lang/ghc: port to powerpc64

[arrowd]

Summary

bootstrap for 12.0 available here http://mikael.urankar.free.fr/FreeBSD/powerpc64/ghc-8.6.3-boot-powerpc64-freebsd.tar.xz

Patches come from https://gitlab.haskell.org/trommler/ghc/commits/wip/T15916
Most of them are already commited upstream.

these 2 patches are not yet committed:
https://ghc.haskell.org/trac/ghc/ticket/15411 (Disable-unboxed-arrays.patch)
https://gitlab.haskell.org/trommler/ghc/commit/19731a77ed203870f76a53eaf01758efbb5144d3
For the record, upstream PR for FreeBSD/ppc64: https://ghc.haskell.org/trac/ghc/ticket/15916

Test Plan

ok on 12.0 ppc64, 8.6.2 and 8.6.3 (the 8.4.x branch is too old)
ok on amd64

the 'llvm backend' is broken (not sure if it's supposed to work, I'll submit a pr upstream)

[arrowd]

lang/ghc/bsd.ghc.mk

13 ↗

(On Diff #53037)

Why do we need that?

[mat]

lang/ghc/Makefile

21 ↗

(On Diff #53037)

Why +=?

lang/ghc862/Makefile

21 ↗

(On Diff #53037)

Why +=?

[mikael]

lang/ghc/Makefile

21 ↗

(On Diff #53037)

I don't know, it seems weird to have ONLY_FOR_ARCHS= powerpc64, it looks like the port is restricted to ppc64 until you look at bsd.ghc.mk

lang/ghc/bsd.ghc.mk

13 ↗

(On Diff #53037)

I need a recent compiler, I probably just need compiler:c11 (or put use_gcc in the ppc64 'if block')

[mikael]

address comments

[tcberner]

lang/ghc/Makefile

18 ↗

(On Diff #53111)

^ what?

[tcberner]

lang/ghc/bsd.ghc.mk

8 ↗

(On Diff #53111)

^ why not add powerpc64 here? instead having that confusing line in lang/ghc/Makefile

[mikael]

lang/ghc/Makefile

18 ↗

(On Diff #53111)

mat didn't like the ONLY_FOR_ARCHS+= ppc64 notation.
you don't like ONLY_FOR_ARCHS= ppc64
should I remove O_F_A in bsd.ghc.mk and list the allowed arches in each Makefile?

lang/ghc/bsd.ghc.mk

8 ↗

(On Diff #53111)

It's only for ghc-8.6.x

[mat]

lang/ghc/Makefile

18 ↗

(On Diff #53111)

I did not say I did not liked it, I asked why, because the variable was not set before, and doing += means you are adding to it, which was weird, because it was not defined before.

You should add a comment saying other archs are added later, because it looks weird like this.

[mikael]

address comment

[tcberner]

Looks good to me.

[arrowd]

Rework ONLY_FOR_ARCHS handling.

[mat]

lang/ghc/bsd.ghc.mk

183 ↗

(On Diff #54412)

I wonder if this should not be changed to positive tests with the architectures that actually need this instead of excluding the ones that do not.

239–240 ↗

(On Diff #54412)

Same here.

[arrowd]

Simplify checks.

[arrowd]

I want to note that this Diff was about adding powerpc64 support to lang/ghc. It is seems to be moving into "refactor bsd.ghc.mk" direction now. I admit there is much to be done here, but it can be done iteratively and later.

[arrowd]

Fix condition.

[tcberner]

lang/ghc/bsd.ghc.mk

91 ↗

(On Diff #54569)

For a future revision, couldnt we do something like

_BOOT_GHC_VESRSION_amd64= 8.6.4
_BOOT_GHC_VESRSION_i386=8.6.4
[...]
_BOOT_GHC_VERSION_${ARCH}?=8.4.3
BOOT_GHC_VERSION=${_BOOT_GHC_VERSION_${ARCH}}

[arrowd]

Final fix.

[linimon]

You seem to have forgotten adjusting ONLY_FOR_ARCHS :-)

I've fixed that locally and will see what happens overnight.

[arrowd]

In D18886#415682, @linimon wrote:

You seem to have forgotten adjusting ONLY_FOR_ARCHS :-)

Hum? Where?

[tcberner]

lang/ghc/bsd.ghc.mk

8 ↗

(On Diff #54571)

^ this seems to breaks it:

you cannot include bsd.port[.pre].mk twice

[arrowd]

Bring IGNORE= line back to its place, because checking for OSVERSION at the
top requires including bsd.port.pre.mk, but it should be included after all
other variables are set.

[linimon]

In D18886#415699, @arrowd wrote:

Hum? Where?

OK, I understand it now. The assingments were confusing.

[tcberner]

looks good to me again