Bug 2380086

Summary: perl-BSON-1.12.2-16.fc43 FTBFS: t/mapping/double.t fail with perl 5.42.0
Product: [Fedora] Fedora Reporter: Petr Pisar <ppisar>
Component: perl-BSONAssignee: Petr Pisar <ppisar>
Status: CLOSED RAWHIDE QA Contact: Fedora Extras Quality Assurance <extras-qa>
Severity: unspecified Docs Contact:
Priority: unspecified    
Version: rawhideCC: perl-devel, ppisar
Target Milestone: ---   
Target Release: ---   
Hardware: Unspecified   
OS: Unspecified   
URL: https://koschei.fedoraproject.org//package/perl-BSON
Whiteboard:
Fixed In Version: perl-BSON-1.12.2-17.fc43 Doc Type: ---
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2025-07-15 12:12:34 UTC Type: Bug
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Bug Depends On:    
Bug Blocks: 2339432    

Description Petr Pisar 2025-07-15 07:00:07 UTC 
perl-BSON-1.12.2-16.fc43 fails to build in Fedora 43 because tests fail like this:

t/mapping/dbref.t ................ ok
Possible precedence problem between ! and string eq at /home/test/fedora/perl-BSON/perl-BSON-1.12.2-build/BSON-v1.12.2/blib/lib/BSON.pm line 99.
t/mapping/decimal128.t ........... ok
Possible precedence problem between ! and string eq at /home/test/fedora/perl-BSON/perl-BSON-1.12.2-build/BSON-v1.12.2/blib/lib/BSON.pm line 99.

#   Failed test 'Inf as double->double'
#   at t/mapping/double.t line 55.
#          got: 'NV'
#     expected: 'PVNV'

#   Failed test '-Inf as double->double'
#   at t/mapping/double.t line 55.
#          got: 'NV'
#     expected: 'PVNV'

#   Failed test 'NaN as double->double'
#   at t/mapping/double.t line 55.
#          got: 'NV'
#     expected: 'PVNV'

#   Failed test 'Inf as BSON::Double->BSON::Double'
#   at t/mapping/double.t line 69.
#          got: 'NV'
#     expected: 'PVNV'

#   Failed test '-Inf as BSON::Double->BSON::Double'
#   at t/mapping/double.t line 69.
#          got: 'NV'
#     expected: 'PVNV'

#   Failed test 'NaN as BSON::Double->BSON::Double'
#   at t/mapping/double.t line 69.
#          got: 'NV'
#     expected: 'PVNV'
# Looks like you failed 6 tests of 39.
t/mapping/double.t ...............
Dubious, test returned 6 (wstat 1536, 0x600)
Failed 6/39 subtests

This is probably triggered by upgrading perl from 4:5.40.2-517.fc43 to 4:5.42.0-519.fc43.
Comment 1 Petr Pisar 2025-07-15 08:50:01 UTC 
The warning is unrelated.

The failure appeared in 5.41.7 and this perldelta seems related:

        When assigning from an SVt_IV into a SVt_NV (or vice versa),
        providing that both are "bodyless" types, Perl_sv_setsv_flags will
        now just change the destination type to match the source type.
        Previously, an SVt_IV would have been upgraded to a SVt_PVNV to
        store an NV, and an SVt_NV would have been upgraded to a SVt_PVIV to
        store an IV. This change prevents the need to allocate - and later
        free - the relevant body struct.

That probably corresponds to perl commit fbebf96ce083dc15b63c04d5695997d6e2657b03 ("Perl_sv_setsv_flags: handle mixed IV and NV fast case").


When minimizing the failing test too much:

  my $hash = decode( encode( { A => unpack("d<",pack("H*","000000000000f0ff")) } ) );
  is( ref(B::svref_2object( \$hash->{A} )), 'B::PVNV', "double->double" );

the type becomes NV not only on perl 5.42.0 but also on perl 5.40.2.

If one calls:

  my $hash2 = decode( encode( { A => 3.14159 } ), wrap_numbers => 1);

before the test, the test returns PVPN on old perl.

I conclude the test or the BSON code builds on a side effect of the interpreter implementation which does not hold true anymore.

Normally I would adjust the test, but here BSON task is to reproduce Perl data structures and it does not reproduce them anymore.

I'm keen to remove perl-BSON and perl-Mongo from Rawhide. It's unmaintained since 2020 by upstream.
Comment 2 Petr Pisar 2025-07-15 08:56:41 UTC 
Actually BSON::decode(, wrap_numbers => 1) is documented to "ensure fields can round-trip if unmodified". So any decoding without wrap_numbers == 1 does not guarantee preserving the type. Maybe the change in the decoding is acceptable.
Comment 3 Petr Pisar 2025-07-15 09:21:35 UTC 
Actually new perl behaves better: Input in the test is NV. Output in the new perl is also NV, while in the old perl it was PVNV.