Better GDC/mingw32 support

h3r3tic added a patch to patches/

What is needed for Tango itself and/or build system these days to have Tango fully working on Mingw?

Patch for building with mingw32

I've attached a patch for a small bug when compiling tango with mingw32. It assumes that only Linux systems use GNU compilers.

The patch switches a version block from "linux" to "GNU" since the code applies to GNU compilers and not just Linux systems.

I also have another problem with Threads and mingw32. The assembly code in lib/common/tango/core/Thread.d references the FS segment, which gdc doesn't like for some reason:

/tmp/ccpWsEPO.s: Assembler messages: /tmp/ccpWsEPO.s:139: Error: bad memory operand `' /tmp/ccpWsEPO.s:153: Error: bad memory operand `'

Commenting out the 6 references to FS (3 push and 3 pop) gets it to compile, but I suspect this isn't the correct fix. I haven't used threads yet to verify that it crashes.

I have quite a bit of assembly experience in the past, but not with GCC assembly so I'm not sure what the problem is. It might be a bug in GDC.

see patches/build-posix-mingw.patch for the trick for compiling Thread.d

Those were the only code problems I've found when building tango with mingw32.

However, cross-compiling is a *huge* mess. All of the GDC scripts and makefiles assume Posix which obviously isn't correct for mingw32. Also there are quite a few places where binary names are hardcoded when it needs a prefix for cross compiling.

Before I start digging into the build system, would there be any arguments against a patch that makes a significant change to handle cross-compiling correctly? I think much of the makefiles can stay, but the GDC build scripts will need some major overhaul.

jerdfeldt: great effort! as for cross-compilation, there is the build-gdc-x.sh script, is there very much more work than that needed?

When that is said, I'm interested in improvements to the build scripts, but preferably without any increase in (outwards) complexity.

Also, as for the version (linux) to version (GNU) - note that linux also means DMD on Linux - so just that switch may not be entirely correct?

Looked at the Thread patch - it seems to remove ThreadASM from the build, then modify the fs stuff a tad? It is very much defined as a hack there though.

The build-gdc-x.sh script has a few problems.

The first being it doesn't set the AR make variable so the built static libraries don't have the correct indexes for the symbols in the archive (since my Linux system ar doesn't understand windows objects I guess).

The second, which is the bigger issue, is the use of the gdc-posix makefile. Windows is not Posix and as a result, lots of hacking needs to be done to change Posix version references to mingw32.

Also, the build-tango.sh script assumes the local system is the target system, which is very much incorrect. Removing the use of uname is necessary here.

Much of this can be fixed. I was planning on merging build-gdc-x.sh into build-gdc.sh and fixing up build-tango.sh and install-gdc.sh similarly.

I would very much appreciate patches in this direction.