Re: [PATCH 8/8] OpenMP: Fortran "!$omp declare mapper" support

[Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>]

On 18 September 2023 12:19:17 CEST, Julian Brown <julian@codesourcery.com> wrote:
>On Thu, 14 Sep 2023 17:13:02 +0200
>Bernhard Reutner-Fischer via Gcc-patches <gcc-patches@gcc.gnu.org>
>wrote:
>
>> On Tue, 5 Sep 2023 12:28:28 -0700
>> Julian Brown <julian@codesourcery.com> wrote:
>> 
>> > +  static bool
>> > +  equal (const omp_name_type<gfc_typespec *> &a,
>> > +	 const omp_name_type<gfc_typespec *> &b)
>> > +  {
>> > +    if (a.name == NULL_TREE && b.name == NULL_TREE)
>> > +      return a.type == b.type;  
>> 
>> I'm curious if (and why) the type comparison above is safe and does
>> not use gfc_compare_types () ?
>> 
>> thanks,
>
>Probably ignorance on my part! It works for (derived, class) types which
>are canonicalized to exactly the same gfc_typespec, but you're likely
>right that a more Fortran-ish notion of type equality should be used
>here instead when comparing "declare mapper"s.

Sounds like a word2vec question.

>
>Or maybe using gfc_compare_types would smush too many types together
>into one? E.g. if b.type is an extension of a.type, do we want those
>types to be able to have separate mappers?

I fear this really depends.

>
>I'll have a look at addressing this when it's time to reroll these
>patches.
>
>Thanks,

TIA,

PS: I'm not thrilled about those recent get_identifier("") as temporary additions fed into the hasher. I know it's not funny, but please let's avoid that. Said the cat.

>
>Julian