Re: [PATCH 3/5] Fortran: Narrow return types [PR78798]

[Harald Anlauf <anlauf@gmx.de>]

Am 13.11.22 um 11:39 schrieb Bernhard Reutner-Fischer via Gcc-patches:
> On Sun, 13 Nov 2022 12:13:26 +0200
> Janne Blomqvist <blomqvist.janne@gmail.com> wrote:
> 
>> On Sun, Nov 13, 2022 at 1:47 AM Bernhard Reutner-Fischer via Fortran
>> <fortran@gcc.gnu.org> wrote:
>>> --- a/gcc/fortran/arith.cc
>>> +++ b/gcc/fortran/arith.cc
>>> @@ -1135,7 +1135,7 @@ compare_complex (gfc_expr *op1, gfc_expr *op2)
>>>      strings.  We return -1 for a < b, 0 for a == b and 1 for a > b.
>>>      We use the processor's default collating sequence.  */
>>>
>>> -int
>>> +signed char
>>>   gfc_compare_string (gfc_expr *a, gfc_expr *b)
>>>   {
>>>     size_t len, alen, blen, i;
>>> @@ -1162,7 +1162,7 @@ gfc_compare_string (gfc_expr *a, gfc_expr *b)
>>>   }
>>
>> Hmm, really? PR 78798 mentions changing int to bool, where
>> appropriate, which I think is uncontroversial, but this?
> 
> Well we could leave this or all spots alone where a bool is
> insufficient, if you prefer.
> 
> In the case of gfc_compare_string, the only user is simplify which only
> looks at ge/gt/le/lt 0

My reading of the mentioned PR is that there is a fundamental
disagreement with the subject:

Bug 78798 - [cleanup] some int-valued functions should be bool

I see that as an issue of (a minor lack of) conciseness;
it is *not* about narrowing.

Replacing "int" by "signed char" adds confusion and makes code
less understandable, so I would oppose it, as we don't solve a
real problem and rather add confusion.