Re: git archive and glob pathspecs

[Peter Wu <peter@lekensteyn.nl>]

On Wednesday 03 September 2014 13:21:06 Duy Nguyen wrote:
> On Wed, Sep 3, 2014 at 5:17 AM, Peter Wu <peter@lekensteyn.nl> wrote:
> > Hi,
> >
> > The `git archive` seems to accept a pathspec judging from the error message (git
> > version 2.1.0):
> >
> >     git archive HEAD -- :x
> >     fatal: pathspec 'x' did not match any files
> >
> > When I try to use deeper glob specs however, it throws an error (this also
> > happens if I use `:(glob)**/Makefile`, tested in the git source tree):
> >
> >     $ git archive HEAD -- ':(glob)*/Makefile'
> >     fatal: pathspec '*/Makefile' did not match any files
> >
> > Strange enough, command `git log -- ':(glob)*/Makefile'` works. Any idea what is
> > wrong?
> 
> There may be something wrong. This patch seems to make it work for me,
> but it includes lots of empty directories. I'll have a closer look
> later (btw it's surprising that negative pathspec works too..)

I can confirm that this patch shows Makefile's, but also includes a lot of empty
directories.

As for why this happens, my guess is that write_archive_entries() recurses the
full tree and adds every encountered directory (via read_tree_1, via
write_archive_entry()).

To fix this, write_archive (write_tar_archive, etc.) should be taught to handle
glob patterns, or parse_pathspec should expand globs (and then
parse_pathspec_arg might have to validate the remaining patterns).

Kind regards,
Peter

> diff --git a/archive.c b/archive.c
> index 3fc0fb2..a5be58d 100644
> --- a/archive.c
> +++ b/archive.c
> @@ -221,6 +221,7 @@ static int path_exists(struct tree *tree, const char *path)
>   int ret;
> 
>   parse_pathspec(&pathspec, 0, 0, "", paths);
> + pathspec.recursive = 1;
>   ret = read_tree_recursive(tree, "", 0, 0, &pathspec, reject_entry, NULL);
>   free_pathspec(&pathspec);
>   return ret != 0;
> @@ -237,6 +238,7 @@ static void parse_pathspec_arg(const char **pathspec,
>   parse_pathspec(&ar_args->pathspec, 0,
>         PATHSPEC_PREFER_FULL,
>         "", pathspec);
> + ar_args->pathspec.recursive = 1;
>   if (pathspec) {
>   while (*pathspec) {
>   if (**pathspec && !path_exists(ar_args->tree, *pathspec))