[PATCH] PR fortran/100950 - ICE in output_constructor_regular_field, at varasm.c:5514

Jakub Jelinek jakub@redhat.com
Fri Aug 20 09:16:18 GMT 2021

On Fri, Aug 20, 2021 at 08:48:57AM +0200, Harald Anlauf via Gcc-patches wrote:
> Hi!
> > Gesendet: Freitag, 20. August 2021 um 02:21 Uhr
> > Von: "H.J. Lu" <hjl.tools@gmail.com>
> > This may have broken bootstrap on 32-bit hosts:
> > 
> > https://gcc.gnu.org/pipermail/gcc-regression/2021-August/075209.html
> I do not understand the error message:
> ../../src-master/gcc/fortran/simplify.c: In function ‘bool substring_has_constant_len(gfc_expr*)’:
> ../../src-master/gcc/fortran/simplify.c:4557:22: error: unknown conversion type character ‘l’ in format [-Werror=format=]
>  4557 |           gfc_error ("Substring start index (" HOST_WIDE_INT_PRINT_DEC
>       |                      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>  4558 |                      ") at %L below 1",
>       |                      ~~~~~~~~~~~~~~~~~
> ../../src-master/gcc/fortran/simplify.c:4557:22: error: format ‘%L’ expects argument of type ‘locus*’, but argument 2 has type ‘long long int’ [-Werror=format=]
>  4557 |           gfc_error ("Substring start index (" HOST_WIDE_INT_PRINT_DEC
>       |                      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>  4558 |                      ") at %L below 1",
>       |                      ~~~~~~~~~~~~~~~~~
>  4559 |                      istart, &ref->u.ss.start->where);
>       |                      ~~~~~~
>       |                      |
>       |                      long long int
> ../../src-master/gcc/fortran/simplify.c:4557:22: error: too many arguments for format [-Werror=format-extra-args]
> Is there an issue with HOST_WIDE_INT_PRINT_DEC on 32-bit hosts?
> What is the right way to print a HOST_WIDE_INT?
> It works on 64-bit without any warning.

gfc_error etc. aren't *printf family, it has its own format specifier
handling (e.g. it handles %L and %C), and it is even different
from the error/warning handling in middle-end/c-family FEs/backends.

HOST_WIDE_INT_PRINT_DEC is wrong in the above spot for 2 reasons:
1) gfc_error etc. argument is automatically marked for translation
   and translated, having a macro in there that expands to various things
   depending on host makes it impossible to mark for translation and
   a lottery for translation.
2) hwint.h defines:
#define HOST_LONG_FORMAT "l"
#define PRId64 GCC_PRI64 "d"
   but xm-mingw32.h overrides
   so effectively HOST_WIDE_INT_PRINT_DEC is "%ld", "%lld" or "%I64d"
   Now, gfc_error does handle %d or %ld, but doesn't handle %lld nor
   %I64d, so even if the 1) issue didn't exist, this explains why
   it works only on some hosts (e.g. x86_64-linux where %ld is used).
   But e.g. on i686-linux or many other hosts it is %lld which
   gfc_error doesn't handle and on Windows that %I64d.

Now, the non-Fortran FE diagnostic code actually has %wd for this (w
modifier like l modifier), which takes HOST_WIDE_INT/unsigned HOST_WIDE_INT
argument and prints it.

So, either you get through the hops to support that, unfortunately it isn't
just adding support for that in fortran/error.c (error_print) and some
helper functions, which wouldn't be that hard, just add 'w' next to 'l'
handling, TYPE_* for that and union member etc., but one needs to modify
c-family/c-format.c too to register the modifier so that gcc doesn't warn
about it and knows the proper argument type etc.

The other much easier but uglier option is to use a temporary buffer:
  char buffer[21];
  sprintf (buffer, HOST_WIDE_INT_PRINT_DEC, hwint_val);
  gfc_error ("... %s ...", ... buffer ...);
This works, as it uses the host sprintf i.e. *printf family for which
HOST_WIDE_INT_PRINT_DEC macro is designed.


More information about the Gcc-patches mailing list