Re: [PATCH v9 03/11] ref-filter: implement an `align` atom

> On Sun, Aug 9, 2015 at 4:09 AM, Karthik Nayak <karthik.188@gmail.com> wrote:
>> On Sun, Aug 9, 2015 at 1:34 PM, Eric Sunshine <sunshine@sunshineco.com> wrote:
>>> On Sun, Aug 9, 2015 at 2:55 AM, Karthik Nayak <karthik.188@gmail.com> wrote:
>>>> On Sun, Aug 9, 2015 at 9:12 AM, Eric Sunshine <sunshine@sunshineco.com> wrote:
>>>>> That raises another question. Why are 'struct ref_formatting_state',
>>>>> 'struct align', 'struct atom_value', etc. defined in ref-filter.h at
>>>>> all? Aren't those private implementation details of ref-filter.c, or
>>>>> do you expect other code to be using them?
>>>>
>>>> I guess struct ref_formatting_state and struct align could be moved to
>>>> ref-filter.c. About struct atom_value its referenced by ref_array_item()
>>>> so any reader reading about this, would find it easier if atom_value()
>>>> is at the same place.
>>>
>>> Do you expect callers ever to be manipulating or otherwise accessing
>>> the atom_value of ref_array_item? If callers have no business mucking
>>> with atom_value, then one option would be to simply forward declare
>>> atom_value in the header:
>>>
>>>     struct atom_value;
>>>
>>>     struct ref_array_item {
>>>         ...
>>>         struct atom_value *value;
>>>         ...
>>>     };
>>>
>>> which makes atom_value opaque to clients of ref-filter. The actual
>>> declaration of atom_value would then be moved to ref-filter.c, thus
>>> kept private.
>>
>> Also the code that this was done in has been excepted into `next`
>> so either I send a new series for the same, or write a patch just to
>> move this from ref-filter.h to ref-filter.c. So what would you suggest?
>
> To my eye, atom_value seems to encapsulate a bunch of state local to
> and only meaningful to ref-filter's internal workings, so it doesn't
> really belong in the public header. Assuming that you don't foresee
> any callers ever needing to access the properties of atom_value, then
> it might indeed be reasonable to introduce a patch which moves it from
> the .h file to the .c file (while leaving only a forward declaration
> in the .h file).