"I'm still a big fan of partial solutions. But to get a good partial solution, you can't just rush into it without thinking about what the tradeoffs really are." -- Mark-Jason Dominus, on reversible debugging
As the changes Yves Orton made to the regular expression engine get worked over by the smoke testers and other more adventurous souls, a couple of problems have come to light.
Nicholas Clark found that the following snippet caused a segfault during global destruction.
$a = qr/(xx|yy)/;
sub a {'xx' =~ $a and print 'ok'};
threads->new(\&a)->join();
Yves fixed that up with a one-line change (a reference counting adjustment).
Ha ha ha plonk http://xrl.us/nnup
Dave Mitchell caught another one, simply by running a build and
watching ext/re/t/regop.pl lose it, which Rafael Garcia-Suarez
corrected with a guard to re_debug_flags, but wondered why it
was being set to NULL in the first place. Answer: it happens
during global destruction.
Changing of the guards http://xrl.us/nnuq
Jarkko Hietaniemi found some other problems with the trie code on
Tru64 in conjunction with -DDEBUGGING. After a quick examination,
Yves couldn't find a good explanation why, but suggested a couple
of nice places for setting breakpoints.
After running down a few dead ends, Jarkko finally narrowed the problem down to taking the address of something on the stack. Hoisting the declaration outwards fixed the problem.
A class of his own http://xrl.us/nnur
Yves made an appeal for more tests that exercise Unicode pattern matching. Not the funky, contrived examples that the porters come up with to exercise obscure parts of the code base, but real honest-to-goodness matches that happen in Real Life.
If you test it, they will come http://xrl.us/nnus
Tels came back with some late comments on the trie enhancements.
Yves had an answer to all of the questions, going so far as to
suggest that a nice optimisation would be to convert patterns
consisting of a single EXACT node to an index() call, thereby
avoiding regmatch() altogether.
Avanti! http://xrl.us/nnut
Yves caught a bug before anyone could step on it.
The unguarded moment http://xrl.us/nnuu
Dave Mitchell also did some more follow-up work on his efforts to remove recursion from the regular expression engine.
Switched on state http://xrl.us/nnuv
Adam Kennedy had recently encountered two discussions of reversible debugging, and wondered how plausible this could be for Perl. (Reversible debugging is the idea of being able to undo the previous ``r''un or ``n''ext statement, and recover the previous state. (History trivia: Roedy Green wrote a language called Abundance in the 70s that provided support for this mechanism. He called it jaunting)).
Randy W. Sims thought that a system that merely recorded the current system state by snapshot would be good enough. You could not go back in time and change things, but you could at least look at them again. Useful if you missed something.
Some things cannot be reversed anyway: system calls, network writes, different paths might cause cleanup handlers to never be called.
Jesse pointed to Leon Brocard's Devel::ebug, which offers an
undo mechanism. Mark-Jason Dominus mapped out the different
points on the continuum, showing what was easy, simple and fairly
useless all the way to difficult, hard and very useful, but we
should be able to get something good at a reasonable cost. His
suggestion was to teach the debugger to record all the commands
during a session, and then allow one to restart the session, replaying
all the commands up to some point in the stream.
Richard Foley explained that this latter trick was already available
in the current debugger. It's called rerun.
David Nicol suggested forking a new copy of the program at each perl statement, and communicate between the prior state processes to determine the difference in state. This would of course entail immense resource costs. Alternately, a fork and dump on each statement would merely chew disk space, instead of RAM. But at the end of the thread, Adam came back and explained that he didn't really care to go back and re-execute the program from a given point. He just wanted to be able to go back and look at what had happened as a disinterested observer (a bit like TV).
We need a "come from" instruction http://xrl.us/nnuw
perlbrowse Dave Mitchell announced a new release of the perlbrowse tool,
that allows the porters to look at the source code from the point
of view of the repository, and view the changes made to the code
base over time.
Whiter than #ffffff http://xrl.us/nnux
One of the biggest threads to hit p5p in months, which goes to show that there's still interest in the beast.
Yves Orton kicked off the thread, writing about the problem of Perl 5 and Perl 6 and it sounded like a replay of the Osborne Effect (Adam Osborne built a phenomenally successful portable computer in the 1980s, and preannounced the arrival of a new faster model. People stopped buying the current model, waiting for the new one to be released. The competition ate the company).
Some of the main points:
We need to get 5.10 out the door.
A Seal of Approval for qualifying CPAN modules.
Send Andy Lester some papers that he can publish on perlfoundation.org.
Perl is a write-only language. Parentheses, no parenthesis, or
versus ||. (This subthread pushed Tom Christiansen into
penning a couple of missives).
At least two technical book publishers consider that the Perl market is done.
Perl sucks as a desktop applications language (think: Tk, perlWx).
Tk looks ancient, perlWx documentation sucks.
gettext doesn't work very well if the initial language is not English
and you want to add a English translation.
Installing Perl on Windows should be as easy as installing Firefox plug-ins.
Prerequisites specifications for CPAN modules sucks.
Perl releases are slowing down, as are new language enhancements.
Core support for Win32 sucks.
Yeah, but we knew all that http://xrl.us/nnuy
Hash::Util::FieldHash Lots of internals talk here, and not enough time to summarise the ramifications.
It's that U magic http://xrl.us/nnuz
enums Thanks to Intel's optinagging compiler, Andy Lester straightened out
the mess of enums being mixed with non-enums, especially in
relation to svtypes.
http://xrl.us/nnu3
Andy then tidied up S_qsortsvu() and made embed.fnc refer to it
(for error checking) and used the macro'ed version where applicable.
http://xrl.us/nnu4
does it Rafael added the DOES method to UNIVERSAL following on from
chromatic's desire several weeks ago to try and make UNIVERSAL
more useful, or rather, less abused. He then hinted that chromatic
was probably in the best place to write the appropriate documentation.
http://xrl.us/nnu5
Andy slotted it into the right place in embed.fnc.
http://xrl.us/nnu6
And chromatic delivered the documentation goods.
http://xrl.us/nnu7
FAIL(F) hp-ux 11.23/64 (ia64/2 cpu) Dave Mitchell observed that this failure ``was caused by the interesting
fact that a detached thread still counts towards the A thread
exited while %d threads were running warning, in violation of the
docs''. And so he fixed it.
http://xrl.us/nnu8
Sys::Syslog (#35406) Keisuke Hirata filed a bug report about Sys::Syslog and bundled
the patch used to fix the problem. Sbastien Aperghis-Tramoni announced
that it had been included in version 0.16, now available from your
neighbourhood CPAN mirror.
A dream comes true http://xrl.us/nnu9
Both David Landgren and Dr. Ruud tried to make sense of this bug. Sébastien A-T thought that it may be a manifestation of another Cygwin-ism that had caused him grief in the past.
Too Unix to be Windows, too Windows to be Unix? http://xrl.us/nnva
Data::Dumper fails to escape bless class name (#39420) Any users out there still using ' (apostrophe) in their class
names? Well don't, because Data::Dumper has forgotten about it.
http://xrl.us/nnvb
Carp can't find Carp::Heavy (#39440) Funny how things come in waves. This was another manifestation of last week's bug about what happens to Carp when perl runs out of file handles.
FITNR http://xrl.us/nnvc
By another strange coincidence, this bug has also been fixed in the next release.
Waiting for 5.10 http://xrl.us/nnvd
tie %SIG (#39504) John Gardiner Myers discovered a way to make perl dump core, and
suspected that the act of tieing %SIG may have something
to do with it.
Where porters fear to tread http://xrl.us/nnve
And another data point:
Curiouser and curiouser http://xrl.us/nnvf
3 closed and 6 open: 1491 total http://xrl.us/nnvg
They're all here http://rt.perl.org/rt3/NoAuth/perl5/Overview.html
Test-Harness version 2.62 uploaded by Andy Lester.
http://xrl.us/nnvh
Sys-Syslog version 0.15 uploaded by Sébastien Aperghis-Tramoni.
http://xrl.us/nnvi
A new website, win32.perl.org opened its doors this week.
Contribute! Contribute! http://xrl.us/nnvj
Ravi Sastry Kadali, from the IBM zOS USS Development team reported having ported 5.8.7 onto IBM z/OS. The team had to make some changes to the source, and wanted to contribute them back to the porters.
Rafael explained that ideally, they should try and port blead, and
send the required changes back for integration. These changes can then
be ported over to the maintenance branch if not compatibility problems
are encountered.
http://xrl.us/nnvk
Steve Stiert then sent in the patch anyway. Jarkko Hietaniemi had a look at it.
http://xrl.us/nnvm
Daniel Frederick Crisman had yet another shot at reworking quote-like
operators in perlop .
When in doubt, use brute force http://xrl.us/nnvn
Yves sent in a patch to fix some segmentation faults during global
destruction (in relation to his regexp work) and also tweaked
Benchmark to stop it from hitting infinite loops.
Two for the price of one http://xrl.us/nnvo
Tom Schindl was bitten by the map in void context memory wastage
problem, that neither foreach, nor more recent versions of Perl, for
that matter.
http://xrl.us/nnvp
Jarkko reworked the gcc warnings selection mechanism, to allow the porters, and more specifically Andy Lester, to enable all sorts of wacky compiler switches to see what happens.
--warn-if-non-halting http://xrl.us/nnvq
Philip M. Gollucci attempted to perform a speed comparison from
5.6.2 to blead , but the results were flawed because he used perl
binaries compiled with debugging. He promised to redo them again,
without debugging.
http://xrl.us/nnvr
Shlomi Fish wrote in to say that he had found a bug with perl -d
not printing the current code line.
So perlbug it http://xrl.us/nnvs
John E. Malberg found a show stopper in blead concerning Unicode. Sadahiro Tomoyuki proposed a fix, and Craig A. Berry committed the change.
http://xrl.us/nnvt
Yitzchak Scott-Thoennes recalled that in the core there are actual arrays of arrays (and not arrays of references to arrays), but could not recall where.
http://xrl.us/nnvu
Salvador Fandiño took another stab at adding macros to Perl5.
http://xrl.us/nnvv
Andreas J. Koenig discovered that the APC archive is missing files 28373-28377.
We have backups, right? http://xrl.us/nnvw
This summary was written by David Landgren. Last week's summary...
http://xrl.us/nnvx
... attracted a reply from Yves, who followed up on the issue
of pluggable regexp engines for maint. The problem is one of
a bad design call, made in the distant past, for which Nicholas
Clark is pondering a ``deeply evil'' workaround.
Tels also replied, with a plug for his Math::String module,
that lets one perform automagical increments (and decrements)
on just about anything that looks incrementable. And if it
doesn't work with Unicode, file a bug report so that Tels can
fix it.
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:
http://www.landgren.net/perl/
Weekly summaries are published on http://use.perl.org/ and posted on a mailing list, (subscription: perl5-summary-subscribe@perl.org ). The archive is at http://dev.perl.org/perl5/list-summaries/ . 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.