This Week on perl5-porters

In the previous weeks it came to pass that Jerry D. Hedden started working on test.pl to give it an overhaul. The latest version made it into blead. Of great interest is a watchdog timer, that will allow tests that can stumble around in the dark to be killed after an appropriate amount of time.

watchdog() some IO tests

threads::shared 1.25

B-Debug-1.11 - B::Flags CORE test fix

[perl5db.t should not load Term::ReadLine::Gnu

threads::shared 1.25

Correction to change 34138

New and old bugs from RT

Security of ARGV using 2-argument open - It's a feature (#2783)

Joshua ben Jore made a fine argument for axing the too-clever behaviour of open by ignoring the shellish nature of such characters as <, > and |. This was judged as being a little to backwards incompatible in 5.10, but could be considered for 5.12.

Storable seg faults under perl 5.8.0 solaris, maximal 64bit (#18049)

Bram wanted to know if the problem was caused by defining use64bitall, since the other combinations smoke cleanly for the Solaris CPAN testers.

This problem seems to be fixed and a small patch for B::Xref (#18306)

Bug fix in malloc.c (#50856)

Mark Noyes filled in details about this problem, where it transpires that different parts of a program are calling Perl's malloc and the underlying libc malloc. It nearly works, except for a minor detail regarding the rounding-up alignment of the different source blocks.

Feature: regexp modifier to disable interpolation like m'' (#54702)

Yves Orton grudgingly admitted that it adding a modifier to a pattern, such as /$foo/d, to suppress interpolation, is doable. In which case the above would be interpreted as an end-of-line followed by literal 'foo'.

Panic in regex processing (unknown regstclass 0) (#55466)

Yves Orton suggested if that someone could isolate the regular expression in the Net::IP test suite that was giving 5.10 grief, then he would take a look and see if he could fix it.

'once' warnings from xs modules in perl 5.10.0 (#55550)

Rafa�l Garcia-Suarez didn't like Chia-liang Kao's idea of making DynaLoader load warnings and instead suggested a GV_ADDMULTI manoeuvre.

overloading and formats (#55668)

Nicholas Clark demonstrated the interaction between formats and variables (the former consume the latter) and suggested that any attempt to introduce string overloading into the picture was destined to end in tears.

Perl 5.10 breaks \N{} in regex interpolated inside regex (with charnames) (#56444)

YVes Orton gave the rationale behind the change (speed) that caused the above failure to occur. That said, he had no idea what the underlying cause of the problem was, and suggested someone fire up a debugger and see why charnames was being descoped.

m/a{1,0}/ compiles but doesn't match a literal string (#56526)

Yves Orton confirmed that Michael Cartmell's reasoning was sound, as was the patch. As such, someone should go ahead and apply it.

pad_alloc and fold_constants are not members of public API (#56536)

Dmitry Karasik wanted to build B::Generate on MSWin32, but the above two API functions need to go public. Rafa�l agreed to letting fold_constants out into the light, but declined pad_alloc, as it's too low-level.

5.10.0 crash on serialised v-strings (#56606)

John Peacock got to work to fix Data::Dumper breakage when serialising with version objects. Due to the incestuous nature of CPAN.pm, the core and the duality of version.pm, he faced considerable difficulty in getting the change to be heeded. In the end he tracked down the CPAN authors who use Data::Dumper to serialise version numbers (in META.yml files) and suggested that an upgrade of their own would be a wise course of action.

Math::BigInt gives completely (bogus results) when Math::BigInt::GMP is used (#56608)

Marc Lehmann suggested there was a problem; Bram and Tels wanted to know what version he was using before jumping to conclusions. Tels was pretty sure that the symptoms pointed to a signed/unsigned mismatch.

Perl 5.10.0 double-free bug with Regular Expressions (#56610)

Andris Kalnozols pointed to a new bug in 5.10 regarding tied hashes. Chris Madsen slimmed the test case down considerably.

PerlIO resource leaks on open() and then :pop in :unix and :stdio (#56644)

Goro Fuji nailed down some leaks in PerlIO with some patches and tests, attributing the problems to (ahem) "a bad design of PerlIO virtual table".

using $1 evaluates strangely in list context when combined with ?: operator (#56674)

Not a bug when you can describe the reason behind the behaviour, as chromatic did, with some additional material from Aristotle Pagaltzis.

Some bugs in Perl regexp (core Perl issues) (#56690, #56738)

Serge discovered that print "ab" =~ /^a?(?=b)b/ prints nothing, instead of 1, which is rather distressing. A couple of variations on the theme also produced bad results. Abigail could do little more than write some TODO tests that Steve Peters applied.

He also found another bug where $^N and $1 disagreed about what had just been matched. Yves Orton pointed out that as the match used a quantified capture buffer (as in /(foo){2}/), the result is pretty much undefined in the first place (and fixing it would be very difficult).

5.10.0 Attribute::Handlers segmentation fault (#56766)

Naquad found a problem, Vincent Pit whipped up a quick patch to fix the problem (which was in fact in Data::Dumper). Now waiting for application.

\w with use locale does not match Russian letters (#56820)

Peter found a problem on Gentoo, not being able to match Russian characters with \w after setting the locale. On the other hand it worked perfectly on FreeBSD.

Perl-5.8.8 compilation on AIX 5.1 (#56826)

Kawaljeet Kaur had problems building perl on AIX, and was unnerved by all the "Note (probably harmless): No library found for -lfoo" messages. Unfortunately, two tests failed in the test suite, which was the really interesting bit, except we didn't get to find out which ones they were.

Andy Dougherty amended the documentation to minimise this possibility for confusion, and earns a gold star for achieving less documentation at the end of the process.

No test coverage for the DO_UTF8() branch of case FF_CHECKCHOP in pp_formline (#56834)

Nicholas Clark found some code that he thought was probably incorrect and certainly inefficient, but was loathe to make the changes without a regression test that actually exercised the code path in question.

He dropped a few hints as to how one might write a test that would hit the right branch.

Taint bug with \G and pos() on hash entries (#56842)

Matt Sergeant found another bug where pos gets lost on the value of a hash key... when taint mode is active.

can't use v[0-9]+ as label (vstring) (#56880)

Trying to drop a v1: as target for a next and its ilk will cause a syntax error (on 5.8 through 5.11).

regex utf8 "uninitialized value" error (#56902)

Ben Bullock related the tale of a regular expression generating spurious "uninitialized value" warnings according to the current setting of UTF-8.

DBI memory leak (#56908)

lav identified a leak in DBI that occurs in 5.10 but not in 5.8. Since DBI plays fast and loose with the internals for speed, and the internals were heavily reworked during the 5.9 development track, it's really not that surprising.

There were no summaries for the two previous weeks as Real Life imposed its own constraints in a big way. I feel very guilty about this, please accept my sincere apologies.

About this summary

If you found this summary useful, please consider contributing to the Perl Foundation or attending a YAPC to help support the development of Perl.