Re: [GSoC][PATCH] submodule: port submodule subcommand 'add' from shell to C

From: Shourya Shukla <shouryashukla.oo@gmail.com>
Date: Aug 26, 2020, 09:15 UTC
Message-ID: <20200826091502.GA29471@konoha>
In-Reply-To: <xmqq8se36gev.fsf@gitster.c.googlers.com>

On 24/08 11:35, Junio C Hamano wrote:

Show 38 quoted lines

> Shourya Shukla <shouryashukla.oo@gmail.com> writes:
> 
> > 	if test -z "$force"
> > 	then
> > 		git ls-files --error-unmatch "$sm_path" > /dev/null 2>&1 &&
> > 		die "$(eval_gettext "'\$sm_path' already exists in the index")"
> > 	else
> > 		git ls-files -s "$sm_path" | sane_grep -v "^160000" > /dev/null 2>&1 &&
> > 		die "$(eval_gettext "'\$sm_path' already exists in the index and is not a submodule")"
> > 	fi
> 
> Hmph.  So,
> 
>  - if we are not being 'force'd, we see if there is anything in the
>    index for the path and error out, whether it is a gitlink or not.
> 
>  - if there is 'force' option, we see what the given path is in the
>    index, and if it is already a gitlink, then die.  That sort of
>    makes sense, as long as the remainder of the code deals with the
>    path that is not a submodule in a sensible way.
> 
> > This is what I have done in C:
> >
> > 	if (!force) {
> > 		if (is_directory(path) && submodule_from_path(the_repository, &null_oid, path))
> > 			die(_("'%s' already exists in the index"), path);
> 
> The shell version would error out with anything in the index, so I'd
> expect that a faithful conversion would not call is_directory() nor
> submodule_from_path() at all---it would just look path up in the_index
> and complains if anything is found.  For example, the quoted part in
> the original above is what gives the error message when I do
> 
> 	$ git submodule add ./Makefile
> 	'Makefile' already exists in the index.
> 
> I think.  And the above code won't trigger the "already exists" at
> all because 'path' is not a directory.

Alright. That is correct. I tried to use a multitude of functions but did not find luck with any of them. The functions I tried:

    - index_path() to check if the path is in the index. For some
      reason, it switched to the 'default' case and return the
      'unsupported file type' error.

    - A combination of doing an OR with index_file_exists() and
      index_dir_exists(). Still no luck. t7406.43 fails.

    - Using index_name_pos() along with the above two functions. Again a
      failure in the same test.

I feel that index_name_pos() should suffice this task but it fails in t7406.43. The SM is in index since 'git ls-files --error-unmatch s1' does return 's1' (s1 is the submodule). What am I missing here?

Show 12 quoted lines

> > 	} else {
> > 		int err;
> > 		if (index_name_pos(&the_index, path, strlen(path)) >= 0 &&
> > 		    !is_submodule_populated_gently(path, &err))
> > 			die(_("'%s' already exists in the index and is not a "
> > 			      "submodule"), path);
>
> Likewise.  The above does much more than the original.
>
> The original was checking if the found cache entry has 160000 mode
> bit, so the second test would not be is_submodule_populated_gently()
> but more like !S_ISGITLINK(ce->ce_mode)

Using this results in failure of t7506.[33-40]. I implemented this in two ways:

    1. Use stat() to initialise the stat st corresponding to the 'path'.
       Then do a '!S_ISGITLINK(st.st_mode)'.

    2. Run a for loop:
		for (i = 0; i < active_nr; i++) {
		const struct cache_entry *ce = active_cache[i];

		if (index_name_pos(&the_index, path, strlen(path)) >= 0 &&
		    !S_ISGITLINK(ce->ce_mode))
			die(_("'%s' already exists in the index and is not a "
			      "submodule"), path);
        }

        Still the tests failed. What is meant by 'active_nr' BTW? I am
        not aware of this term.

Where am I going wrong for both the if-cases?