bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[李丁]

[-- Attachment #1: Type: text/plain, Size: 602 bytes --]

Hi,

I tried to compile emacs 24.2 on Windows 7 with Visual C++ 11.0 Express,
and I found that EnumSystemLocales in w32proc.c and w32select.c were
compiled as is (not as __stdcall function with postfix decorations), which
causes the linking failure.

I read through the sources and found that config.h defined _WIN32_WINNT as
0x0400, which corresponds to Windows NT4. However MSDN says that
EnumSystemLocales requires at least Windows 2000 (0x0500). So you should
probably either require a higher Windows version or use some other
mechanism to accomplish what EnumSystemLocales does.

Regards,
Li Ding

[-- Attachment #2: Type: text/html, Size: 689 bytes --]

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[Eli Zaretskii]

> Date: Tue, 13 Nov 2012 19:11:34 +0800
> From: 李丁 <iamliding@gmail.com>
> 
> I tried to compile emacs 24.2 on Windows 7 with Visual C++ 11.0 Express,
> and I found that EnumSystemLocales in w32proc.c and w32select.c were
> compiled as is (not as __stdcall function with postfix decorations), which
> causes the linking failure.
> 
> I read through the sources and found that config.h defined _WIN32_WINNT as
> 0x0400, which corresponds to Windows NT4. However MSDN says that
> EnumSystemLocales requires at least Windows 2000 (0x0500). So you should
> probably either require a higher Windows version or use some other
> mechanism to accomplish what EnumSystemLocales does.

Would adding a correct prototype for EnumSystemLocales to w32term.h
(included by both source files you mention) do the trick?

(I don't want to bump up _WIN32_WINNT, because that might produce an
executable which won't run on Windows 9X, which we still try to
support.)

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[李丁]

[-- Attachment #1: Type: text/plain, Size: 2038 bytes --]

Sure, adding a correct prototype is just like including the correct header
file if Windows 9X
really supports this function. But you should be careful checking compiler
versions, in case that
redefinition error occur.

Actually, I encountered several redefinition problems during compilation.
For example in w32term.c:

    #ifndef GLYPHSET
    /* Pre Windows 2000, this was not available, but define it here so
       that Emacs compiled on such a platform will run on newer versions.
 */
    ...
    #endif

VC 11.0 does not define GLYPHSET either (or not included), but it does have
the definitions,
and above code leads to redefinition error.

And this in w32term.c too:

    /* Reportedly, MSVC does not have this in its headers.  */
    #ifdef _MSC_VER
    DECLARE_HANDLE(HMONITOR);
    #endif

VC 11.0 does have HMONITOR, and another redefinition error.

Maybe VC 11.0 has not been widely used, but these small problems should be
resolved.


2012/11/14 Eli Zaretskii <eliz@gnu.org>

> > Date: Tue, 13 Nov 2012 19:11:34 +0800
> > From: 李丁 <iamliding@gmail.com>
> >
> > I tried to compile emacs 24.2 on Windows 7 with Visual C++ 11.0 Express,
> > and I found that EnumSystemLocales in w32proc.c and w32select.c were
> > compiled as is (not as __stdcall function with postfix decorations),
> which
> > causes the linking failure.
> >
> > I read through the sources and found that config.h defined _WIN32_WINNT
> as
> > 0x0400, which corresponds to Windows NT4. However MSDN says that
> > EnumSystemLocales requires at least Windows 2000 (0x0500). So you should
> > probably either require a higher Windows version or use some other
> > mechanism to accomplish what EnumSystemLocales does.
>
> Would adding a correct prototype for EnumSystemLocales to w32term.h
> (included by both source files you mention) do the trick?
>
> (I don't want to bump up _WIN32_WINNT, because that might produce an
> executable which won't run on Windows 9X, which we still try to
> support.)
>

[-- Attachment #2: Type: text/html, Size: 2899 bytes --]

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[李丁]

[-- Attachment #1: Type: text/plain, Size: 2425 bytes --]

Sorry, this piece of code is in w32fns.c (line 143-146), not w32term.c:
   /* Reportedly, MSVC does not have this in its headers.  */
    #ifdef _MSC_VER
    DECLARE_HANDLE(HMONITOR);
    #endif
And it leads to redefinition error under MSVC 11.0.


2012/11/14 李丁 <iamliding@gmail.com>

> Sure, adding a correct prototype is just like including the correct header
> file if Windows 9X
> really supports this function. But you should be careful checking compiler
> versions, in case that
> redefinition error occur.
>
> Actually, I encountered several redefinition problems during compilation.
> For example in w32term.c:
>
>     #ifndef GLYPHSET
>     /* Pre Windows 2000, this was not available, but define it here so
>        that Emacs compiled on such a platform will run on newer versions.
>  */
>     ...
>     #endif
>
> VC 11.0 does not define GLYPHSET either (or not included), but it does
> have the definitions,
> and above code leads to redefinition error.
>
> And this in w32term.c too:
>
>     /* Reportedly, MSVC does not have this in its headers.  */
>     #ifdef _MSC_VER
>     DECLARE_HANDLE(HMONITOR);
>     #endif
>
> VC 11.0 does have HMONITOR, and another redefinition error.
>
> Maybe VC 11.0 has not been widely used, but these small problems should be
> resolved.
>
>
> 2012/11/14 Eli Zaretskii <eliz@gnu.org>
>
>> > Date: Tue, 13 Nov 2012 19:11:34 +0800
>> > From: 李丁 <iamliding@gmail.com>
>> >
>> > I tried to compile emacs 24.2 on Windows 7 with Visual C++ 11.0 Express,
>> > and I found that EnumSystemLocales in w32proc.c and w32select.c were
>> > compiled as is (not as __stdcall function with postfix decorations),
>> which
>> > causes the linking failure.
>> >
>> > I read through the sources and found that config.h defined _WIN32_WINNT
>> as
>> > 0x0400, which corresponds to Windows NT4. However MSDN says that
>> > EnumSystemLocales requires at least Windows 2000 (0x0500). So you should
>> > probably either require a higher Windows version or use some other
>> > mechanism to accomplish what EnumSystemLocales does.
>>
>> Would adding a correct prototype for EnumSystemLocales to w32term.h
>> (included by both source files you mention) do the trick?
>>
>> (I don't want to bump up _WIN32_WINNT, because that might produce an
>> executable which won't run on Windows 9X, which we still try to
>> support.)
>>
>
>

[-- Attachment #2: Type: text/html, Size: 3963 bytes --]

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[Eli Zaretskii]

> Date: Wed, 14 Nov 2012 09:14:45 +0800
> From: 李丁 <iamliding@gmail.com>
> Cc: 12878@debbugs.gnu.org
> 
> Sure, adding a correct prototype is just like including the correct header
> file if Windows 9X
> really supports this function. But you should be careful checking compiler
> versions, in case that
> redefinition error occur.

A repeated prototype can never trigger redefinition warnings or errors.

> Actually, I encountered several redefinition problems during compilation.
> For example in w32term.c:
> 
>     #ifndef GLYPHSET
>     /* Pre Windows 2000, this was not available, but define it here so
>        that Emacs compiled on such a platform will run on newer versions.
>  */
>     ...
>     #endif
> 
> VC 11.0 does not define GLYPHSET either (or not included), but it does have
> the definitions,
> and above code leads to redefinition error.
> 
> And this in w32term.c too:
> 
>     /* Reportedly, MSVC does not have this in its headers.  */
>     #ifdef _MSC_VER
>     DECLARE_HANDLE(HMONITOR);
>     #endif
> 
> VC 11.0 does have HMONITOR, and another redefinition error.

What is the value of _MSC_VER for this compiler?

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[李丁]

[-- Attachment #1: Type: text/plain, Size: 1321 bytes --]

The value is 1700


2012/11/14 Eli Zaretskii <eliz@gnu.org>

> > Date: Wed, 14 Nov 2012 09:14:45 +0800
> > From: 李丁 <iamliding@gmail.com>
> > Cc: 12878@debbugs.gnu.org
> >
> > Sure, adding a correct prototype is just like including the correct
> header
> > file if Windows 9X
> > really supports this function. But you should be careful checking
> compiler
> > versions, in case that
> > redefinition error occur.
>
> A repeated prototype can never trigger redefinition warnings or errors.
>
> > Actually, I encountered several redefinition problems during compilation.
> > For example in w32term.c:
> >
> >     #ifndef GLYPHSET
> >     /* Pre Windows 2000, this was not available, but define it here so
> >        that Emacs compiled on such a platform will run on newer versions.
> >  */
> >     ...
> >     #endif
> >
> > VC 11.0 does not define GLYPHSET either (or not included), but it does
> have
> > the definitions,
> > and above code leads to redefinition error.
> >
> > And this in w32term.c too:
> >
> >     /* Reportedly, MSVC does not have this in its headers.  */
> >     #ifdef _MSC_VER
> >     DECLARE_HANDLE(HMONITOR);
> >     #endif
> >
> > VC 11.0 does have HMONITOR, and another redefinition error.
>
> What is the value of _MSC_VER for this compiler?
>

[-- Attachment #2: Type: text/html, Size: 1987 bytes --]

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[Eli Zaretskii]

> Date: Wed, 14 Nov 2012 09:14:45 +0800
> From: 李丁 <iamliding@gmail.com>
> Cc: 12878@debbugs.gnu.org
> 
> Sure, adding a correct prototype is just like including the correct
> header file if Windows 9X really supports this function. But you
> should be careful checking compiler versions, in case that
> redefinition error occur.

I added the prototype of EnumSystemLocales to one of the w32 headers.
Please try the latest emacs-24 branch (revision 110902 or later) and
see if you still have problems building it.

> Actually, I encountered several redefinition problems during compilation.
> For example in w32term.c:
> 
>     #ifndef GLYPHSET
>     /* Pre Windows 2000, this was not available, but define it here so
>        that Emacs compiled on such a platform will run on newer versions.
>  */
>     ...
>     #endif
> 
> VC 11.0 does not define GLYPHSET either (or not included), but it does have
> the definitions,
> and above code leads to redefinition error.

This is no longer in the development sources of the emacs-24 branch.
This code fragment is now guarded by "#if _WIN32_WINNT < 0x0500",
which I think should work for you, as long as you don't define
_WIN32_WINNT to a value higher than 0x0400.

> And this in w32term.c too:
> 
>     /* Reportedly, MSVC does not have this in its headers.  */
>     #ifdef _MSC_VER
>     DECLARE_HANDLE(HMONITOR);
>     #endif
> 
> VC 11.0 does have HMONITOR, and another redefinition error.

This is now guarded by "#if defined (_MSC_VER) && _WIN32_WINNT < 0x0500"
so again, I don't think that it should give you trouble with
_WIN32_WINNT set at 0x0400.

Please try the latest emacs-24 branch, and if it works for you, I will
close the bug.

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[李丁]

[-- Attachment #1: Type: text/plain, Size: 3126 bytes --]

Hi,
Thank you for fixing the bug.

I checked out the latest branch (r 110913), it seems that you miss a
semicolon at line 761 in w32term.h.

Previous problems are fixed, but I still can not compile the latest branch.
There are more problems:

1. I need `nmake bootstrap' but in 24.2 release I can compile directly with
`nmake'

2. In lisp.h it begins to define ARRAY_MARK_FLAG as PTRDIFF_MIN, however,
    definition of PTRDIFF_MIN is in stdint.h, which is not included. And
even if included, in nt/inc/stdint.h,
    various *_MIN (including PTRDIFF_MIN) are not defined

3. In w32.c DeviceIoControl, it uses a FSCTL_GET_REPARSE_POINT that is only
defined for _MSC_VER >= 0x0500

4. In xdisp.c, start_hourglass function, w32_note_current_window is
declared (as extern) and used in the middle of
    the function, causing an error message. Move the declaration to the
beginning of the function solved the problem.

After fixing the above problems, finally I can make a temacs.exe, but while
loading subr.el, temacs reported an
error message saying `Invalid funtion: "DEAD"'. I do not know where the
function `dead' comes from.

Maybe you can close the previous bug, but for the latest branch, more fixes
are needed.



2012/11/18 Eli Zaretskii <eliz@gnu.org>

> > Date: Wed, 14 Nov 2012 09:14:45 +0800
> > From: 李丁 <iamliding@gmail.com>
> > Cc: 12878@debbugs.gnu.org
> >
> > Sure, adding a correct prototype is just like including the correct
> > header file if Windows 9X really supports this function. But you
> > should be careful checking compiler versions, in case that
> > redefinition error occur.
>
> I added the prototype of EnumSystemLocales to one of the w32 headers.
> Please try the latest emacs-24 branch (revision 110902 or later) and
> see if you still have problems building it.
>
> > Actually, I encountered several redefinition problems during compilation.
> > For example in w32term.c:
> >
> >     #ifndef GLYPHSET
> >     /* Pre Windows 2000, this was not available, but define it here so
> >        that Emacs compiled on such a platform will run on newer versions.
> >  */
> >     ...
> >     #endif
> >
> > VC 11.0 does not define GLYPHSET either (or not included), but it does
> have
> > the definitions,
> > and above code leads to redefinition error.
>
> This is no longer in the development sources of the emacs-24 branch.
> This code fragment is now guarded by "#if _WIN32_WINNT < 0x0500",
> which I think should work for you, as long as you don't define
> _WIN32_WINNT to a value higher than 0x0400.
>
> > And this in w32term.c too:
> >
> >     /* Reportedly, MSVC does not have this in its headers.  */
> >     #ifdef _MSC_VER
> >     DECLARE_HANDLE(HMONITOR);
> >     #endif
> >
> > VC 11.0 does have HMONITOR, and another redefinition error.
>
> This is now guarded by "#if defined (_MSC_VER) && _WIN32_WINNT < 0x0500"
> so again, I don't think that it should give you trouble with
> _WIN32_WINNT set at 0x0400.
>
> Please try the latest emacs-24 branch, and if it works for you, I will
> close the bug.
>

[-- Attachment #2: Type: text/html, Size: 4208 bytes --]

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[Eli Zaretskii]

> Date: Mon, 19 Nov 2012 13:59:54 +0800
> From: 李丁 <iamliding@gmail.com>
> Cc: 12878@debbugs.gnu.org
> 
> I checked out the latest branch (r 110913), it seems that you miss a
> semicolon at line 761 in w32term.h.

Oops!  Fixed.

> There are more problems:
> 
> 1. I need `nmake bootstrap' but in 24.2 release I can compile directly with
> `nmake'

Do you mean that you need "nmake bootstrap" for _every_ change you
make in any of the files?  If so, please tell what error messages do
you see if you make some change in one C file and run Nmake normally,
without bootstrapping.

> 2. In lisp.h it begins to define ARRAY_MARK_FLAG as PTRDIFF_MIN, however,
>     definition of PTRDIFF_MIN is in stdint.h, which is not included. And
> even if included, in nt/inc/stdint.h,
>     various *_MIN (including PTRDIFF_MIN) are not defined

stdint.h is included by inttypes.h, which _is_ included by lisp.h.  I
added to nt/inc/stdint.h a suitable definition of PTRDIFF_MIN.

> 3. In w32.c DeviceIoControl, it uses a FSCTL_GET_REPARSE_POINT that is only
> defined for _MSC_VER >= 0x0500

I added the missing definitions to w32.c.

> 4. In xdisp.c, start_hourglass function, w32_note_current_window is
> declared (as extern) and used in the middle of
>     the function, causing an error message. Move the declaration to the
> beginning of the function solved the problem.

Fixed.

> After fixing the above problems, finally I can make a temacs.exe, but while
> loading subr.el, temacs reported an
> error message saying `Invalid funtion: "DEAD"'. I do not know where the
> function `dead' comes from.

This indicates some problems with GC, perhaps alignment.  See

  http://debbugs.gnu.org/cgi/bugreport.cgi?bug=10749
  http://lists.gnu.org/archive/html/bug-gnu-emacs/2012-02/msg00565.html

> Maybe you can close the previous bug, but for the latest branch, more fixes
> are needed.

Please try the latest emacs-24 branch.  Revision 110917 shoulod have
the above fixes.

Thanks.

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[Eli Zaretskii]

> Date: Tue, 20 Nov 2012 12:18:25 +0800
> From: 李丁 <iamliding@gmail.com>
> 
> In nt/inc/stdint.h, INTPTR_MIN is still not defined (though PTRDIFF_MIN is
> defined as INTPTR_MIN).

Fixed.

> Tried GC_LISP_OBJECT_ALIGNMENT=1(or 2), still has the `Invalid function
> "DEAD"' problem.

Sorry, can't help you here.

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[Eli Zaretskii]

> Date: Tue, 20 Nov 2012 12:18:25 +0800
> From: 李丁 <iamliding@gmail.com>
> 
> Tried GC_LISP_OBJECT_ALIGNMENT=1(or 2), still has the `Invalid function
> "DEAD"' problem.

This could be due to reallocation of 'environ' in editfns.c.  See the
discussion and the patches in bug #13070.

bug#12878: 24.2; Compiling failed on Windows 7 with VC 11 Express: _WIN32_WINNT version too low

[npostavs]

tags 12878 wontfix
close 12878
quit

> I tried to compile emacs 24.2 on Windows 7 with Visual C++ 11.0
> Express

This building method is no longer supported.