Slash Boxes
NOTE: use Perl; is on undef hiatus. You can read content, but you can't post it. More info will be forthcoming forthcomingly.

All the Perl that's Practical to Extract and Report

This Week on perl5-porters - 4-10 May 2008

posted by grinder on 2008.05.15 16:32   Printer-friendly

This Week on perl5-porters - 4-10 May 2008

Today, I was doing some mucking about with class data, generated packages, and Class::ISA, and I started to wail and gnash my teeth. I cried, "Why can't I just replace the darn object system?" -- Ricardo Signes, a man with a method.

Topics of Interest

fascinating paper: heapsort with n log(n) - 0.9n comparisons

Last week, Marc Lehmann had tossed out a link to an interesting paper he had read on heapsorts, with a new implementation that purported to require fewer comparisons.

This week John P. Lindeman wrote a very thoughtful reply that's worth reading in its own right. In essence, one of the fundamental problems facing heapsort these days is that it is very cache-unfriendly.

Quicksort and mergesort tend to process data sequentially, and that plays nicely with how caches behave. Heapsort, on the other hand, tends to skip around all over the place, and that leaves the cache lines cold and unloved. And so John feels that heapsort will be unlikely to beat the other algorithms consistently, especially when large datasets are processed, which is where the correct use of cache comes into play.

  welcome to the real world 

Helping 5.8.9

Nicholas Clark asked for some volunteers to help with the mundane pumpking tasks that need doing when a new release goes out the door. Such things as checking author/email names, module versioning, dual-life synchronisations and making suggestions for perl589delta.pod.

Yitzchak Scott-Thoennes checked out most of the CPAN/core discrepancies. Steffen Müller did the author names.

  out-sourcing pumpkingship 

add Porting/

Jim Cromie started to play around with git, and the more he looked at it, the more he liked it. He proposed a script to deal with ignoring files not under revision control, but Rafael thought that he had the logic completely backwards.

  git down 

Adding new keywords

Simon Wistow, in a flash of inspiration, realised that it would be fun to dive into the guts and figure out how to add a new keyword to the language, such as method, which would start out being a simple alias for sub. He was quite surprised at how easy it was, and on which side things that can be implemented both inside or outside the core should be.

Jonathan Rockway muttered about Devel::Declare and Method::Signatures. Jesse Vincent revealed that Chia-liang Kao and Yuval Kogman were working on Perl6-Declare that was much more robust and flexible than Method::Signatures.

Matt S. Trout thought that Jesse's recent work on MAD and Yuval's start on B::Simple should be a big help in making the perl5 syntax more malleable.

  anyone for "with" 

TODO Tests

Bram saw what he thought was a flaw in the infrastructure for dealing with TODO tests. Imagine a function that should return 1. Except it doesn't, because a bug that no-one understands makes it return 2.

So we write a TODO test to say "this should return 1". Now imagine that the bug was tickled, and now the function returns 3. Currently there's no way of detecting that the value, while still incorrect, has changed. Some people were dubious about the value of detecting this.

  different kinds of wrong 

Documentation bug or brain bug? (or, what \k<name> and $+{name} refer to)

Sébastien Aperghis-Tramoni had a couple of questions about the new 5.10 regexp extensions (or rather, the documentation that explains their behaviour). After a couple of porters worked the the problems in the wording, Aristotle Pagaltzis wrote up a new description, that Yves Orton thought was a good description. 

Understanding refcounts

Simon Wistow was looking at Devel::Gladiator, a module that examines memory allocations in arenas (geddit?), and wondered why, given two nearly identical snippets, one leaked massively while the other didn't.

Naturally, Dave Mitchell gave a cogent explanation as to why, and demonstrated a simpler case in a couple of lines of Perl, no modules needed. 

Thoughts about overloading method calls

Ricardo Signes want to do some exotic things with classes, objects and overloading. Aristotle Pagaltzis admitted to having wanted to do as much in the past. Ævar Arnfjörð Bjarmason had thought of doing even more bizarre things, and I would say more, but alas this paragraph is too small for it to fit.

The thread is well worth reading in its entirety, for a view on the current thinking on allowing more flexible syntaxes. 

Recursive inheritance detected while looking for method

Lyle ran into a problem with recursive inheritance and wondered why things didn't work in a way that would make it possible.

Nicholas Clark explained that the code that looks after this was heavily rewritten thanks to Brandon Black and his MRO work. He also thought that it should be possible to plug in a custom MRO handler that would do what Lyle wanted. 

File::Path::mkpath() incompatibility in perl-5.10

Gisle Aas demonstrated a problem in the 2.x version of File::Path's mkpath creating unwanted directories. David Landgren offered a work-around, and began to work a better solution.

  the right path 

Should potentially confusing labels emit a warning?

Following up on Andy Lester's blog about the grief caused by BEGIN: { }, which is a labelled lexical scope rather than a BEGIN block, there was a tentative suggestion to make such things warn.

Most people weren't too comfortable with the idea, suggesting that if it was really a problem then it would have wound up on the TODO list at some point.


Special blocks below the top-level scope

In playing around with BEGIN blocks while researching things from the above thread, Aristotle Pagaltzis came across some some strange behaviour with a BEGIN block within a subroutine.

Nicholas explained how it was quite consistent given the rules concerning compile-time and run-time effects, and Jan Dubois wrapped out the remaining doubts. 

On the almost impossibility to write correct XS modules

Marc Lehmann had raised the problem of char * as an interface to character strings in XS last week. Yves Orton pointed out that this is used internally by the core, and thus faces the same issues. Win32 is particularly sensitive to the issue. 

TODO of the week

chromatic proposed a patch for the TODO from last week. Yay!

  the yada yada yada operator 

and a revised patch to add the ..., !!!, and ??? operators

  the yada yada yada trinity 


system() accepts a LIST syntax (and a PROGRAM LIST syntax) to avoid running a shell. readpipe() (the function behind qx//) could be similarly extended.

Patches of Interest

Add Some Links to External (WWW) Resources to the Perldocs

David Landgren thought that the patch to add web page links to the Perl documentation should be refused on the grounds of the likelihood of them becoming obsolete and therefore contributing to the maintenance burden.

  looked at your 1996 bookmarks recently? 

I must be mad

Jesse Vincent moved MAD's libraries under lib/ and placed them under the MAD:: namespace. Other than that he tidied up Gerrard Goossen's work on the test suite and generally brought the bitrot level down close to zero.

Gerrard wondered whether, because of the dependency on XML::Parser, if it wouldn't be better to spin the libraried out completely as CPAN modules. Jesse felt that pulling MAD too far away from the core would all but ensure its demise.

  what, me worry? 

Documentation typo fix for MRO

Florian Ragwitz suggested a documentation patch for MRO but it was ignored.

  the next self can 

Call SvMAGICAL_off in mg_free

In the ongoing saga concerning mg_free and mg_clear, Bram identified only one other place in the core where SV*MAGICAL_off() was called after mg_free. So he cleaned things up, and Rafael Garcia-Suarez applied his patch. 

In the process of working with magic, Bram wanted to know how to downgrade a PVMG to a PV. John Peacock wondered why on earth someone would want to do such a thing, to which Bram replied that it would make testing easier. John suggested a different approach that, while more work for Bram, better reflected what was going on under the hood.

  got magic? 

Add SvMORTAL to check if a SV is mortalized

Claes Jakobsson was grovelling deep in XS code, trying to figure out whether SV were mortal or not, and was surprised to discover that there was no simple check available. So he wrote a macro to allow SVs to be tested as such.

Rafael wondered whether it would be more interesting to wrap it up as a module so that older perls could use it. In turn, Claes wondered if the hassle of setting up the module and having the XS author bring it in and use it was worth it.

Nicholas was hesitant to load up the core with yet another addition to the API, and when Claes thought about writing regression tests for it, he realised that the best place for it would be on CPAN.

  mortal combat 

New and old bugs from RT

Bug in File::Find on Windows when target dir is bare drive letter and no_chdir = 1 (#41555)

Last week, Bram fixed this bug, but wondered how one could write some tests for it. This week, David Landgren sketched out an approach, but no-one commented on its viability. 

Ambiguous Configure run/make error (#49946)

Stefan Schlotterbeck-Macht filed a bug back in January concerning a cross-compilation failure, the exact problem being a make target <command-line> not having a rule in the Makefile. Andy Dougherty noticed it this week, and suggested that it was quite possible that just deleting the target in the Makefile would be sufficient. 

NDBM_File fails to build (#50180)

Also back in January, Devogon filed a report with NDBM_File not being built on 5.10.0 under Debian. Andy had a look at things, and saw that they had changed both the name of the header file (now called gdbm-ndbm.h) and the library file (gdbm-compat).

Andy wanted to patch Configure to do the right thing and so made a request for comments. Currently there are three different header files to search, as well as three different library names. The question was whether there were any other possibilities lurking about on other platforms.

  speak up 

Crash when localizing a symtab entry (#52740)

In April, Niko Tyni wrote a very detailed bug report concerning a segfault in a mod_perl environment. This provided sufficient information to Rafael Garcia-Suarez to make a small change to sv.c, which was later officialised as change #33807.

  one good turn deserves another 

perl-5.10.0-33733 assertion with JSON::XS-2.2 (#53244)

This is the thread about the correct use of SvCUR. Nicholas Clark revived the thread when he explained that one of the reasons the code broke was that because when gcc is instructed to add the -g compiler switch for debugging purposes, Configure piggybacks this and enables a raft of code assertions, and JSON::XS was tripping over one of them.

Many new assertions had been added to the code by Nicholas when he worked on slimming down the SV bodies, and he needed to be sure that nothing broke out in the wild (due to a cavalier approach to the internals). Andreas König's long BBC smoke campaign showed that by and large Nicolas's changes were completely transparent.

Marc Lehmann compared 5.8.8 and 5.10.0, running bare, or with POSIX loaded, and noticed that while the bare binary was smaller, the size when POSIX was loaded showed a larger increase on 5.10 compared to 5.8. This appeared to contradict the statement in perldelta, and this left Marc feeling a bit confused. 

Bug in if(open(my $fh,...)) { } scoping (#53504)

Aristotle Pagaltzis showed how Perl 6 deals with the problem of scoping lexicals to blocks, and that the rules there are much simpler. 

bug with index() matching beyond end of string when \0 bytes are involved (#53746)

Giuseppe Insana filed a fairly comprehensive report detailing his surprise at some results with index when NUL bytes appear in the string.

Bram simplified matters, which made it clear to see what the problem was. Abigail wrote a series of TODO tests so that some budding internals hacker who wanted to take a stab at fixing pp_index in pp.c would know if their efforts were successful or not.

  hint hint 

64-bit Integers -- inexact division gives odd result when is large (#53784)

Chris Hall ran into difficulties with the results of large integer divisions being upgraded to floating point, with the loss of accuracy that that implies. He and Dominic Dunlop kicked the problem around a bit but were not able to come up with an explanation for what they were seeing.

  but -Mbigint ok 

No complain about bareword (#53806)

Ronald Fischer complained thusly when bitten by print merrily accepting a bareword as its first argument and silently stringifying it. Bram promised some TODO tests for this.

Something tells me this is not going to be simple to fix.

  the magic of print 

perlpodspec typo "Encoding::Supported" (#53908)

Kevin Ryde found an instance of Encoding::Supported that he suspected should read Encode::Supported. And indeed it should. Patched by H.Merijn Brand. 

pod2man loses =head2 starting [with] ' or . (#53910)

Kevin also discovered that Pod::Man failed to properly escape POD headings that begin with a quote or period, letting them be interpreted by *roff and thus ignored. He proposed a simple patch that should be enough to resolve the issue.

  more POD love 

Perl5 Bug Summary 

New Core Modules

Encode 2.25

Released by Dan "the Encode maintainer" Kogai. 
IO::Compress::* 2.010

Paul Marquess synched blead with the current CPAN release, which should make things play more nicely on 5.10 
threads::shared 1.19

Jerry D. Hedden pushed the latest threads::shared out the door. The main change is saner semantics for ref() on shared objects. 
Thread::Queue 2.07

The above module in turn allowed Jerry to release a version of Thread::Queue that allows one to enqueue structures with circular references. 

In Brief

Bhavna Yadav wondered how to port a Perl script on Vxworks. Dominic Dunlop pointed out that Wind River ported Perl independently to their Vxworks operating system... seven years ago, and as far as Dominic was aware, no porter had any technical knowledge of the platform. 

Thanks to Reini Urban, one can build a static perl on Cygwin.

  no shared libraries 

Alexandr Ciornii pinged the list with an 18 month old Net::SMTP documentation patch that had never been applied. 

In the Taint (PL_tainting, SvTAINTED_on, SvTAINTED_off, SvTAINT) thread, Paul Fenwick explained that late-acting tainting left PERL5LIB wide open.

  this is a problem 

Vadim Konovalov cross-posted a question regarding Perl cross-compilation on arm-linux but no-one took him up on it. 

Abigail tweaked File::Copy to use the 3-arg open instead of a slightly crufty and vaguely deprecated 2-arg form. 

Abigail also wrote an initial cut to teach File::Copy to respect permission bits. This caused problems on Cygwin, so, for want of a better solution, Jerry D. Hedden changed some of the failing tests into TODO tests for that platform. 

Alexandr Ciornii supplied a Data::Dumper patch to make the latest blead version compile on older perls. 

Ricardo Signes fixed up a POD link for perlunitut . 

David Nicol wrote a wafer thin patch to underscore the semantic importance of the empty pattern (//). 

Last week's summary

  This Week on perl5-porters - 28 April-3 May 2008 

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 or attending a YAPC to help support the development of Perl.

The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
More | Login | Reply
Loading... please wait.