This Week on perl5-porters - 17-23 July 2006

This Week on perl5-porters - 17-23 July 2006

"I'd like perlcc to die because it's an attractive nuisance. People keep finding this, finding it doesn't work, and posting "how can I make this work" posts" -- Joshua ben Jore, out for blood.

Topics of Interest

Is the -std=c89 switch worth the pain?

It started when H.Merijn Brand noticed that the recent addition of the gcc switch (-std=c89, to ensure conformance with the C89 ISO/ANSI standard), caused the Cygwin compiles to develop an interesting smokey flavour.

  That's not supposed to happen

After he fixed that up, he then noticed that an HP/UX box was also appeared to be rather distressed by the same switch. Despite the fact that the gcc compiler in question was a much more recent edition.

  Uh oh

Jarkko Hietaniemi felt H.Merijn's pain, and recast the patch to enable the switch only where it was proven beyond reasonable doubt not to cause problems. Much the same way that -ansi -pedantic cannot be used everywhere.

Nicholas Clark thought that it was safe now and should ever more be so, on FreeBSD. Rafael described some of the creepy-crawly things he had encountered in his sojourns through glibc headers, and concluded that Linux should be scratched off the list altogether. Jarkko fumed and went to stand in the corner.

Sadly, at the end of the day, no more remained, save for a footnote buried in the depths of perlhack.

  Good if you can get it

Thread Exit

Jerry D. Hedden summarised the ongoing thread about threads, and how they should be exited, and how much they should exit. Of note was a post from Eric Rybski, who has been using the ithreads model intensively over the past couple of years, and injected an interesting point of view into the debate. In a nutshell: should not impose global semantics on how threads operate.

  Last exit to Threadlyn

Before going too far down the wrong road, Jerry reported on what he had come up with, and asked for feedback. Rafael Garcia-Suarez appreciated the backwards-compatible aspects.

Jan Dubois was a bit uneasy about the behaviour of a thread calling die.

  This way out

And so it came to pass that Jerry released threads version 1.37. As a bonus, the reworking of Perl_ihtread_hook appears to have fixed bug #39839, although Dave Mitchell expressed doubt as to whether the race condition that the bug cites has just been pushed out somewhere else.


And some additional thready goodness also for 1.37.

In preparing this patch, Jerry encountered what he thought was a bug in under MSWin32. It turned out to be a problem with newlines in fragments of -e code.

perlbug on Windows

Steve Hay was puzzled why perlbug was not (apparently) working on Windows, and traced it down to a Unixism that leads to a two minute timeout on Windows. He considered a fix that would work correctly in his corporate environment (using the USERDNSDOMAIN environment variable), but wondered if this was applicable to the world at large.

  Code set free

B:: renovation, and the end of the perlcc compiler

Joshua ben Jore produced a patch to overhaul B::, to drag it kicking and screaming into the 21st century (the module, not Joshua, although...)

One of his biggest gripes was with B::C and B::CC, that he would rather see them jettisoned out the airlock, along with perlcc compiler, saying that all perlcc really does is fixes bugs (in the same way that flypaper fixes flies). Rafael agreed that this was a very tempting suggestion.

Nicholas wondered if the cascade of events would lead to the removal of ext/Byteloader as well. Joshua put forward the best argument for ditching all of it, and that was if anyone had a burning desire to get anything along these lines working, their best chance lay in hooking up with the Parrot and Pugs teams, because that was the only way they would stand a chance of attracting the attention of anyone who might be interested in seeing the task achieved. In the meantime, the perl tarball would shrink, which is a bit of a bonus.

  At the end of the line

Using SV as key in hash

Elizabeth Mattijsen posted this question on Perlmonks but received no useful answers, so she came and asked the porters. She wanted to set a hash key to the contents of an SV. In theory it would appear that the API supports this, however, it doesn't work in practice. Unfortunately, the porters didn't appear to have any special insight into the matter, either.


Patches of Interest

Problem with autogenerated ppport.h files (Change 28513)

Steve Hay encountered a minor problem due to the changes induced by generating ppport.h files on the fly. Marcus Holland-Moritz welcomed the information, and whipped up a fix to solve the problems. Steve gave it the green light, and Marcus uploaded a new version of Devel::PPPort.

  globby badness

z/OS: pp_sys.c, reg*.c, toke.c and utf8.c

Jarkko continued to untangle the useful bits from the z/OS patch that the IBM developers posted a while back. He punted on regexec.c, citing too many differences in the codebase between then and now, as well as philosophical differences.

New and old bugs from RT

localtime(3) calls tzset(3) but localtime_r(3) may not (#26136)

Benjamin Holzman delivered another version of the consolidated patch to fix this up, but apparently it fell on the floor.


Switch module bug (#39789)

Steve Peters wondered whether it might be able to transmute the Switch module into the switch feature now in blead. Nicholas Clark thought not, citing problems backwards incompatibility, and wondered if anybody was particularly interested.

  Not bug-compatible, for a start

eval and hash access in subroutine (#39816)

Thomas Ziehmer had a problem with a lexical going out of scope in a package. David Landgren offered a work-around, using an our or package variable. Dave Mitchell explained why the behaviour was the way it was.

Rick Delaney took Dave's example and tweaked it to produce a different message that seemed to be more in line with the behaviour that Thomas had encountered, and after a follow-up from Dave, we all understood what was happening.

  Seeking closure

PERL5SHELL is not checked for tainted data (#39832)

There was a lingering suspicion as to whether $ENV{SystemRoot} and $ENV{windir} required taint checks on Win32, but Jan Dubois thought that their use was benign as far as Perl was concerned.

  Caring for the environment

UTF-8-related crash (#39870)

"vvv" supplied a small program demonstrating a core dump with UTF-8. Andreas Koenig explained that the best way to solve this problem was to use use utf8 instead of use encoding 'utf8', whilst cheerfully admitting that it was indeed a bug in use encoding, that continues to reside even in a recent version of blead (somewhere around patch 28603).

  Tis a bug

-0.0 loses signedness upon numeric comparison (#39875)

Zefram had a problem with -0.0 becoming 0.0 after a numeric comparison. Yitzchak Scott-Thoennes proposed an interim fix, but asked for a Configure probe to check for the existence of a C routine (signbit()) needed to deal with the problem.

H.Merijn thought that this would mean that the bug would persist on platforms that lacked the function, and Yitzchak agreed, and asked H.Merijn whether an alternative function (copysign()) would be better. H.Merijn ran some code and produced some statistics of its availability.

Time::gmtime 1.02 examples patch (#39876)

Alexandr Ciornii found some glaring errors in the the documentation to Time::gmtime. Steve Peters applied the fixes, and bumped the version up to 1.03.

  About time

printf and stat out-of-order (#39880)

Someone thought they had encountered a problem with stat, printf and temporary variables interacting in weird ways, but 'twas merely another case of crypto-context.

  Foiled again

Inconsistent list slice behaviour (#39882)

Bo Lindbergh ran into problems with slices of empty lists, and found the different behaviours of sort, grep, do and company rather inconsistent.

  There was probably a reason in the beginning

ISA cache leaks into and out of Safe compartments (#39887)

Marc Lehmann discovered, through a quirk in the method caching code, that it is possible to have methods names leak into and out of Safe compartments. He proposed a quick work-around to avoid his specific problem, and a tactical nuclear strike to solve the problem in the general case, but wondered whether the latter was too heavy-handed. At the other end of the spectrum there is the possibility to simply document the behaviour and move on.

Nobody had sufficient motivation to comment one way or the other.

  Safety first

perl 5.8.8: RT 3.6.0: segfault in perl sv.c:5810 (#39893)

Peter Karl Mueller encountered some MySQL/UTF-8 grief in RT, and asked for help.

Docs for close don't mention effect of stream error status (#39902)

Timbo took the time to explain that close does The Right Thing, no really, but the documentation doesn't clarify that it does, nor what it is.

  The path to success

perl C compiler Can't locate object method IVX via package B::NV (#39903)

Tim Liim reported an error in some code that was compiled with the perlcc compiler (see also: Topics of Interest this week).

  I'm sorry Tim, I'm afraid I can't do that

Regular expression whitespace character has different meanings in UTF-8 and single-byte modes (#39912)

"shadrinovdd" had a problem with character 160, which, when interpreted according to ISO Latin-1 rules, is a non-breaking space. And this causes different user-visible behaviours to emerge, depending on whether the UTF-8 flag is switched on for the string.

Much discussion ensued as to what the right course of action should be.

  Lost in space

Perl5 Bug Summary

  Back below 1500 again

New Core Modules

In Brief

Jarkko Hietaniemi was all in favour of making Math::Complex more bignum-savvy, but admitted to not knowing how to make it so.

Steve Hay produced a series of test cases to isolate a mod_perl 1.x problem using bleadperl. Rafael thought he had the right fix, but that didn't work. Dave Mitchell was too busy in the real world to look. The problem remains.

Gabor Szabo suggested adding some links to standard modules in perlfunc.

Jarkko made magic vtables const if PERL_GLOBAL_STRUCT_PRIVATE is defined, perhaps in an attempt to revive the Symbian port.

Yves Orton had a stab at silencing more warnings when compiling the regexp engine, but gnashed his teeth at recalcitrant printf format strings that refused to play the game.

Bill Kurland wondered whether the mailing list was the best place to ask question regarding regular expression test failures.

  Probably, yes.

Ashish Agarwal had some trouble with the Convert-IBM390-0.22 module, which was in fact an interesting case study of Unix encountering EBCDIC.

Tim Bunce needed help spotting a leak in the DBI internals. Nicholas Clark spotted something that looked fishy. Tim was thrilled, then saddened, since afterwards, while the code leaked less, it continued to leak a bit.

  The story of my bathroom tap repairs

Andreas Koenig later reported a problem with DBI breaking on blead. Not having a copy handy, Tim asked Andreas to try a one-liner fix. After this failed to work, Tim promised to pull down the latest blead and see what needed to be done.

  After OSCON

David Nicol put forward a magical vtable proposal. Nicholas Clark said that it sounded like a serious undertaking with the possibility that the code could be scrapped after much toiling. So he didn't feel like doing it.

Hugo van der Sanden spotted some code in Net::Cmd that was performing a last when it should really have been performing a return, and wondered if anyone agreed with his analysis.


Jarkko taught DEC OSF to use -g3 if both -O and -g are specified.

About this summary

This summary was written by David Landgren.

The p5p summaries will now have a four week recess. I'm off to Australia to go traipsing around rain forests, the Barrier Reef and Sydney Harbour. With a bit of luck, I won't have to turn a computer on at all.

Writing these summaries has been fun, albeit exhausting at times, and I'm looking forward to having a rest for a while. I hope you have enjoyed reading them. I'll be back in time for YAPC::EU in Birmingham, where I shall try to summarise the the highlights between now and then, jet-lagged out of my mind in the back of the auditorium.

If you want a bookmarklet approach to viewing bugs and change reports, there are a couple of bookmarklets that you might find useful on my page of Perl stuff:

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.