t/uni/overload.t: a7ff7aca38 causes test failures on FreeBSD-13 threaded builds

[jkeenan]

In a FreeBSD-13 VM I use for smoke-testing, I am seeing new failures in t/uni/overload.t on threaded builds. Sample smoke-test report

The failures look like this:

$ uname -mrs
FreeBSD 13.0-RELEASE amd64

$ clang --version
FreeBSD clang version 11.0.1 (git@github.com:llvm/llvm-project.git llvmorg-11.0.1-0-g43ff75f2c3fe)
Target: x86_64-unknown-freebsd13.0
Thread model: posix
InstalledDir: /usr/bin

$ git describe
v5.37.2-181-g9b8ba59f2c

$ ./perl -Ilib -V:config_args
config_args='-des -Dusedevel -Duseithreads -Doptimize=-O2 -pipe -fstack-protector -fno-strict-aliasing';

$ cd t;./perl harness -v uni/overload.t; cd -
ok 1 - length of 'ASCII'
ok 2 - length of 'ASCII'
ok 3 - length of 'ASCII'
[snip]
ok 30 - e acute -> E acute
ok 31
ok 32 - e acute -> E acute
ok 33
not ok 34 - E acute -> e acute
ok 35
not ok 36 - E acute -> e acute
ok 37
not ok 38 - E acute -> e acute
ok 39
not ok 40 - e acute -> E acute
ok 41
not ok 42 - e acute -> E acute
ok 43
not ok 44 - e acute -> E acute
ok 45
not ok 46 - E acute -> e acute
ok 47
not ok 48 - E acute -> e acute
ok 49
not ok 50 - E acute -> e acute
ok 51
not ok 52 - e acute -> E acute
ok 53
not ok 54 - e acute -> E acute
ok 55
not ok 56 - e acute -> E acute
ok 57 - print 
[snip]
ok 190
ok 191
not ok 192 # TODO & SKIP RT #3054: Recursive operator overloading overflows the C stack
not ok 193 - RT \#3270: Overloaded operator that returns an lvalue can be used as an lvalue # TODO RT #3270: Overloaded operators can not be treated as lvalues
# Failed test 34 - E acute -> e acute at uni/overload.t line 114
#      got "�"
# expected "�"
# Failed test 36 - E acute -> e acute at uni/overload.t line 117
#      got "�"
# expected "�"
# Failed test 38 - E acute -> e acute at uni/overload.t line 120
#      got "�"
# expected "�"
# Failed test 40 - e acute -> E acute at uni/overload.t line 125
#      got "�"
# expected "�"
# Failed test 42 - e acute -> E acute at uni/overload.t line 128
#      got "�"
# expected "�"
# Failed test 44 - e acute -> E acute at uni/overload.t line 131
#      got "�"
# expected "�"
# Failed test 46 - E acute -> e acute at uni/overload.t line 136
#      got "�"
# expected "�"
# Failed test 48 - E acute -> e acute at uni/overload.t line 139
#      got "�"
# expected "�"
# Failed test 50 - E acute -> e acute at uni/overload.t line 142
#      got "�"
# expected "�"
# Failed test 52 - e acute -> E acute at uni/overload.t line 147
#      got "�"
# expected "�"
# Failed test 54 - e acute -> E acute at uni/overload.t line 150
#      got "�"
# expected "�"
# Failed test 56 - e acute -> E acute at uni/overload.t line 153
#      got "�"
# expected "�"
Failed 12/193 subtests 

Test Summary Report
-------------------
uni/overload.t (Wstat: 0 Tests: 193 Failed: 12)
  Failed tests:  34, 36, 38, 40, 42, 44, 46, 48, 50, 52
                54, 56
Files=1, Tests=193,  0 wallclock secs ( 0.04 usr  0.00 sys +  0.05 cusr  0.00 csys =  0.09 CPU)
Result: FAIL

Bisecting with the following invocation:

perl Porting/bisect.pl -Duseithreads -Dcc=clang \
--target t/uni/overload.t \
--start bf2a3dae9f4f828fd1f2f8aaf4769f96520c9552 \
--end f83e59d8ea2781ba717b261b07d37a3df68a853c

... pointed to a7ff7ac

a7ff7aca3826600e5832ffe161fe075182f19a82 is the first bad commit
commit a7ff7aca3826600e5832ffe161fe075182f19a82
Author: Karl Williamson <khw@cpan.org>
Date:   Thu Mar 11 12:24:56 2021 -0700
Commit:     Karl Williamson <khw@cpan.org>
CommitDate: Tue Aug 9 18:05:59 2022 -0600

    locale.c: Use setlocale() for init, not P2008

@khwilliamson, can you take a look? Here's some locale data:

$ locale
LANG=C.UTF-8
LC_CTYPE="C.UTF-8"
LC_COLLATE="C.UTF-8"
LC_TIME="C.UTF-8"
LC_NUMERIC="C.UTF-8"
LC_MONETARY="C.UTF-8"
LC_MESSAGES="C.UTF-8"
LC_ALL=

Thank you very much.

[jkeenan]

Status as of now (v5.37.2-208-gbf4a32d613):

$ ./perl -Ilib -V:config_args
config_args='-des -Dusedevel -Duseithreads -Doptimize=-O2 -pipe -fstack-protector -fno-strict-aliasing';

$ cd t;./perl harness uni/overload.t; cd -
uni/overload.t .. 1/193 # Failed test 34 - E acute -> e acute at uni/overload.t line 114
#      got "�"
# expected "�"
# Failed test 36 - E acute -> e acute at uni/overload.t line 117
#      got "�"
# expected "�"
# Failed test 38 - E acute -> e acute at uni/overload.t line 120
#      got "�"
# expected "�"
# Failed test 40 - e acute -> E acute at uni/overload.t line 125
#      got "�"
# expected "�"
# Failed test 42 - e acute -> E acute at uni/overload.t line 128
#      got "�"
# expected "�"
# Failed test 44 - e acute -> E acute at uni/overload.t line 131
#      got "�"
# expected "�"
# Failed test 46 - E acute -> e acute at uni/overload.t line 136
#      got "�"
# expected "�"
# Failed test 48 - E acute -> e acute at uni/overload.t line 139
#      got "�"
# expected "�"
# Failed test 50 - E acute -> e acute at uni/overload.t line 142
#      got "�"
# expected "�"
# Failed test 52 - e acute -> E acute at uni/overload.t line 147
#      got "�"
# expected "�"
# Failed test 54 - e acute -> E acute at uni/overload.t line 150
#      got "�"
# expected "�"
# Failed test 56 - e acute -> E acute at uni/overload.t line 153
#      got "�"
# expected "�"
uni/overload.t .. Failed 12/193 subtests 

Test Summary Report
-------------------
uni/overload.t (Wstat: 0 Tests: 193 Failed: 12)
  Failed tests:  34, 36, 38, 40, 42, 44, 46, 48, 50, 52
                54, 56
Files=1, Tests=193,  0 wallclock secs ( 0.03 usr  0.00 sys +  0.03 cusr  0.02 csys =  0.09 CPU)
Result: FAIL

[jkeenan]

Note: I'm not getting these test failures on unthreaded builds on this VM, nor am I getting them on threaded builds on my regular FreeBSD-12 machine. On the latter, locales are different from the FreeBSD-13 VM:

$ uname -mrs
FreeBSD 12.3-RELEASE-p1 amd64
$ locale
LANG=
LC_CTYPE="C"
LC_COLLATE="C"
LC_TIME="C"
LC_NUMERIC="C"
LC_MONETARY="C"
LC_MESSAGES="C"
LC_ALL=

(I've never consciously modified locales on these machines. AFAICT, these are "out of the box.")

[khwilliamson]

This turns out to be a freebsd bug
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=265950

[jkeenan]

This turns out to be a freebsd bug https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=265950

@khwilliamson, github did not do the right thing with that link. Suggest:
FreeBSD 265950

Is there anything we can do with either that test file or hints/freebsd.sh while we wait for them to repair that bug?

[khwilliamson]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=265950
seems to work

[khwilliamson]

6c72dca should make the test pass

[jkeenan]

6c72dca should make the test pass

Thanks. Since this was a commit to blead, it will get incorporated into perl-5.37.3 dev release later today. I'll close the ticket in a day or two after watching results.

[jkeenan]

6c72dca should make the test pass

Thanks. Since this was a commit to blead, it will get incorporated into perl-5.37.3 dev release later today. I'll close the ticket in a day or two after watching results.

The test is passing, so closing this ticket.