VCS_INFO_get_data_git to early out with a failure status¹, consequently
$vcs_info_msg_0_ would be empty.
¹ via the 'if [[ -z ]]' block around line 170.
This sets the %b expando to the hash of the before-the-merge HEAD, rather
than to the literal string "detached HEAD". That hash is already available
via the gen-applied-string hook.
When pressing Ctrl-C after `git am`, only `last` exists in
`.git/rebase-apply/`, which is empty.
This patch fixes it to fall back to "no patch applied" then.
This shows, during 'git merge', the revision hashes of the "remote" head
(the one that will become second parent of the commit) in the %m expando.
Review-by: Frank Terbeck
When running git rebase -m and a conflict occurs, the git-rebase-todo
file is not present. This leads to an error from grep every time the
shell prompt is printed when vcs_info is enabled. Avoid this message by
checking if the file exists before trying to grep it.
When patches are applied, let quilt use .pc without forcing the
patch directory, this will fix the unapplied detection when being in
subdir.
When no patches are applied, use zstyle quilt-patch-dir then
QUILT_PATCHES then "patches" for path to search for patches.
Note: prefer setting quilt-patch-dir rather than QUILT_PATCHES for
absolute path because when patches are applied, quilt unapplied will
not return the correct list (i.e. the whole list rather that the one
specified by .pc/.quilt_series).
git-am also uses .git/rebase-apply for patch list but
the file original-commit does not exist (as no commit exist).
This patch handles both git rebase and git am. Also:
- get the first line (rather than the first char) when the message
contains only one line;
- remove unused function (ironically that should have been used here).
This will detect changes to submodules from the superproject's
perspective, e.g. after `git rm submodule`.
>From GIT-DIFF-INDEX(1)/GIT-DIFF(1):
Using "dirty" ignores all changes to the work tree of submodules,
only changes to the commits stored in the superproject are shown
(this was the behavior until 1.7.0).
Since a rebase contains a list of patches to re-apply, re-use the
facility for stgit to have the same mechanism.
The patch list given to the gen-{un,}applied-string hooks is an array
with the sha1 and the subject of the commit. On rebase merge, the
applied patches prior to current contains only a number and "?".
In empty repositories, HEAD is an unresolvable symbolic ref. Start computing
stagedstr/unstagedstr in that case; for the former, use a different method
than the non-empty-repository case.
The previous code would fail to detect the wcroot with Subversion 1.7+
when cwd is at least two levels below the root (i.e., ../../.svn exists
and ../.svn doesn't), and would then pass to the hook the revision and
basename of cwd rather than of the wcroot.
Previously, the value of the wc root would be used. In Subversion,
it makes more sense to use the revision of cwd, since all commands
(e.g., 'svn ci', 'svnversion') operate only on cwd and below, not on
wcroot and below.
This makes %b expand to a refname rather than a sha1 when HEAD is detached but
happens to match some ref (branch, tag, etc). The resulting output will
typically contain a slash (e.g., "tags/v1.0.2", "heads/mybranch"), which helps
distinguish it from the output in the "HEAD is a symbolic ref" case.
Guilt uses the same internal directory for keeping state as stgit, but
it doesn't use the same files (not surprisingly). This caused error
messages due to missing files.
This fixes that by making the "stgit-active?" test stricter.
Reported-by: Axel Beckert <abe@debian.org>
