This Week on perl5-porters - 6-12 November 2006

This Week on perl5-porters - 6-12 November 2006

"This is like sitting your mother down to teach her computers and beginning by explaining little endian vs big endian and the intimate details of protected mode memory" -- Michael G. Schwern, firmly believing that there must be a better way of installing perl.

Topics of Interest

INC handlers and __FILE__

(I didn't get around to summarising this thread last week).

Ron Isaacson had reported that he used code references in @INC, to great effect, but regretted that the modules loaded no longer really knew where they came from, __FILE__ and related things would refer to the module loader module, rather than their own paths.

Rafaél Garcia-Suarez concocted a quick fix in pp_ctl.c for blead, which improves the situation considerably. Glenn Linderman wondered if a cheap and nasty hack, a #line directive at the top of each file, could do as a temporary work-around.

Nicholas Clark suggested Ron take a look at an Acme module of his, that might do more or less what he needed (see, sometimes Acme modules are useful).

Failures in /run/cloexec.t on Solaris and Tru64

The new t/run/cloexec.t test blew a bit of smoke out the O-rings. It could mean that this is testing the operating system, rather than perl, in which case it might wind up being removed. Andrew Savige, father of the original patch, tried to teach it some manners.

Building perl with VC++ 2005 (VC8)

Steve Hay, Steve Peters and Jan Dubois continued to work this week on getting Visual C++ from Microsoft to compile the perl source, and had lots of fun, or not, playing around with I/O, DLLs and the C runtime library. By the end of the week all appeared to have been straightened out.

Later on in the week, Steve Hay started having problems with test failures on a VC8 build. These were apparently due to problems in interacting with (or rather, finding) the C runtime library.

After a bit of a discussion Steve was moving in the right direction.

Installing Perl shouldn't feel like a rite of passage

Michael G. Schwern felt that the Configure process was a bit too verbose, asked to many questions, and sometimes gave some very scary replies. He felt that it would be a good idea to be able to reduce the verbosity level. Some way down in the thread, he showed the README from Ruby, which, it must be said, is short, sharp, and to the point.

Andy Dougherty noted that the INSTALL file had amassed considerable cruft over the years, due to various stupid bug reports that had wasted considerable time of those doing front-line support at the time.

Nonetheless, as Michael pointed out, the perl install process is geared to catering for people with weird system setups, are in the business of packaging Perl for distributions or are using it on non-mainstream systems. The README and the Configure step reflect this: it's hard to know what's important, and what's noise.

H.Merijn Brand felt that it might be worthwhile to put more emphasis on, the idea being that if you've sweated out a configuration trial-by-fire once, you can at least save that file away in a safe place, and reuse it next time you build a perl.

At the end of a fairly long, but always interesting thread (lots of good stuff from chromatic), Jarkko Hietaniemi injected a dose of reality into the discussion, pointing out that Configure is not going to be redesigned any time soon, it's hard to produce an interface to it that makes end-user programmers and system administrators happy, and so yes, to a certain extent it *will* be difficult to avoid building perl from being a rite of passage.

  655000 lines of C can't be wrong

Andy Dougherty produced an initial tweak to Configure to tone down the *** WHOA THERE!!! *** messages to something a bit more laid back.

  goodness gracious

Jesse Vincent updated README to make it a little friendlier, and feared that the resulting discussion would evolve into a discussion of the colour of the bikeshed. Rafaël JFDI'ed the patch.

There were a number of useful suggestions made, and Michael reported back that the target guinea pig for installing Perl liked the new version.

Failures while attempting to install Plagger with bleadperl

Steve Peters was delighted to discover that Plagger has even more CPAN dependencies than Catalyst or Jifty. Therefore, installing that one module causes a large slab of CPAN to be pulled down and tested, which is just the ticket when you're taking blead for a spin.

Alas, twenty eight modules failed their test suite, and three more failed to build altogether.

  Uh, could you run prove -vb on that, please?

Taking a look at the OS X hintfile for 10.4

Michael G. Schwern also thought that the hintfile for OS/X was rather dusty, especially in terms of the 10.4 release, and thought that a touch of spring cleaning was in order.

Andy Dougherty, Dominic Dunlop, Bo Lindbergh and Edward Moy had a look at the issues, and it appears that the main ones are locales and poll(). Considerable head scratching and archive diving was observed, and Andy hoped that the patch file would be commented with pointers to this and other discussions, so that when the problem is revisited, say, in 2011, it will be easy to zoom in on the previous debate.

  those who do not learn from history

Warnings from Pod/t/man.t in blead

It seems that Pod::Man is also falling foul of the new Variable length character upgraded in print diagnostic. Rafaël fixed the problem in blead, but Russ Allbery noted that it wouldn't work for maint, as it uses modern bleading features. He suggested a possible maint-compatible technique, and wanted to know if it would fly.

As proposed, the technique caused the tests on blead to fail, but Rafaël produced a variant that should work on both codebases. It did lead him to doubt whether the warning was actually useful in general.

  Modern man

Chatting about sort sanity checking

David Nicol began to see how easy it would be to add warnings to catch custom sort comparison functions that don't fulfill the contract of what such a function is supposed to do (return a negative or positive value, or zero). The low-hanging fruit would be to check for <, > operators and the like.

John P. Linderman thought this would be a good idea, but found it difficult to reconcile with the constraint that there should be no runtime penalty.

  Out of sorts

MRO, sub_generation and so on

Brandon Black wrote Class::C3, which provides an alternate technique for method lookups, and he had come to the conclusion that for best results, it needed to be more tightly integrated into the core (or rather, the core needed to expose some aspects to the outside, that his module could tap into).

Specifically, he had problems with the fact that a single change somewhere can blow away the entire method name cache. What would be nice would be to a way to blow away the methods on a per-package basis, as it would less the cost of method cache invalidations.

Hugo van der Sanden was extremely interested in the approach, since he has lots of code that creates methods on the fly, and this winds up continually trashing the cache. Joshua ben Jore thought that Moose and Class::Trait would benefit from the concept.

Nicholas Clark was somewhat sceptical, wondering how often people run around adding new methods to code (once it's up and running, after some preliminary initialisation phase). Be that as it may, he wasn't particularly familiar with that part of the codebase, and suggested looking through the lens of perlbrowse, to find out who has been working on that code in the recent past.

He did just that, and discovered that most of the code dates back to beginning of the revision history, and thus the details are literally lost in the mists of time. After taking another, closer look, Nicholas was able to give Brandon some reasonable directions on how to proceed.

  A new order

Perl Debugger resets the $main::{"_<$filename"} upon eval with a #line $filename directive

Shlomi Fish found a problem with the debugger, that only showed up in non-threaded 5.8.8 builds. After poring over the changes in perlbrowse, he was able to pin-point the likely culprits. Once this legwork was done, it was a small matter for Rafaël to come up with a fix.

Shlomi was most impressed, and asked if Rafaël had come up with a test, to ensure that the bug didn't come back again. Rafaël admitted to not being sure how to go about that, and Shlomi said that he had an idea for one. So he produced a stand-alone test that did the deed, and again it was a small matter for Rafaël to recast it in terms of a test for the test suite.

  debugging the bug in the debugger

24714 broke Term::ReadLine::Gnu

Andreas König discovered that a recent change to optimise lazy copying in sv_setsv() caused Term::ReadLine::Gnu to go belly up. Nicholas Clark rejigged the patch to restore the old behaviour to callers outside the core, but Andreas still wasn't getting satisfaction.

Patches of Interest

New regexp syntax omnibus

Yves has been speaking with Larry Wall about the work on the regular expression engine. Since all the new (?..) verbs do something quite different to the older zero-width assertions, it makes sense to make them look different. So now they are introduced by (*...). (Makes me glad I haven't been writing code that uses them).

Anyway, the plot is beginning to be revealed at last. Yves wants to build Perl 6 rules on Perl 5. This is going to be interesting.

Additional changes followed, and all of it wound up being landed in blead.

Joshua ben Jore was having trouble keeping Yves Orton's new regular expression control verbs straight, since he kept getting mixed up with Prolog. He also had trouble keeping up with the current thread dealing with these matters, since he replied to a previous thread on the subject. He thought that ERROR should become CUT, and CUT become something else. Either that, or use whatever Perl 6 uses.

  the hobgoblin of little minds

Static linkage for perl.exe for win32

Vadim Konovalov added a new target and few small tweaks to the Win32 Makefile, to allow a static perl.exe to be built (no more perl.dlls). Steve Hay suggested better names for the resulting executable.

  No more DLL hell

Vadim would really like to know how to push this patch forward, since it would make his life a lot easier.

Make s?printf reject tainted formats

A few months ago, the idea was put forward that printfy format strings should be subject to taint checks. Rafaël wrote the code to implement this, and then stood back to see what people thought of the concept. Dave Mitchell warmed to the concept and Johnathon Stowe suggested that strftime ought to receive similar treatment.

Rafaél couldn't see how to abuse strftime effectively, so that was left alone, but the code for the printf family went in.

  taint fair

Perl_die() / Perl_croak()

Philip M. Gollucci finally scratched his itch and fixed up the printf compiler attribute checks, by disabling them. This allows modperl 2 to be built under blead again.

Robin Barker was under the impression that it is distressingly easy to break modperl when working with blead, and wondered if there was some way of capturing in core tests the assumptions that modperl relies upon.

Make if (%hash) {} act the same as if (keys %hash) {}

Yves Orton noticed that, while semantically equivalent, testing for %hash or keys %hash to see if a hash contains something do not run at the same speed. He patched the core to make the former become about as fast as the latter.

Rick Delaney reminded Yves to watch out for tied hashes, in case his patch changed existing behaviour. Worse, using if (%hash) now resets the iterator, which was more or less the final nail in the coffin for the idea.

Yves then took a different approach, and created a new opcode boolkeys to be used in this if (%hash) context, made it do the right thing in the face of tied hashes, and as a result made it even faster than using if (keys %hash).

  faster %pussycat

New and old bugs from RT

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

Back in July, Benjamin Holzman supplied a patch to fix up the mess when playing around with tzset and the TZ environment variable. H.Merijn Brand dusted it off and applied it to blead as change #29209.

Yves Orton wondered whether it was good for Windows, since it caused a number of tests to fail. Furthermore, it was unlikely to work in any event, since recent overhauls to the environment variable handling on Windows has decoupled the changing of environment variables within Perl space from affecting the underlying C runtime library's notion of the environment.

Sadahiro Tomoyuki suggested that the safest course of action was to leave the code, and skip the tests on Windows, and supplied a patch to do just that.

  time on my side

/^/m doesn't match after a newline at the end of the string (#27053)

Yves then responded to a February 2004 bug filed by zefram, who didn't like the behaviour of ^ in a certain context. He thought that while it could be construed as a bug, it has since managed to become sufficiently widespread to have evolved into a feature. While fixing it would be theoretically good, it would cause a lot of existing code to fail, starting with important pieces of the Test:: namespace itself.

This made Steve Peters ponder the idea of adding a "Won't fix" bug status, which would allow bugs like this to be no longer counted towards the open bugs total. Steve thought that a number of auto-vivification feature-bugs could go this way as well.

  sort of like unemployment statistics

Cwd::chdir() and handles (#38466)

Steve Peters pinged Ken Williams on a patch for PathTools.

Data::Dumper fails to dump all hash keys when iterator is not at start (#40668)

Jerry D. Hedden fixed this bug, bumped the version, and Rafaël applied it.

Bleadperl crash in Perl_pp_entersub() (#40681)

Steve Hay and Nicholas Clark got to the heart of the matter caused by modperl 1.x blowing up on blead. And when Nicholas finally came face to face with the bug, he didn't know what the best way to solve it would be.

Different behaviour of precompiled regular expression (#40684)

Claus Fischer reported a bug that Yves noted was already fixed in blead.

Attribute::Handlers not fully functional at runtime (#40695)

Johnpc discovered that attribute handlers don't work quite perfectly for routines that are pulled in dynamically, via require. His work-around was to place the require in a BEGIN block. Rafaël suggested trying a different approach, and asked if that would work instead.

Weird behaviour in tainted hash with large integer, SvIV/SvPV mismatch (#40708)

payerle wasn't sure, but he thought he had encountered a bug in Perl, and so he filed a slightly apologetic bug report. Dave Mitchell cheerfully confirmed that it was in fact really and truly a bug in Perl. He simplified the code, and showed that 5.8.7 and blead gets it wrong, but, somewhat miraculously, 5.8.8 gets right.

At this point, Dave punted, and asked Nicholas to take a look.

perl parser bug leading to memory corruption/segmentation fault or simply parse errors (#40718)

Dave Mitchell fixed the parser so that it no longer lost its mind.

Perl5 Bug Summary

  22 up, 18 down, now 1539

New Core Modules

In Brief

Nicholas Clark uses an Intel-based Macintosh.

Mike Schilli fixed a bug in Applied.

Robin Barker worked on suppressing the df warning in stat.t on OS X but Rafaël beat him to it.

But he was first to fix t/op/inccode.t failing under minitest.

and silenced a signed/unsigned mismatch warning,

not once, but twice,

Rafaël made Text::Wrap warn less aggressively.

he also added a whole lotta goodness to Attribute::Handlers, based on last week's discussion.

Jarkko Hietaniemi made a couple of small tweaks based on smoke signs, that is, errors produced in daily stream of smoke tests.

Michael G. Schwern uncovered a blead configuration that prevented strict from being installed.

  Not a feature

About this summary

This summary was written by David Landgren.

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.