Re: [GSoC][PATCH v5 1/3] submodule: fix buggy $path and $sm_path variable's value

From: Stefan Beller <sbeller@google.com>
Date: May 30, 2017, 23:29 UTC
Message-ID: <CAGZ79kYMA6Me_ZnBZQitW7ZSJ0kfvb-LPnH=1gTwhRN-KOe5pA@mail.gmail.com>
In-Reply-To: <56bcf006-06f1-1851-87de-5697f988cb08@ramsayjones.plus.com>

On Tue, May 30, 2017 at 4:07 PM, Ramsay Jones <ramsay@ramsayjones.plus.com> wrote:

Show 29 quoted lines

>
>
> On 30/05/17 22:53, Stefan Beller wrote:
>> On Fri, May 26, 2017 at 6:10 PM, Ramsay Jones
>> <ramsay@ramsayjones.plus.com> wrote:
>>> On 26/05/17 18:07, Stefan Beller wrote:
>>>> On Fri, May 26, 2017 at 9:31 AM, Ramsay Jones
>>>> <ramsay@ramsayjones.plus.com> wrote:
>
>>> Back in 2012, the submodule list was generated by filtering the
>>> output of 'git ls-files --error-unmatch --stage --'; but I don't
>>> recall if (at that time) git-ls-files required being at the top
>>> of the working tree, or if it would execute fine in a sub-directory.
>>> So, it's possible that the documentation of $path was wrong all along.
>>> ;-)
>>>
>>> At that time, by definition, $path == $sm_path. However, you know this
>>> stuff much better than me (I don't use git-submodule), so ...
>>
>> Don't take that stance. Sometimes I shoot quickly from the hip without
>> considering consequences (Figuratively).
>>
>> In a foreach command I can see value both in the 'displaypath'
>> (what sm_path would become here) and the 'submodule path'
>> from the superproject. The naming of 'sm_path' hints at that it ought
>> to be the 'submodule path'.
>
> Well, since I introduced it, I can confidently proclaim that it is
> indeed the 'submodule path'. :-D

ok. :)

> As I said above, I can't remember how git-ls-files worked back then,
> but it seems that I thought of it as the path to the submodule from
> the root of the working tree. Again, by definition, $sm_path == $path
> (as documented). Of course, that may have changed since then.

Documented in 64394e3 (git-submodule.sh: Don't use $path variable in eval_gettext string, by yourself)

What I intended to say above was "documented to the end user", and I do not count our commit messages as such. The end user facing documentation only talks about path, not mentioning sm_path.

Show 18 quoted lines

>>>> With this patch it becomes less non-sensey and could be documented as:
>>>>
>>>>     $sm_path is the relative path from the current working directory
>>>>     to the submodule (ignoring relations to the superproject or nesting
>>>>     of submodules).
>>>
>>> OK.
>>>
>>>>                      This documentation also fits into the narrative of
>>>>     the test in t7407.
>>>
>>> Hmm, does it?
>>
>> After rereading that test, I would think so?
>
> Really? So, if it differs from $path, then something changed between
> commit 64394e3ae9 and commit 091a6eb0fe. I haven't really read that
> commit/test, so take what I say with a pinch of salt ...

Well yes. I am specifically reading 091a6eb0fe, the changes to t7407.

In that test sm_path contains the relative path from $PWD to the submodule. (It does NOT: "$[sm_]path is the name of the submodule directory relative to the superproject" as documented but rather ... relative to the $PWD)

Show 10 quoted lines

>
>> Thanks for keeping discussing this.
>>
>> So maybe we want to
>> * keep path=sm_path
>
> As I said in commit 64394e3ae9, $path was part of the API, so I could
> not just rename it, without a deprecation period, etc ... So, I was
> 'crossing my fingers' that nobody would export $path in their user
> scripts (not very likely, after all).

Ok. So another approach to get away in the C conversion:
* export the sm_path as all other environment variables
* for "$path" we do not export it into the environment, but
  prefix the command with it, i.e. we'd ask our shell to run
  "path=%s; %s", sm_path, argv[0]
  to preserve the historic behavior.

Show 5 quoted lines

>
>> * fix the documentation via s/$path/$sm_path/g in that section quoted above.
>
> So, "$path is the name of the submodule directory relative to the
> superproject", as currently documented in the man page, yes?

No, the documentation does not match reality. The reality is that both sm_path as well as path give the display path.

> So, $sm_path == $path, at least for some period?

yes that is current reality.

Show 5 quoted lines

>
>> * Introduce a new variable sm_display_path that contains what this patch
>>   proposes sm_path to be.
>
> So, this would be the path from the cwd to the submodule, yes?

yes.

>
> I don't know how windows folks will feel about simply
> removing $path, ...

I agree that this is a bad idea now. As said above we'd just not export the path as an env variable and should be "fine" in the sense that we do not break historical expectations, but have to deal with slightly messy code.

Just digging further, there was ea2fa1040d (submodule foreach: correct path display in recursive submodules, 2016-03-29), which is tangent to the issue of whether we think it is a display path or the superproject->submodule path.

Thanks, Stefan