[PATCH] PR fortran/50549 - should detect different type parameters in structure constructors

[Harald Anlauf]

Hi Tobias,

Am 28.03.22 um 12:05 schrieb Tobias Burnus:
> Thanks for the patch! LGTM and I think GCC 12 is still okay.
> 
> However, I have a nit:
> 
>> --- a/gcc/fortran/resolve.cc
>> +++ b/gcc/fortran/resolve.cc
>> @@ -1375,11 +1375,22 @@ resolve_structure_cons (gfc_expr *expr, int init)
>> ...
>> +           long len_a, len_b;
>> +           len_a = mpz_get_si (comp->ts.u.cl->length->value.integer);
>> +           len_b = mpz_get_si 
>> (cons->expr->ts.u.cl->length->value.integer);
>> +           gfc_error ("Unequal character lengths (%ld/%ld) for pointer "
>> +                      "component %qs in constructor at %L",
>> +                      len_a, len_b, comp->name, &cons->expr->where);
> 
> 'long' might be int32_t instead of int64_t (e.g. on Windows, I think both
> MinGW32 and MinGW64, but I am not quite sure). Thus, I wonder whether it
> makes more sense to use:
> 
>    HOST_WIDE_INT, gfc_mpz_get_hwi() and '%wd'
> 
> I note that '%wd' (and '%lld') is only supported since last August
> (commit https://gcc.gnu.org/r12-3044-g1b507b1e3c5 ), but now that it is,
> I think we should use it at places where the value can be larger than 
> INT_MAX.

using HOST_WIDE_INT as in the updated patch (sort of) works, but for
some reason I do not yet understand the format check kicks in for
gfc_error, producing:

../../gcc-trunk/gcc/fortran/resolve.cc: In function 'bool 
resolve_structure_cons(gfc_expr*, int)':
../../gcc-trunk/gcc/fortran/resolve.cc:1388:43: warning: unknown 
conversion type character 'w' in format [-Wformat=]
      la, lb, comp->name, &cons->expr->where);
                                            ^
../../gcc-trunk/gcc/fortran/resolve.cc:1388:43: warning: unknown 
conversion type character 'w' in format [-Wformat=]
../../gcc-trunk/gcc/fortran/resolve.cc:1388:43: warning: format '%s' 
expects argument of type 'char*', but argument 2 has type 'long int' 
[-Wformat=]
../../gcc-trunk/gcc/fortran/resolve.cc:1388:43: warning: format '%L' 
expects argument of type 'locus*', but argument 3 has type 'long int' 
[-Wformat=]
../../gcc-trunk/gcc/fortran/resolve.cc:1388:43: warning: too many 
arguments for format [-Wformat-extra-args]

This would likely lead to a bootstrap error.

Do I need to add some forgotten include?  Or some annotation to
suppress the warning?

Or should I rather convert the character lengths via sprintf first
before generating the error message?  (That would be the quick fix.)

> I think at some point, we should also check the rest of the code and
> change those mpz_get_si to gfc_mpz_get_hwi which can exceed INT_MAX.
> Likewise, some of the %ld/%lu or %lld/%llu code should be also converted 
> to %wd/%wu.
> 
> Tobias
> 
> PS: For using HWI with 'sprintf' instead of diagnostic's error/warning,
> HOST_WIDE_INT_PRINT_DEC exists and has to be used.

All current cases of printing a HOST_WIDE_INT in gcc/fortran/ use
'sprintf', and I did not find any other use of %wd/%wu.  So the
mentioned implementation is not really stressed yet... ;-)

Thanks,
Harald

> -----------------
> Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 
> 80634 München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: 
> Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: München; 
> Registergericht München, HRB 106955
> 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-Fortran-character-length-of-pointer-assignments-in-s.patch
Type: text/x-patch
Size: 3184 bytes
Desc: not available
URL: <https://gcc.gnu.org/pipermail/fortran/attachments/20220328/ebaab6fb/attachment-0001.bin>