This Week on perl5-porters - 30 December 2007-5 January 2008

This Week on perl5-porters - 30 December 2007-5 January 2008

"That's now twice in this thread that I've been accused of favouring languages that I actually dislike [...] . I think the Java model of exception signatures is awful." -- Zefram (I hope you know this will go down on your permanent record).

Topics of Interest

Exception roles, take 1

Ricardo Signes wrapped up the year with a first cut at doing exception roles in Perl (the idea being that one would manage errors with some formal mechanism, anything, than matching $! with regexps).

Zefram quoted lisp back at him for more background on handling error conditions.

The discussion continued in the new year, with people discussing how to avoid creating elaborate Exception hierarchies that would wind up looking like Java.


Lexicals used only once should warn (redux I)

This thread from last week, err month, um year continued in full swing this week. There are two main points to come out of it. Firstly, given the following statement:

  my $opaque = xyzzy();

if $opaque appears nowhere else in the current scope, it is not possible to determine at compile time whether or not its purpose is to hold a reference to an acquired resource (and thus cannot be considered "unused").

Secondly, and in light of the above, is it worth expending so much effort to hunt down and carp about truly unused lexicals? From a purist's point of view, the answer is yes, but from a pragmatist's point of view, the better solution lies in a lint-like analysis.

Lexicals used only once should warn (redux II)

Another part of the thread from last week reminded Fergal Daly how much he would like, in a loop like

  for my $x (@list) { ... }

to obtain a trace that not only shows the call stack, but also shows the value of $x. Even knowing that the loop was in the Nth iteration would be better than nothing. chromatic suspected that if code were written to handle that, it would come in handy for dealing with tail-call optimisations.

That reminded Yves Orton about a $^SUB variable which would provide a hinting mechanism to the compiler, thereby side-stepping the issue of introducing unnecessary slowdowns in the general case, and putting the onus on the programmer to get things right.

  RAII: not an Italian TV station

sub 2007{ ... goto &2008 }

David Nicol broke out of the monster warnings thread and wondered about tail recursion the light of RAII. The trouble is that an opaque scalar in @_ holding a resource lock would get wiped out during the tail recursion call. He was of the opinion that tail recursion could not be done automatically by the compiler, but that it could be possible to set it up yourself, if one was able to set up a new @_ explicitly, just before transfer.

  finding the right syntax

In the new year, Aristotle Pagaltzis cautioned against diddling with @_ since it is built for speed and thus behaves oddly in a number of edge cases. Better to invent some sort of syntax that looks like a regular function call, and deals with @_ itself away from prying eyes.

Jim Cromie pondered whether some sort of pp_goto/pp_entersub mashup would do the trick.

Optimising opcodes

Rafaël Garcia-Suarez, Nicholas Clark and Paul Johnson continued to ponder ways to obtain line number information for warning messages if all the nullops were thrown away.

  pointer games

Negated smart match

Ron Blaschke wanted to know why there was no !~~ negated smart match operator. There's one in Perl 6. In Perl 5.10, it gets parsed as the slightly useless !~ ~. In the meantime, one can get by with !($thud ~~ @qux).

  one way to do it

strictly by default

The debate continued over how and when strict would be enabled by default in future versions of Perl. The key was to specify some sort of use 5.12 pragma or a feature, but there was disagreement over whether they should behave the same, or differently.

SV leak

Nicholas Clark looked at some code in XS_PerlIO_get_layers and realised that it was probably leaking SVs. This made him wonder if other XS code committed the same sin.

Marcus Holland-Moritz seemed to think so, and committed change #32816 it. And then committed #32817 when he realised that the routine's dTARGET was now unused.

Vincent Pit had a look around and came up with a number of other places where the same kind of misbehaviour was occurring. He whipped up a patch, that Marcus applied.

Handling of SET magic in mX?PUSH macros

After the above adventure, Marcus noted that there was no need to handle the setting of magic in the mPUSH family of macros, since they are creating new mortals that cannot have any magic on them (yet). Which makes for less make-work code.

MRO and av_clear

Torsten Schönfeld was having problems with perl 5.10 and XS code diddling @ISA to change package hierarchies. As it happens, Torsten was using the av_clear API call to clear out @ISA.

The problem is that @ISA has a certain amount of magic associated with it. Rick Delaney had a look at what Torsten was trying to do, and once he understood what was needed, was able to cook up a patch and toss in a couple of regression tests to wrap things up.

  introducing magic_clearisa()

When will perl 5.10 be stable?

Alberto Simões asked why stable.tar.gz refers to 5.8.8 and not 5.10.0. A long discussion ensued. Most people were happy to accept that it is probably premature to label 5.10.0 to be stable, but it's going to happen sooner or later. So when?

Michael G. Schwern suggested adopting Debian's stable/testing/unstable/experimental labels. The main problem was that people had difficulty trying to match Perl releases into the above four categories. Dave Mitchell came up with an alternative eminently pragmatic approach.

  5.8 is the new 5.6

SvOOK() now doesn't (ab)use SvIVX

Nicholas Clark, looking more closely at how macros expand, put forward an alternative technique to deal with strings that get clipped from the beginning. Instead of recopying the string, perl has always kept the string as is, and moved a pointer forwards to point to the new beginning.

Until now, the macro expanded to some bit-twiddling and possibly a function call. By rearranging things, Nicholas was able to get rid of the function call, but wondered if there was a way to trip things up because of that.

After having played with it a bit more, Nicholas determined that it was simpler to store an offset.


Solving the ~~ changing behaviour after using ==

Following on from the thread from last week where Gabor Szabo reported that smart match could return a differing results from the same inputs, Nicholas changed the behaviour in bleadperl so that "42x" (a numeric value with trailing garbage) never gets the IOK or NOK flags set. Thus solving the problem neatly.

It turns out that doing so didn't break the test suite, but the question to ask is whether there were no tests for it. In which case, careful analysis will be required to see whether it is safe to backport to the 5.10 line.

Patches of Interest

mg_magical() sometimes turns SvRMAGICAL on when it shouldn't

Vincent Pit detected problems in the chain of magic whereby different ordering of magic would produce different results. Steve Peters wanted to see some tests.

refactor PL_opargs generation in and fix helem

Marcus Holland-Moritz was in the mood for adding a new op flag, and suffered considerable pain when he gazed upon, as well as some 8-year old code contributed by Ilya Zakharevich which he thought was a "can't happen" scenario.

In tightening things up, he discovered a dormant bug that meant that helem had an incorrect specification so he corrected it.

  all applied

Loading a "loadable object" with a non-standard file extension

One of things that was pushed for in 5.10 was to embed platform-specific decisions into when it was generated during the build of Perl instead of deferring things until run-time. The move proved to be a shade too aggressive and broke established behaviour in 5.6 and 5.10.

Jan Dubois restored the old behaviour with a patch that was applied by Rafaël.

Clean up File::Temp test file

Jerry D. Hedden fixed up a leaking temporary file in lib/File/Temp/t/fork.t. Applied by Rafaël.

Clean up lib/B

He also ensured that the realclean target removed lib/B. Not applied.

~~ is not a feature

Jerry also redelivered a Warnocked patch which, happily, was applied the second time around.

  if at first you do not succeed

File::Temp::_gettemp should ignore dir -w test on Cygwin

Jari Aalto could not install CPAN modules on Cygwin because of a pointless check to see whether the directory was writable (which it always is). Applied.

This is the BBC

Params::Validate and Clone

Andreas König, Rafaël Garcia-Suarez, Nicholas Clark and Steve Peters had a closer look at this failure and tried to figure what could be done to blead to reduce the breakage. To a certain extent, however, some changes have been advertised for a long time, patches have been sent to authors of problematic modules, but few distributions have seen new releases.

  not much we can do

New and old bugs from RT

say behaves as just print on tied filehandle (#49264)

Ambrus Zsbán noticed that say on a tied filehandle lacks the \n tacked on the end, and traced the problem as far as pp_hot.c but didn't know how to fix it.

Schwern weighed in with a first cut at a patch. Graham Barr saw that it leaked. Rafaël and Nicholas started debating internals, discussing hitherto unknown macros (at least to the summariser). Something was applied, in any event

  say can you see

IO::Handle method say should ignore $\ (#49266)

Ambrus, on a roll, found another edge case where say misbehaved. This was either ignored, or solved by the same patch that fixed bug #49264.

  say it ain't so

B::Deparse fails to deparse a reference to an anonymous hash (#49298)

David Leadbeater noticed that B::Deparse was incapable of dealing with coderef that returns a reference to an anonymous array or hash. Rafaël muttered something about someone having to teach something about something, and then did just exactly that.

  special ops

[[:print:]] versus \p{Print} (#49302)

According to the documentation, any [[:...:]] and \p{Is....} pair should match the same thing. Robin Barker showed that this was not always the case.

segfault in 5.10 (and earlier) (#49322)

A bug report from Will Coleda showed that

  @r=eval {@c=(@n=(1,2) && ($n[1],$n[0]))};
  @r=eval {@c=(@n=(1,2) && ($n[1],$n[0]))};

will crash on any number of different versions of Perl.

  so that's pretty sick code

Steve Peters noted that in 5.10... it still dumps core, but with a new error message!

  progress at last

Segfault with with tie and STDOUT (#49366)

Steve Peters noted that if you are not careful when creating tied objects that print, and the thing tied is STDOUT, perl goes into a loop of infinite recursion and dumps core (after exhausting its C stack). Ways to have interpreter deal with the situation more gracefully foundered on the problem of determining the maximum stack size in C.

  and portably, while you're at it

Perl5 Bug Summary

Ticket Counts: 310 new + 1470 open = 1780 (8 created, 4 closed this week)

New Core Modules

Math-Complex 1.38
  documentation and test tweaks, courtesy Jarkko Hietaniemi (applied)
Sys-Syslog 0.24
  tests that skip, courtesy Sébastien Aperghis-Tramoni (applied)
constant 1.15
  tests that behave on 5.8.[0-3], also Sébastien, also applied
ExtUtils-MakeMaker 6.43_01
  lots of bug fixes, courtesy Michael G. Schwern

In Brief

Nicholas Clark thought that having gcc -pedantic on by default would be useful more for debugging builds than production builds.

Jan Dubois traced a perl 5.10 failure on Linux 2.4 to a bug report and its corresponding patch. Even so, he couldn't see why it caused the failure, but an environment variable tweak to the system provided a reasonable work-around.

  more fun with glibc

Nicholas responded to Larry's remarks about the fact that the Perl 5 smart match was not quite the same as the Perl 6 smart match. The problem is that there is insufficient cross-pollination between the two development camps. Certainly, there have been no patches from Perl 6 developers to adjust the Perl 5 implementations of Perl 6 ideas to keep them in line with the functionality du jour.

  obscured by crowds

Vincent Pit thought that DEBUG_S should meet the thin end of a chainsaw.

  remnants of 5.005 threads

After having manually expanded macros once too often, Nicholas finally got fed up enough to write a short Perl program to automate the task.

  good laziness

Rafaël killed the v-string portability warning in 5.10, declaring that it would no longer be present in 5.10.1.

  no-one shed a tear

Robin Barker's consting goodness to Compress::Raw::Zlib and Filter::Util::Call were applied.

Sébastien Aperghis-Tramoni delivered some small documentation tweaks, applied.

Marcus Holland-Moritz eliminated some magic numbers in NewOp() calls. Code that invents new ops is likely to break.

  but code that invents new ops is unlikely

Michael G. Schwern tweaked t/ to make it resistant to changes to $\, $" and $,.

  no more havoc

The "strict by default for 5.12" discussion got bogged down in details of whether it should be a feature and how should it really be enabled but sometimes we don't unless we do although maybe we might if we should.

  or words to that effect

About this summary

  get last week's here

This summary was written by David Landgren. It does not exactly cover the entire week, as I want to move from Monday through Sunday to Sunday through Saturday. So some threads will be dealt next week.

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.