bug#6126: 24.0.50; Segmentation fault when w32-shell-execute try to open an unassociated file

[Eli Zaretskii <eliz@gnu.org>]

> From: Lennart Borgman <lennart.borgman@gmail.com>
> Date: Fri, 7 May 2010 02:00:49 +0200
> Cc: 6126@debbugs.gnu.org
> 
> - In w32_error the argument error_no has type int. It should be more
> easy to understand if it had the type DWORD which is what GetLastError
> returns. Will using int be correct on all w32 platforms?

Yes.  DWORD is an unsigned 32-bit integer type on all versions of
Windows, even on 64-bit Windows.  (I agree that it would be better to
use `unsigned int' rather than just `int', though.)

> - The call to error in w32-shell-execute has only two arguments. Is
> that correct?

Yes, the other arguments are optional, see the doc string.

> error in eval.c takes four arguments.

`error' accepts a variable-size argument list, like `printf'.  This is
stated in the commentary in eval.c.

> - The parameter lpBuffer to FormatMessage has the type LPTSTR. Is it
> correct to call that with *char (ie buf)?

Yes.  LPTSTR is defined as a `char *' in non-Unicode builds, and as a
`wchar_t *' in Unicode builds (which we don't yet support in Emacs,
but we should, some day).

> It looks in the backtrace
> like even the argument a1 to error is incorrect.

If you mean these parts of the backtrace:

    a1=0x40008048 <Address 0x40008048 out of bounds>,
    a2=0x40008048 <Address 0x40008048 out of bounds>,
    a3=0x40008048 <Address 0x40008048 out of bounds>) at eval.c:2078

then it looks like `error' handles that just fine, because the value
of `args[]' computed from them is correct:

        args = {
          0x137bc48
"\317\265\315\263\325\322\262\273\265\275\326\270\266\250\265\304\316\304\274\376\241\243\r\n",
          0x465fc78 "C:\\abc.ttt", 0x0}

Maybe the problem happens because the error string (args[0]) is
encoded in a locale-specific encoding, so perhaps calling build_string
on it is not TRT.