Re: [Tuxpaint-devel] Hidden bug on Windows?

[TOYAMA Shin-i. <sh...@wm...>]

Hi,

I investigated more to understand why this happened by checking 
original 0.9.26's code.

In old build system for Windows 2000/XP,
----------------
msys:$ make obj/onscreen_keyboard.o

...Compiling on screen keyboard support...
msys:$
----------------

You see no warning and no error.

However, in the latest MinGW/MSYS2 build environment,
----------------
mingw64:$ make obj/onscreen_keyboard.o

...Compiling on screen keyboard support...
src/onscreen_keyboard.c: In function 'mtw':
src/onscreen_keyboard.c:76:16: warning: passing argument 2 of 'libiconv' from 
incompatible pointer type [-Wincompatible-pointer-types]
   76 |   iconv(trans, (const char **)&tok, &in, &wrptr, &out);
      |                ^~~~~~~~~~~~~~~~~~~
      |                |
      |                const char **
In file included from src/onscreen_keyboard.c:55:
C:/msys64/mingw64/include/iconv.h:82:43: note: expected 'char **' but argument 
is of type 'const char **'
   82 | extern size_t iconv (iconv_t cd,  char* * inbuf, size_t *inbytesleft, 
char* * outbuf, size_t *outbytesleft);
      |                                   ~~~~~~~~^~~~~
src/onscreen_keyboard.c:78:18: warning: passing argument 2 of 'swprintf' makes 
integer from pointer without a cast [-Wint-conversion]
   78 |   swprintf(wtok, L"%ls", ui16);
      |                  ^~~~~~
      |                  |
      |                  const short unsigned int *
In file included from C:/msys64/mingw64/x86_64-w64-mingw32/include/wchar.h:1192,
                 from src/onscreen_keyboard.h:1,
                 from src/onscreen_keyboard.c:23:
C:/msys64/mingw64/x86_64-w64-mingw32/include/swprintf.inl:34:41: note: expected 
'size_t' {aka 'long long unsigned int'} but argument is of type 'const short 
unsigned int *'
   34 | int swprintf (wchar_t *__stream, size_t __count, const wchar_t 
*__format, ...)
      |                                  ~~~~~~~^~~~~~~
src/onscreen_keyboard.c:66:10: warning: variable 'n' set but not used 
[-Wunused-but-set-variable]
   66 |   size_t n, in, out;
      |          ^
mingw64:$
----------------

Interestingly, changing the second argument of 'swprintf()' to 
the integer valiable results completely the opposite.

This result says that the type of second argument to swprintf() 
becomes completely different in MinGW/MSYS2.

Although I am not sure why this works on Windows 8 and before, 
it must be one of the reason for the crash on Windows10.

And you see another warning regarding iconv() which is now 
unused in onscreen_keyboard.c but still used in tuxpaint.c

Until now, I have just ignored such warnings when compiling 
TuxPaint on windows, but it might be better to look them more 
carefully and try to reduce warnings as possible I can.

Thanks!

TOYAMA Shin-ichi wrote in 
<616b70f2.4975%sh...@wm...>
>Hi Bill,
>
>Pere Pujal i Carabantes wrote in 
><041...@gm...>
>>> Packages compiled as 0.9.26-5 are available in 
>>> 
>>>  https://z1.plala.jp/tuxpaint/release/0.9.26-5/  (5!!)
>>> 
>>> Could you give them final tests ?
>>> I pray that this will work out this time.
>>
>>As far as I've tested it works fine in W10:
>>No need to enable compatibility to W8 to make it running.
>>No crashes with labels.
>>No crashes with onscreen keyboard.
>>Drawings can be exported/imported from/to Linux/Windows without
>>labels being corrupted.
>
>So, I believe  those two crash bug on windows10  have been fixed 
>finally.
>
>By the way, patched 0.9.26 also compiled on the old build system
>for 2000 and XP, but it did not run.
>Then, I think that  the initial tweak for windows on the OSK was
>not a mistake.
>
>I guess that this bug has been existing since when I changed the 
>build system to MinGW/MSYS2 and become apparent in Windows10.
>
>I think  every windows user, except XP/2000,  would be suggested
>to upgrade to 0.9.26-5.
>
>Thanks.

-- 
TOYAMA Shin-ichi mailto:sh...@wm...