Re: Building progress.

[Jean-Baptiste Boric <jblbeurope%gmail.com@localhost>]

> >> [%] If nothing else, I consider gcc 4.9.0 unsuitable for open-source
> >>     work such as I am interested in doing.
> > As far as I know, GCC 4.9 is the only modern compiler that can target
> > the R5900 inside the Playstation 2 for now.  So we're stuck with it
> > for the time being.
> 
> How did the playstation2 ports in NetBSD 4.* and 5.* work, then?  Or
> did they never work because they never had a working compiler?

The PlayStation 2 port of NetBSD was compiled with Sony's heavily patched
GCC 2.95 as an external toolchain.

The ports did seemed to work a long time ago, but they were abandoned when
it was becoming truly impossible to compile NetBSD with such a hopelessly
out-of-date GCC by then.

> In any case, if 4.9 is the only known compiler capable of generating
> suitable code (something which remains to be determined as far as I
> know as I write this), I guess I'll have to abandon NetBSD/ps2 stuff
> until I find/create another one; you may be willing to ignore the
> licensing issues, but I am not.

If I recall correctly, Sony used a heavily patched GCC 2.95 in the official
SDK (and Linux PS2 Kit). The PS2DEV guys seems to use GCC 3.2.2, with a
+5000 lines patch for GCC and ~12000 lines patch for Binutils. And we have
GCC 4.9 now.

In theory, you might get away with it by using another MIPS compiler (maybe
LLVM?) that can generate a STRICT subset of the R5900's instruction set.
That might work for userspace, but if this worked for kernel stuff then the
port wouldn't have died just of toolchain issues.

As for licensing issues, except for touching one line in GCC to deactivate
fixincludes (and yet that part is actually public domain), I didn't touched
GPL stuff. As for philosophical incompatibilities with GPL, your only
practical solution would be to add R5900 support to LLVM.