Re: Suspicious warning in W64 build

[Richard Copley]

On 9 September 2017 at 17:07, Eli Zaretskii <address@hidden> wrote:
>> From: Richard Copley <address@hidden>
>> Date: Sat, 9 Sep 2017 12:17:14 +0100
>> Cc: Angelo Graziosi <address@hidden>, Emacs Development <address@hidden>
>>
>> > I see a lot of warnings in that log.  It's a pity no one reports them,
>> > let alone works on fixing them.  (I don't see any of them on my
>> > systems.)
>>
>> I assumed nobody was interested, since there were so many. My
>> mistake, sorry.
>>
>> Reporting the warnings by email and answering questions about what
>> happens when one makes various changes isn't the most convenient
>> edit-compile-test cycle I've seen. You have your reasons for using an
>> old version of the compiler.
>
> I use the latest GCC version provided by mingw.org's MinGW
> distribution.  Currently, that's 6.3.0.
>
>> Can you install MSYS2 and MinGW-W64
>> somewhere just for build testing? It would be less frustrating.
>
> Sorry, this is unlikely to happen.  Being a co-maintainer eats up all
> of my free time, so entertaining yet another incompatible development
> environment and keeping it in good shape is not something I can
> afford.

No sweat.

> I expect others who use MinGW64 to care enough to report and fix these
> problems.

I care and I'm encouraged by your words. I will do what I can.

>> At least some of the "-Wformat=" warnings are misleading.
>> Emacs has to use MSVC's rules for format strings, but GCC
>> warns based on the C standard rules implemented in GCC.
>
> Actually, these are the most worrisome, because they seem to tell your
> MinGW headers might mismatch your GCC version.  Or maybe this is a
> general MinGW64 problem that should be solved by MinGW64 developers.
> Consider this warning, which is quite typical, and is seen in your log
> many times:
>
>     CC       frame.o
>   frame.c: In function 'make_terminal_frame':
>   frame.c:1098:46: warning: unknown conversion type character 'l' in format 
> [-Wformat=]
>      fset_name (f, make_formatted_string (name, "F%"pMd, ++tty_frame_count));
>                                                 ^~~~
>   In file included from 
> C:/msys64/mingw64/x86_64-w64-mingw32/include/inttypes.h:299:0,
>                    from C:/projects/emacs/nt/inc/inttypes.h:24,
>                    from lisp.h:31,
>                    from frame.c:29:
>   C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw_print_pop.h:77:19: 
> note: format string is defined here
>    #define PRIdMAX "lld"
>                      ^
>   frame.c:1098:46: warning: too many arguments for format 
> [-Wformat-extra-args]
>      fset_name (f, make_formatted_string (name, "F%"pMd, ++tty_frame_count));
>                                                 ^~~~
>
> How come the compiler doesn't recognize format specifiers defined on
> the system headers?  And note that as result GCC ignores some
> arguments of fset_name, which might mean it actually generates wrong
> code for this function.
>
> This should be taken up with MinGW64 developers ASAP, because I don't
> see how we can fix this in Emacs.

I don't disagree. My impression is that Alexey and co. at MinGW-W64
are well aware of it. I don't know if there's a solution in the
pipeline. I'm not the best person to press the point there (but ask me
one more time and I will try).

> Another class of similar warnings is like this:
>
>     CC       keyboard.o
>   keyboard.c: In function 'cmd_error':
>   keyboard.c:957:23: warning: format '%d' expects argument of type 'int', but 
> argument 3 has type 'EMACS_INT {aka long long int}' [-Wformat=]
>     sprintf (macroerror, "After %"pI"d kbd macro iterations: ",
>                          ^~~~~~~~~
>   keyboard.c:957:35: note: format string is defined here
>     sprintf (macroerror, "After %"pI"d kbd macro iterations: ",
>                                 ~~~~~^
>                                 %"pI"lld
>
> You may think the compiler doesn't understand %lld, but it also
> doesn't seem to understand the MS native %I64d:
>
>   print.c: In function 'safe_debug_print':
>   print.c:833:24: warning: unknown conversion type character 'I' in format 
> [-Wformat=]
>          fprintf (stderr, "#<%s_LISP_OBJECT 0x%08"pI"x>\r\n",
>                           ^~~~~~~~~~~~~~~~~~~~~~~~
>   In file included from print.c:25:0:
>   lisp.h:98:16: note: format string is defined here
>    #   define pI "I64"
>                   ^
> This leaves us in a conundrum, because I don't understand what printf
> format spec will MinGW64 understand and process correctly when a
> 64-bit integral value has to be printed.

Yep. Sucks.

> Another problem to be taken up with MinGW64 developers is this:
>
>     CCLD     addpm.exe
>   C:/projects/emacs/nt/addpm.c:42:0: warning: "_WIN32_WINNT" redefined
>    #define _WIN32_WINNT _WIN32_WINNT_WIN7
>
>   In file included from 
> C:/msys64/mingw64/x86_64-w64-mingw32/include/crtdefs.h:10:0,
>                    from 
> C:/msys64/mingw64/x86_64-w64-mingw32/include/stdlib.h:9,
>                    from C:/projects/emacs/nt/addpm.c:37:
>   C:/msys64/mingw64/x86_64-w64-mingw32/include/_mingw.h:225:0: note: this is 
> the location of the previous definition
>    #define _WIN32_WINNT 0x502
>
> These are all MinGW system headers, so it sounds like they contradict
> one another?  Maybe there's something Emacs does to trigger this, but
> what is that?

This is due to a local patch. I'm really very sorry to have distracted you
with it, when you don't have the source.

If you want, I will get you a transcript from the unpatched Emacs git
master branch as I should have done in the first place.

> Anyway, I fixed some warnings, so you should see fewer of them.
> Hopefully, I didn't introduce new warnings aor problems.  If/when the
> MinGW64 folks (or someone here who is "in the know") tells how to
> resolve the problems with printf and _WIN32_WINNT, we can fix the
> rest.
>
> There are few warnings which are not specific to MS-Windows; I will
> describe them in a separate message.
>
> Thanks.

Thanks very much.