I've been working on a cross compiler that runs in linux and compiles D code for Windows CE/Windows Mobile on the Arm/Xscale processor.  I have a working C/C++ cross compiler that I configured myself, and I am currently building one with D included.

I've also run into a problem that I feel would be best handled by asking
for help here.  While the xcompiler is building, it gets to the point
where it checks for the C standard mmap function.  mmap does not exist.
  Then the make fails.  I looked at phobos, and it seems like mmap isn't
entirely essential because most of the code can use equivelant Windows
calls, which are available in WinCE.  One exception to that seems to be
in phobos/internal/gc/gcgcc.d, where the alternative to "version
(GC_Use_Alloc_MMap)" has //TODO written in it and no code to show.  I
suppose there could be more if I haven't looked in enough files.

This who thing raises a few questions for me:
- When checks result in "no", why do some cause errors and failure while
others do not?  Is it that only the things absolutely essential to D's
functioning will throw errors in the make process?
- If mmap is not needed or I modify phobos to fix this, how do I get the
configure script/makefile to adjust to this?
- If I modify gdc's phobos (count on it), should I submit my changes
somewhere?  If so, how do I go about that?
- Is this thing even compiling (or going to compile) phobos as a Windows
version, or is it doing something unfortunate for me like assuming the
target is linux?

It might also be nice if someone with more knowledge of this stuff would handle the lack of an alternative to mmap in that gcgcc.d file, if it's necessary.  I will probably try to do it myself pretty soon, but I may get it wrong.

OK so now some possibly helpful info...

This is the configure I used:
sh /root/gcc/gdc-3.3.3/configure \
   --prefix=/usr/local/arm-wince-pe \
   --disable-nls \
   --disable-shared \
   --enable-multilib=no \
   --enable-languages=c,d,c++ \
   --with-build-time-tools=/usr/local/arm-wince-pe/bin \
   --with-as=/usr/local/arm-wince-pe/bin/arm-wince-pe-as \
   --with-ld=/usr/local/arm-wince-pe/bin/arm-wince-pe-ld \
   --with-sysroot=/usr/local/arm-wince-pe \
   --enable-phobos-config-dir=/usr/local/arm-wince-pe/phobos-config \
   --target=arm-wince-pe

Stuff that make printed out, as well as the libphobos config.log are attached.

Chad J > wrote:
> I've been working on a cross compiler that runs in linux and compiles D code for Windows CE/Windows Mobile on the Arm/Xscale processor.  I have a working C/C++ cross compiler that I configured myself, and I am currently building one with D included.
> 
> I've also run into a problem that I feel would be best handled by asking for help here.  While the xcompiler is building, it gets to the point where it checks for the C standard mmap function.  mmap does not exist. 

mmap isn't a standard C function, it's a POSIX function.  Slight difference, but important.

>  Then the make fails.  I looked at phobos, and it seems like mmap isn't entirely essential because most of the code can use equivelant Windows calls, which are available in WinCE.  One exception to that seems to be in phobos/internal/gc/gcgcc.d, where the alternative to "version (GC_Use_Alloc_MMap)" has //TODO written in it and no code to show.  I suppose there could be more if I haven't looked in enough files.

It sounds like you're using the GDC Phobos library--the DMD version doesn't have that file.  If you're targeting WinCE, the compiled code will probably be linking to the DMD Phobos.  It has all the necessary Windows calls in place.  Alternately, you may simply have to move some DMD Phobos code into GDC or compile a slightly different set of object files.  If it's any help, the GC files needed for Windows are:

    gc.obj
    gcbits.obj
    gcx.obj
    win32.obj

gcstats.obj is required as well, but that should build as a part of phobos/std.


Sean

Sean Kelly wrote:
> Chad J > wrote:
> 
>> I've been working on a cross compiler that runs in linux and compiles D code for Windows CE/Windows Mobile on the Arm/Xscale processor.  I have a working C/C++ cross compiler that I configured myself, and I am currently building one with D included.
>>
>> I've also run into a problem that I feel would be best handled by asking for help here.  While the xcompiler is building, it gets to the point where it checks for the C standard mmap function.  mmap does not exist. 
> 
> 
> mmap isn't a standard C function, it's a POSIX function.  Slight difference, but important.
> 
>>  Then the make fails.  I looked at phobos, and it seems like mmap isn't entirely essential because most of the code can use equivelant Windows calls, which are available in WinCE.  One exception to that seems to be in phobos/internal/gc/gcgcc.d, where the alternative to "version (GC_Use_Alloc_MMap)" has //TODO written in it and no code to show.  I suppose there could be more if I haven't looked in enough files.
> 
> 
> It sounds like you're using the GDC Phobos library--the DMD version doesn't have that file.  If you're targeting WinCE, the compiled code will probably be linking to the DMD Phobos.  It has all the necessary Windows calls in place.  Alternately, you may simply have to move some DMD Phobos code into GDC or compile a slightly different set of object files.  If it's any help, the GC files needed for Windows are:
> 
>     gc.obj
>     gcbits.obj
>     gcx.obj
>     win32.obj
> 
> gcstats.obj is required as well, but that should build as a part of phobos/std.


PocketPC compiler? That's great to hear!

To make it easier on yourself, you might try to use Ares instead of Phobos? It's much smaller, much more concise, and Sean is very repsonsive in resolving any problem or question.

http://dsource.org/projects/ares

2 Cents

Sean Kelly wrote:
> 
> mmap isn't a standard C function, it's a POSIX function.  Slight difference, but important.
> 

Ah, good to know.

> 
> 
> It sounds like you're using the GDC Phobos library--the DMD version doesn't have that file.  If you're targeting WinCE, the compiled code will probably be linking to the DMD Phobos.  It has all the necessary Windows calls in place.  Alternately, you may simply have to move some DMD Phobos code into GDC or compile a slightly different set of object files.  If it's any help, the GC files needed for Windows are:
> 
>     gc.obj
>     gcbits.obj
>     gcx.obj
>     win32.obj
> 
> gcstats.obj is required as well, but that should build as a part of phobos/std.
> 
> 
> Sean

Interesting.  I'd like to try that.  One problem still remains though, how do I suppress the checking of functions like mmap so that I can build phobos and install the compiler?

It also makes me wonder why exactly there is a different version of phobos for GDC.

Chad J > wrote:
> 
> One problem still remains though, how do I suppress the checking of functions like mmap so that I can build phobos and install the compiler?
> 

Nevermind the suppression of checking, I figured that out.  I ran into another issue, probably also related to the configure scripts.

I managed to get it to start compiling phobos by commenting out some stuff in the d/phobos/configure script, and also had to add a case for the fragment files (it would only take premade frag- files from mingw... I had to change that).  At that point I noticed that $target_os was set to "pe".  I'm not sure, but that looks like it could cause issues.  When I got phobos to start compiling, it ran into some errors, namely missing definitions for wchar_t in std/c/process.d.  I noticed that those definitions were OS dependant.  So I plopped the following into the std/c/process.d file:

version( Win32 )
{
    pragma(msg, "OS version is Win32");
}
else version( Windows )
{
    pragma(msg, "OS version is Windows");
}
else version( linux )
{
    pragma(msg, "OS version is linux");
}
else version( Unix )
{
    pragma(msg, "OS version is Unix");
}
else
{
    pragma(msg, "OS version is unknown!");
}

Please let me know if I missed an important version.

I ran make again, and surely enough it printed out "OS version is unknown!".

I'd imagine not knowing what the OS is would kill a phobos build very fast.  I also wonder if this problem is caused by what $target_os was set to in the configure script.  I just looked and I don't see a GCC configure switch that lets me specify what the name of the OS is, and to do that so globally might be a bad idea anyways.

So that leaves me with another question, how would I go about setting the reserved version identifiers for the D compiler?

At some point I'll probably look through the configure scripts some more, but I find those things really difficult to follow at times, so it'd help if I at least knew where to look.

I have an update on this:

I managed to get version(Win32) to work by adding a case in d/target-ver-syms.sh for "pe".  While I was at it I noticed the target cpu list, and add in a case for "arm" that makes version(ARM) work.

version(Windows) is still broken though.

I have an update on this:

I managed to get version(Win32) to work by adding a case in
d/target-ver-syms.sh for "pe".  While I was at it I noticed the target
cpu list, and add in a case for "arm" that makes version(ARM) work.

version(Windows) is still broken though.

edit:  I managed to get version(Windows) to work as well.