support arm64 architecture

[longquan7]

I build the docker-perl/5.026.000-64bit/Dockerfile from arm64v8/buildpack-deps:stretch and make success,so I think we can add it to support arm64.

[zakame]

Thanks @longquan7 for the PR!

I took a deeper look at how multiple-architecture images can be supported and it seems we can specify additional Architectures in library/perl to include arm64v8. For now though, I'll have to verify first if our perls build and completely pass tests before proceeding.

[longquan7]

@zakame thank you, if you build and pass tests please let me know.

[zakame]

I got these test failures on my system building perl-5.26.0 on Ubuntu 16.04 aarch64 QEMU/KVM (closest I could get to an arm64v8:

Test Summary Report
-------------------
re/fold_grind.t                                                  (Wstat: 9 Tests: 0 Failed: 0)
  Non-zero wait status: 9
  Parse errors: No plan found in TAP output
re/pat_psycho.t                                                  (Wstat: 9 Tests: 13 Failed: 0)
  Non-zero wait status: 9
  Parse errors: Bad plan.  You planned 15 tests but ran 13.
Files=2560, Tests=1230777, 19982 wallclock secs (6029.22 usr 1776.39 sys + 29608.85 cusr 6851.37 csys = 44265.83 CPU)
Result: FAIL

Will be looking into it further (perhaps find myself a proper arm64v8 hardware and try to test it from there.)

[zakame]

Running re/pat_psycho.t on its own seems to make it pass:

root@0f2171d2f275:/usr/src/perl/t# LD_LIBRARY_PATH=.. ./perl harness re/pat_psycho.t 
re/pat_psycho.t .. ok     
All tests successful.
Files=1, Tests=15, 127 wallclock secs ( 0.28 usr  0.07 sys + 125.83 cusr  0.54 csys = 126.72 CPU)
Result: PASS

Which leads me to think it may be just a system load issue (since the tests are normally run with TEST_JOBS=$(nproc) tests in parallel, instead of one at a time.)

For re/fold_grind.t it seems that it happens too in sparc64, per this Debian bug from July. Running it on its own still fails:

root@0f2171d2f275:/usr/src/perl/t# LD_LIBRARY_PATH=.. ./perl harness re/fold_grind.t 
re/fold_grind.t .. 1/? # Test process timed out - terminating
re/fold_grind.t .. All 116 subtests passed 

Test Summary Report
-------------------
re/fold_grind.t (Wstat: 9 Tests: 116 Failed: 0)
  Non-zero wait status: 9
  Parse errors: No plan found in TAP output
Files=1, Tests=116, 301 wallclock secs ( 0.45 usr  0.08 sys + 300.36 cusr  0.33 csys = 301.22 CPU)
Result: FAIL

[longquan7]

@zakame this problem influence that the image suport arm64 architecture?

[zakame]

@longquan7 at least for myself, I won't be able to support it properly if I can't replicate issues (or have access to an arm64 device.) I'll resolve this first somehow.

[longquan7]

@zakame OK,thanks

[zakame]

For re/fold_grind.t a workaround is to increase the timeout via the PERL_TEST_TIME_OUT_FACTOR environment variable:

root@0f2171d2f275:/usr/src/perl/t# PERL_TEST_TIME_OUT_FACTOR=2 ./perl harness re/fold_grind.t 
re/fold_grind.t .. ok     
All tests successful.
Files=1, Tests=684, 549 wallclock secs ( 1.67 usr  0.10 sys + 547.65 cusr  0.60 csys = 550.02 CPU)
Result: PASS

[longquan7]

@zakame that's great. Is there anything else besides this problem

[zakame]

I tested on Packet.net's Type 2A servers and both 5.26{,-threaded} seem to build and test fine. Now checking on 5.24 and 5.22.

[zakame]

I can confirm that currently supported Perl versions (5.22 through 5.26) can build and test successfully on aarch64/arm64v8. I'll put an update to https://github.com/docker-library/official-images to include building arm64v8 in the perl image builder.

Thanks again!

[zakame]

Closing here as the needed change is on https://github.com/docker-library/official-images/pull/3373/files#diff-600942e61fee1f4174da5a5a7f0eac09

[longquan7]

@zakame OK.thanks