"error: 'JIT_LOCAL4' undeclared" on OpenBSD/powerpc

[juanfra684]

When I try to build Racket 6.4 on OpenBSD/powerpc with GCC 4.9.3 and jit/places/futures enabled, it shows this error during the build:

cc -O2 -pipe    -I./.. -I/usr/ports/pobj/racket-minimal-6.4/racket-6.4/src/racket/src/../include -I/usr/local/include -I/usr/X11R6/include -pthread -I/usr/local/include  -DUSE_SENORA_GC  -c /usr/ports/pobj/racket-minimal-6.4/racket-6.4/src/racket/src/hash.c -o hash.o
cc -O2 -pipe    -I./.. -I/usr/ports/pobj/racket-minimal-6.4/racket-6.4/src/racket/src/../include -I/usr/local/include -I/usr/X11R6/include -pthread -I/usr/local/include  -DUSE_SENORA_GC  -c /usr/ports/pobj/racket-minimal-6.4/racket-6.4/src/racket/src/jit.c -o jit.o
/usr/ports/pobj/racket-minimal-6.4/racket-6.4/src/racket/src/jit.c: In function 'scheme_generate':
/usr/ports/pobj/racket-minimal-6.4/racket-6.4/src/racket/src/jit.c:2469:32: error: 'JIT_LOCAL4' undeclared (first use in this function)
           mz_ld_runstack_base_alt(JIT_RUNSTACK);
                                ^
/usr/ports/pobj/racket-minimal-6.4/racket-6.4/src/racket/src/jit.c:2469:32: note: each undeclared identifier is reported only once for each function it appears in
/usr/ports/pobj/racket-minimal-6.4/racket-6.4/src/racket/src/jit.c: In function 'generate_function_getarg':
/usr/ports/pobj/racket-minimal-6.4/racket-6.4/src/racket/src/jit.c:3465:26: error: 'JIT_LOCAL4' undeclared (first use in this function)
     mz_set_local_p(JIT_RUNSTACK, JIT_RUNSTACK_BASE_LOCAL);
                          ^
*** Error 1 in racket/src (Makefile:217 'jit.o')
*** Error 1 in racket/src (Makefile:136 'all')

Unlike the default compiler on OpenBSD (GCC 4.2), GCC 4.9.3 supports thread-local storage (emulated).

[juanfra684]

I need a fix for this bug. I changed the compiler for the OpenBSD port to GCC 4.9 because I would like to enable places and futures for the architectures with JIT support. I'm holding off the update to 6.4 due to this bug. If you can't fix it just tell me and I will disable places and futures for this architecture.

[mflatt]

Sorry for not getting back to you sooner. Futures and places do not currently work with PPC, so disabling them is the right choice.

It's possible that places could be made to work with a modest effort. Making futures work would be much more difficult, I think, because lots more synchronization is needed for a non-TSO platform.

[juanfra684]

Thanks for the info. I thought that the "error" was something more simple. Don't waste your time working in places/futures for powerpc, in my experience very few users use a openbsd/powerpc machine for real work. Anyway, I updated the OpenBSD package to 6.4 just 5 days after posting my problem here.

I have some problems with Racket 6.5 with JIT enabled on powerpc. Do you know how to fix this error in number.rktl?:

(#<procedure:floor> +inf.0) ==> +inf.0
(#<procedure:ceiling> +inf.0) ==> +inf.0
(#<procedure:round> +inf.0) ==> +nan.0
  BUT EXPECTED +inf.0
(#<procedure:truncate> +inf.0) ==> +inf.0
(#<procedure:floor> -inf.0) ==> -inf.0
(#<procedure:ceiling> -inf.0) ==> -inf.0
(#<procedure:round> -inf.0) ==> +nan.0
  BUT EXPECTED -inf.0
(#<procedure:truncate> -inf.0) ==> -inf.0
(#<procedure:floor> +nan.0) ==> +nan.0
(#<procedure:ceiling> +nan.0) ==> +nan.0
(#<procedure:round> +nan.0) ==> +nan.0
(#<procedure:truncate> +nan.0) ==> +nan.0

I will open new bug reports in the next weeks for the other problems.

[mflatt]

I don't think I ever investigated these for BC, but CS is meant to work on FreeBSD, now. I say "meant to work" because the configuration is now in place, but I haven't tried a build myself. (I have tried FreeBSD on AArch64 and NetBSD on PowerPC32.)

[juanfra684]

Don't worry, Racket has been building fine for years on OpenBSD/macppc with places and futures disabled. I'm closing the issue because it is very old. I will open new bug reports if I find problems with CS on other architectures.

Thanks for taking caring of the less popular architectures (and OS).