Re: [PATCH v2 03/18] ref-filter: make valid_atom as function parameter

[Christian Couder <christian.couder@gmail.com>]

On Thu, Jan 18, 2018 at 7:20 AM, Оля Тележная <olyatelezhnaya@gmail.com> wrote:
> 2018-01-18 1:39 GMT+03:00 Christian Couder <christian.couder@gmail.com>:
>> On Wed, Jan 17, 2018 at 10:43 PM, Jeff King <peff@peff.net> wrote:
>>> On Tue, Jan 16, 2018 at 09:55:22AM +0300, Оля Тележная wrote:
>>>
>>>> > IOW, the progression I'd expect in a series like this is:
>>>> >
>>>> >   1. Teach ref-filter.c to support everything that cat-file can do.
>>>> >
>>>> >   2. Convert cat-file to use ref-filter.c.
>>>>
>>>> I agree, I even made this and it's working fine:
>>>> https://github.com/git/git/pull/450/commits/1b74f1047f07434dccb207534d1ad45a143e3f2b
>>
>> (Nit: it looks like the above link does not work any more, but it
>> seems that you are talking about the last patch on the catfile
>> branch.)
>>
>>>> But I decided not to add that to patch because I expand the
>>>> functionality of several commands (not only cat-file and
>>>> for-each-ref), and I need to support all new functionality in a proper
>>>> way, make these error messages, test everything and - the hardest one
>>>> - support many new commands for cat-file. As I understand, it is not
>>>> possible unless we finally move to ref-filter and print results also
>>>> there. Oh, and I also need to rewrite docs in that case. And I decided
>>>> to apply this in another patch. But, please, say your opinion, maybe
>>>> we could do that here in some way.
>>>
>>> Yeah, I agree that it will cause changes to other users of ref-filter,
>>> and you'd need to deal with documentation and tests there. But I think
>>> we're going to have to do those things eventually (since supporting
>>> those extra fields everywhere is part of the point of the project). And
>>> by doing them now, I think it can make the transition for cat-file a lot
>>> simpler, because we never have to puzzle over this intermediate state
>>> where only some of the atoms are valid for cat-file.
>>
>> I agree that you will have to deal with documentation and tests at one
>> point and that it could be a good idea to do that now.
>>
>> I wonder if it is possible to add atoms one by one into ref-filter and
>> to add tests and documentation at the same time, instead of merging
>> cat-file atoms with ref-filter atoms in one big step.
>>
>> When all the cat-file atoms have been added to ref-filter's
>> valid_atom, maybe you could add ref-filter's atoms to cat-file's
>> valid_cat_file_atom one by one and add tests and documentation at the
>> same time.
>>
>> And then when ref-filter's valid_atom and cat-file's
>> valid_cat_file_atom are identical you can remove cat-file's
>> valid_cat_file_atom and maybe after that rename "ref-filter" to
>> "format".
>
> I think it's important to finish migrating process at first. I mean,
> now we are preparing and collecting everything in ref-filter, but we
> make resulting string and print still in cat-file. And I am not sure,
> but maybe it will not be possible to start using new atoms in cat-file
> while some part of logic still differs.

Ok, you can finish the migration process then.

> And another thoughts here - we were thinking about creating format.h
> but decided not to move forward with it, and now we are suffering
> because of it. Can I create it right now or the history of commits
> would be too dirty because of it?

It would also make it difficult to refactor your patch series if there
is a big move or renaming in the middle.

> Also, do you mean just renaming of
> ref-filter? I was thinking that I need to put formatting-related logic
> to another file and leave all other stuff in ref-filter.

Yeah, you can do both a move and a renaming.

> Anyway, your suggested steps looks good, and I am planning to
> implement them later.

Ok.

> Let's discuss, what behavior we are waiting for
> when atom seems useless for the command. Die or ignore?

We could alternatively just emit a warning, but it looks like there
are a lot of die() calls already in ref-filter.c, so I would suggest
die().

> And, which
> atoms are useless (as I understand, "rest" and "deltabase" from
> cat-file are useless for all ref-filter users, so the question is
> about - am I right in it, and about ref-filter atoms for cat-file).

For now and I think until the migration process is finished, you could
just die() in case of any atom not already supported by the command.

> I have never written any tests and docs for Git, I will try to explore
> by myself how to do that, but if you have any special links/materials
> about it - please send them to me :)

I think that looking at the existing documentation and tests is
probably the best way to learn about it.