build: fix DESTCPU detection on non-Intel platforms

[richardlau]

Checklist

• the commit message follows commit guidelines

Affected core subsystem(s)

build

Description of change

make binary attempts to auto detect DESTCPU if not set, but was
assuming being on an Intel architecture (i.e. it was checking for x64
and assuming x86 otherwise).

• sample failure for Linux on arm
• sample failure for Linux on ppc64le

[richardlau]

cc @mhdawson

[mhdawson]

Just wondering which case this covers:

+ifeq ($(findstring powerpc,$(shell uname -p)),powerpc) +DESTCPU ?= ppc +else 

otherwise LGTM

[richardlau]

That case covers AIX.

[jbergstroem]

@nodejs/build

[richardlau]

Note about the implementation:
I chose to extend what was already there in the makefile. Using variants of uname I don't see a way of detecting 32-bit/64-bit on AIX.

With regards to host architecture detection there are two other places this is done in the build scripts:

• in host_arch_cc() in configure -- configure sets the target arch to the detected host arch if --dest-cpu is not specified (in the case of make binary, configure is run with --dest-cpu=$(DESTCPU))
• in GuessArchitecture() in tools/utils.py -- This is used by tools/test.py and for some reason only selects 32-bit architectures

[richardlau]

@mhdawson I've updated the commit so on AIX it defaults to 64-bit as we discussed on the phone.

[mhdawson]

LGTM

[mhdawson]

@jbergstroem just wondering if you can take a quick look and let me know if you have any concerns with this.

[joaocgreis]

I can't claim to know enough of this to stamp it, but since it uses only ?= it seems quite safe to me.

However, at least ARM64 is missing:

$ uname -m aarch64

[richardlau]

Thanks, rebased onto latest master and added aarch64.

[MylesBorins]

ci: https://ci.nodejs.org/job/node-test-pull-request/2614/

overall the changes to the makefile LGTM

How was ppc + arm being detected and built correctly before?

[rvagg]

sgtm

might be time to pull out all of this ARCH and DESTCPU stuff into a script (bash or perl) in tools

[jbergstroem]

@rvagg we could use python's os stuff. Would align well with our build stack.

[rvagg]

@jbergstroem sounds good.

To clarify though, that shouldn't hold up this PR if the IBM folks agree.

@thealphanerd we manually specify both ARCH and DESTCPU in all (?) our build machines so we don't get caught up by this, it's obviously a concern for end-users on any of these platforms, however.

[jbergstroem]

@rvagg we don't manually specify it on all our machines; mainly release-related stuff -- I think we should avoid it where possible since we'd be testing the configure process as well. With that said, in cases we have to, we should obviously override (esp DESTCPU).

[jbergstroem]

LGTM from me as well btw. Just haven't tried all variations of uname.

[mhdawson]

Looks like we have enough LGTMs to land, will do that now

[mhdawson]

Landed as 830a726