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

From: Kaartic Sivaraam <kaartic.sivaraam@gmail.com>
Date: Sep 1, 2020, 20:35 UTC
Message-ID: <31e40c63bbac03d261ac6f46a0d2f6ae90a21038.camel@gmail.com>
In-Reply-To: <20200831130448.GA119147@konoha>

On Mon, 2020-08-31 at 18:34 +0530, Shourya Shukla wrote:

Show 26 quoted lines

> On 31/08 01:28, Kaartic Sivaraam wrote:
> 
> This is what I have done finally:
> ---
> 	if (read_cache() < 0)
> 		die(_("index file corrupt"));
> 
> 	if (!force) {
> 		if (cache_file_exists(path, strlen(path), ignore_case) ||
> 		    cache_dir_exists(path, strlen(path)))
> 			die(_("'%s' already exists in the index"), path);
> 	} else {
> 		int cache_pos = cache_name_pos(path, strlen(path));
> 		struct cache_entry *ce = the_index.cache[cache_pos];
> 		if (cache_pos >= 0 && !S_ISGITLINK(ce->ce_mode))
> 			die(_("'%s' already exists in the index and is not a "
> 			      "submodule"), path);
> 	}
> ---
> 
> I did not put the 'cache_pos >= 0' at the start since I thought that it
> will unnecessarily increase an indentation level. Since we are using
> 'cache_{file,dir}_exists' in the first check and 'cache_name_pos()' in
> the second, the placement of check at another indentation level would be
> unnecessary. What do you think about this?
>

Interestingly. 'cache_dir_exists' seems to work as expected only when the global ignore_case whose value seems to depend on core.ignorecase. So, we can't just rely on 'cache_dir_exists to identify a directory that has tracked contents. Apparently, the 'directory_exists_in_index' in 'dir.c' seems to have the code that we want here (which is also the only user of 'index_dir_exists'; the function for which 'cache_dir_exists' is a convenience wrapper.

The best idea I could think of is to expose that method and re-use it here. Given that my kowledge about index and caching is primitive, I'm not sure if there's a better approach. If others have a better idea for handling this directory case, do enlighten us.

Show 32 quoted lines

> > This is more close to what the shell version did but misses one case
> > which might or might not be covered by the test suite[1]. The case when
> > path is a directory that has tracked contents. In the shell version we
> > would get:
> > 
> >    $ git submodule add ../git-crypt/ builtin
> >    'builtin' already exists in the index
> >    $ git submodule add --force ../git-crypt/ builtin
> >    'builtin' already exists in the index and is not a submodule
> > 
> >    In the C version with the above snippet we get:
> > 
> >    $ git submodule add --force ../git-crypt/ builtin
> >    fatal: 'builtin' does not have a commit checked out
> >    $ git submodule add ../git-crypt/ builtin
> >    fatal: 'builtin' does not have a commit checked out
> > 
> >    That's not appropriate and should be fixed. I believe we could do
> >    something with `cache_dir_exists` to fix this.
> > 
> > 
> >    Footnote
> >    ===
> > 
> >    [1]: If it's not covered already, it might be a good idea to add a test
> >    for the above case.
> 
> Like Junio said, we do not care if it is a file or a directory of any
> sorts, we will give the error if it already exists. Therefore, even if
> it is an untracked or a tracked one, it should not matter to us. Hence
> testing for it may not be necessary is what I feel. Why should we test
> it?

I'm guessing you misunderstood. A few things:

- We only care about tracked contents for the case in hand.

- Identifying whether a given path corresponds to a directory
  which has tracked contents is tricky. Neither 'cache_name_pos'
  nor 'cache_file_exists' handle this. 'cache_dir_exists' is also
  not very useful as mentioned above.

So, we do have to take care when handling that case as Junio pointed out.

-- 
Sivaraam