This Week on perl5-porters - 24-29 February 2008

This Week on perl5-porters - 24-29 February 2008

"Is this a bug? Or why is this the expected behaviour?" -- Steffen Ullrich, playing with signal handlers.

Topics of Interest

use encoding 'utf8' bug for Latin-1 range

The thread about use encoding continued this week. Juerd Waalboer gave one of the best concise explanations as to why the current model Perl uses for dealing with Unicode is broken, which is that the \x hex escape is overloaded for bytes (\x2b versus \x{d0b2}), and that it takes place too early, while the source is being read.

The result of which is that a source code file encoded in an Asian language cannot embed a latin-1 character like an e-acute.

Much discussion of remarkable civility followed, regarding what to do about the matter. Glenn Lindemann put forward the following ideas:

No-one appeared to lament the idea of letting encoding go.

Yves Orton pointed out that Microsoft managed to get their Unicode handling more or less right, albeit at a certain cost to their API, and regretted that Unix-like operating systems supplied the absolute strict minimum, pushing all the work onto each and every client program. Which meant that nothing really worked at all, not even the so-called shebang line.

Juerd and Nicholas put forward that there is a case to be made for perl to figure out itself whether a given source file is in ASCII, Latin-1 or UTF-8. It turns out that it's just about impossible to construct a sensible Latin-1 file that also turns out to be be valid UTF-8. The idea is to start out in 7-bit ASCII and carry on until a byte with the high bit set is encountered.

If this byte introduces a valid UTF-8 character, the rest of the file must be, too. Any invalid byte sequences thereafter trigger a fatal compile-time error. Otherwise it means it must be Latin-1, in which case similar but different rules apply which also cause the compilation to halt if encodings change mid-stream. The key issue is to determine that the encoding does indeed change.

EBCDIC was also mentioned in passing. Sadly, Perl no longer runs on EBCDIC due to a general lack of nurturing. Then again, if it was important, Nicholas felt that someone from IBM would have been in touch at some point.

  for some reason I now have a splitting headache

Interrupting system() with signal depends on signal handler

Steffen Ullrich noticed that an alarm signal handler that does a syswrite as opposed to a print behave differently. After diving in through pp_sys.c, he noticed that he could make the print version (which was working correctly) behave the same incorrect way, by setting $! to undef.

He produced a one-line patch that fixed the behaviour (hmm, did we get a test?) and Rafael applied it as change #33408.

  handle with care

CPAN NetBIOS broadcasts

Linda W was scratching her head wondering why CPAN installations on cygwin were glacially slow. After running a network trace, she discovered that what had been a path /var/cache/cpan was being interpreted as a UNC path (/cache/cpan on host //var).

This caused the local host to send out plaintive calls for host //var to please call home. Michael G. Schwern thought that this sounded like the same problem described in CPAN bug #32813, as did Linda.

Yves Orton, current maintainer of ExtUtils::Install, which is were the problem originated, pushed out a new version and Linda confirmed that it solved the problem.

Ken Williams was not around to comment on how hard it is to use File::Spec correctly.

  not quite Unix, not quite Windows

Google summer of code

Eric Wilhelm got the ball rolling on Perl's participation in Google's Summer of Code project. But you've probably heard about this in other venues. All hail Eric.

The Perl 5 Wiki is place to go for the latest information.

  summertime fun

Patches of Interest

sv.c consting goodness

Steven Schubiger's consting patch number 4 from the beginning of the month was applied. This lead to patches 5, 6, 7, 8 and 9, all applying ever more consting to sv.c being issued by Steven, which in turn were all applied by various porters.

no archlib in otherlibdirs

After some long, hard thought, Andy Dougherty remembered why Reini Urban's plan for organising site and vendor libraries on Cygwin wouldn't work in the general case. So Reini withdrew his patch but would continue to use it locally.

On the other hand, his enhancements to B::Debug made it in.

  win some, lose some

warning message for -M:Foo, extended and revised

Robin Barker finally settled on "Invalid module name :Foo with -M option: contains single ':'", which was good enough for Rafael


More diagnostics for

Slaven Rezic enhanced Fatal to name the builtin that could not be overridden in its dying message.

  if I told you I would have to kill you

Thread patches

Jerry D. Hedden is doing so much work on threads at the moment, he deserves his own section.

First off, the patch to not install threads on non-thread builds was reverted (Michael G. Schwern killer argument being that at least that way you get a nice error message).

Then the CPAN 1.69 version of threads was synch'ed with blead.

As was threads::shared 1.17.

At the end of the week, he also delivered version 1.18, which added some diagnostics to help track down what's going wrong when t/stress.t decides to go belly up.

Moving along, Thread::Semaphore 2.07 checked in.

and last but not least, Thread::Queue 2.06 did too.

Watching the smoke signals

It looked like t/stress.t in the threads module failed, and so Jerry asked if there was any chance of seeing what the new diagnostics had to say. Steve Hay discovered that the problem was in fact a TODO test that had started to pass, and Test::Smoke got confused and recorded it as a failure.

  Smoke [5.11.0] 33390 FAIL(F) MSWin32 WinXP/.Net SP2 (x86/2 cpu)

New and old bugs from RT

Segfault when calling ->next::method on non-existing package (#51092)

David Landgren thought that the test that Rafael Garcia-Suarez added as part of the fix for this bug should have had the RT bug number embedded in it somewhere. In other other news, we discovered that there are 485 subscribers to perl5-porters.

Perl5 Bug Summary

  288 new + 1500 open = 1788 (+3 -2)

New Core Modules

ExtUtils::Install version 1.45

This was the fix for the //var problem noted by Linda W. (But stay tuned next week for exciting new developments).
ExtUtils::MakeMaker 6.44

Michael G. Schwern rolled out 6.34_01 plus Yves's EU::I 1.45 as version 6.44. Other assorted bugfixes made it in, but Michael announced that he had declined to put in the fixes required to make paths with whitespace work correctly, saying that he wanted to think about a better solution.

In Brief

Last week, Jim Cromie had the newfound ability to hook XML analysis to a test suite (via the PERL_XMLDUMP environment variable). This week, Jim wrote a patch to test -Dmad's PERL_XMLDUMP= output. It was not applied.

  truly madly

On the other hand, Rafael did apply his optimisation of the OP_IS_(FILETEST|SOCKET) macros, with some OP */int fuzz.

The exact recipe for signalling a non-met prerequisite (such that a perl build without threads should not attempt to require threads) was nailed down and codified on the CPAN Testers wiki.

Salvador Fandiño found that the documentation made no mention of av_delete calling sv_2mortal on the returned SV. Yet av_pop and av_shift don't and so the documentation should probably point out the difference.

  quirk quirk

Craig Berry reported that maint-5.8 was not compiling on VMS, largely due to incorrect prototypes in re.xs. Nicholas Clark determined that a subsequent integration fixed the problem.

  a matter of time

Steve Peters wanted to know why quad words on Win32 weren't configured, since all the pieces were in place to allow them to be. Jan Dubois thought that it wasn't much of a problem since you really need to have IVSIZE defined to be 8 to take any advantage of them.

  mmm, bignums

Nicholas Clark hacked perlbug to allow it to send thank-you messages back to the porters.

  send more money

Nicholas also got his languages mixed up trying to write else if in C macros. Fortunately there are only four or five distinct syntaxes to master for writing else if constructs in all computer languages.

  as if

About this summary

This summary was written by David Landgren. I chopped a day off this week; it makes it easy to start next week on the first of the month.

  17-23 February 2008

Weekly summaries are published on and posted on a mailing list, (subscription: The archive is at Corrections and comments are welcome.

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