Re: [hatari-devel] printf for 64 bit value and %PRId64/ %I64d

[Thomas Huth <th.huth@xxxxxxxxx>]

Am Sat, 17 Jul 2021 17:25:33 +0200
schrieb Nicolas Pomarède <npomarede@xxxxxxxxxxxx>:

> Hi
> 
> it seems that with recent version of mingw/gcc that I'm using there's 
> now an error when using the portable %PRId64 macro to print 64 bit 
> integer and crosscompiling under linux to do a Windows build :
> 
> 
> /home/npomarede/src/win/src/gemdos.c: In function 
> 'restore_file_handle_info':
> /home/npomarede/src/win/src/gemdos.c:891:24: warning: unknown
> conversion type character 'l' in format [-Wformat=]
>    891 |   Log_Printf(LOG_WARN, "GEMDOS '%s' handle %d cannot be 
> restored, seek to saved offset %"PRId64" failed for: %s\n",
>        | 
> ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> In file included from /home/npomarede/src/win/src/gemdos.c:42:
> /usr/x86_64-w64-mingw32/sys-root/mingw/include/inttypes.h:33:18:
> note: format string is defined here
>     33 | #define PRId64 "lld"
>        |                  ^
> 
> One possible fix for this is to do :
> 
> #define PRId64 "I64d"
> 
> only when compiling for windows (as %I64d seems to be more supported 
> under Windows)
> 
> Any other idea how to fix this in the most clean/secure way ?

IIRC I once hit this problem already, too, when the Fedora container in
the gitlab-CI got upgraded to F34 (see commit 89ce85e772bc5ccd1) ... I
had a look at it, but did not come to a clean solution. I think it has
something to do with MinGW supporting both, the printf-style functions
from the MS runtime, and the C99-style macros, at the same time.

I did not find a clean solution back then and simply downgraded the
container to F33 to "fix" it. But I think you could try playing with
__attribut__((format(gnu_printf, 2, 3))) instead of the normal "printf"
attribute - but beware, it might not be supported by Clang. Or maybe
check the __USE_MINGW_ANSI_STDIO define ... but I think it also did not
work for me back then.
Alternatively cast the value to (long) and use "%l" instead (if
it's unlikely that the value is bigger than 32-bit)?

 Thomas