Re: [PATCH 1/6] writer: Emit definitions of declarations when they are present

[Dodji Seketeli <dodji@redhat.com>]

Giuliano Procida <gprocida@google.com> writes:

> Hi.
>
> On Mon, 30 Nov 2020 at 15:56, Dodji Seketeli <dodji@redhat.com> wrote:
>
>> Hello Giuliano,
>>
>> Giuliano Procida via Libabigail <libabigail@sourceware.org> a écrit:
>>
>> > Hi Dodji.
>> >
>> > I had a quick scan through this.
>> >
>> > Do we need  look_through_decl_only_enum  somewhere as well?
>>
>> Good question.
>>
>> Right now, when the equals function compares enums, it doens't look
>> through the declaration to get the definition of the enum.  It will only
>> use the decl-only part of the enum in the comparison, even if the enum
>> decl was fully resolved to its definition.  For classes (and unions)
>> however, equals always looks through the declaration.
>>
>> So, at ABIXML write time, if we happen to not saving the definition and
>> we only save the decl-only part, the comparison should be done between
>> the decl-only part of both enums (the one in the IR coming from the
>> binary and the one from the IR coming from the ABIXML) being compared.
>> So it shouldn't yield an ABI change.
>>
>> Now, the behaviour for classes/unions can be said to be inconsistent with
>> the behaviour for enums.  So we might indeed want to always serialize
>> the definition of declarations of enum if we have it.
>>
>> But then, we'd need update at least the 'equals' comparison function
>> accordingly, I believe.
>>
>> But as this is like getting into the "feature" territory (kind of) I'd
>> wait for releasing 1.8 before doing this.
>>
>> What do you think?
>>
>>
> I agree. TBH, in retrospect, all the forward-declared enum changes could
> have waited. I can paste your comment into Bugzilla, but you might prefer
> something else.

Yeah, you can paste it there if you feel like it's useful :-)  I don't
think strongly either way :-) I live you decide.

Thanks!


-- 
		Dodji