Traditional exception handling is more like equipping each Klingon with a dead-man's switch attached to a nuke. Dying by accident in someone else's nuke blast would be a death without honor. So the point of the Perl 6 exception model is to try to keep as many warriors alive as long as possible in order to inflict the maximum possible damage on the enemy, which is our highest goal, after all. :) -- Larry Wall, on the issue of dealing gracefully with errors.
IO::Dir destructor Peter Dintelmann looked at Zefram's fix for IO::Dir (which was
clobbering global variables like $!) and realised that the same
criticism could be levelled at DirHandle. So he adopted Zefram's
initial approach to DirHandle as well, and Steve Peters made
sure it got into blead.
http://xrl.us/b7fqo
Paul Fenwick laid out a more detailed description of his exceptions hierarchy for Perl 5. chromatic thought that an attempt at hierarchies, whether they be singly or multiply rooted, was doomed to failure. A much better approach would be to define a set of roles, and any given exception would implement one or more roles. Larry Wall explained what was going on over in Perl 6 and detailed a most interesting shift in the design philosophy of handling exceptions, in that errors are treated as data rather than interruptions to control flow.
Mark J. Reed offered a useful Klingon directive (yIQap pagh yIHegh!).
Joshua ben Jore reminded people of the danger of inspecting $@ and
wiping it in the process, thus resulting in a error-on-error condition
known as exception suicide. He went on to show how paranoid you
really need to be in order to avoid exceptions vanishing through
carelessness.
they *are* out to get you http://xrl.us/b7fqq
autodie.pm design questions Paul, no Bruce, Fenwick had a couple of questions to ask the porters.
The first one was how to apply autodie semantics to system, which
is more difficult that one might imagine since it has some really
weird variant calling conventions. The second question was the
possibility to grant autodie core status at some point (the
problem being that it currently depends four non-trivial non-core
modules).
Roland Giersig accidentally opened old wounds asking why it couldn't
be possible to retrofit the new autodie semantics onto Fatal,
but Aristotle Pagaltzis and Abigail made short order of that issue.
Of the prerequisite modules needed by autodie, Roland thought that
IPC::System::Simple sounded like something useful in its own right.
As to the system issue, Abigail thought that it would be nice to
autodie along with the rest, as long as there was an no autodie
'system' escape hatch.
Paul, err Bruce, no Paul sneaked in a subsequent question, which was
should a bare use autodie enable the Fatality of everything it
knew how to lay its hands on, or just a conservative (but nevertheless
broad) subset. In other words, does it mean use autodie ':all' or
use autodie ':default'? The latter approach is innately conservative,
forcing the client code to ask explicitly for any semantics for which
no solid concensus has emerged.
Paul dumped his current thoughts into POD here:
http://github.com/pfenwick/autodie/tree/master/DESIGN.pod
and admitted it would be better to put it in a wiki, but the tram he was in at the time didn't have wireless (loser!).
Another issue (the bloke just kept sneaking them in) is that old
Fatal messages sucked the big one, and Paul wanted to make them
prettier in the new version. Aristotle Pagaltzis argued that their
ugliness had never hindered widespread adoption of the module in
the first place. Therefore it was important to retain the bletcherously
ugly error messages of Fatal, and introduce soothing, elegant
messages in autodie, since that would only increase the velocity
of adoption of the latter. And you never could eliminate the
possibility that someone, somewhere, was scraping the exact output
of a Fatal error message.
a fatal abstraction http://xrl.us/b7fqs
PERL5OPT -M processed after command line -M? David Golden was surprised to discover that the effect of a -M
switch in the PERL5OPT environment variable takes place after
a -M switch on the command line. And I imagine that few people
would have expected the contrary.
Rafaël Garcia-Suarez argued that it was intentional, since
traditionally, the command-line overrides the environment and
when one thinks about the precedence of module loading it sort
of makes sense. On the other hand -T gets noticed straight away.
David agreed that while that was useful behaviour, the inconsistency
bothered him. What was lacking was the documentation that would
explain which switches overrode which context.
http://xrl.us/b7fqu
Yuval Kogman reported on some research he had conducted with llvm-gcc (the Low Level Virtual Machine project) by using it to compile the 5.10 perl codebase. He saw an improvement of around 15% (which is a serious improvement) but had trouble trying to get XS modules to work (which is a serious disadvantage). On the other hand, it may open the door to putting Perl apps on Apple's iPhone, which earns serious cred points in the mindshare department.
Steve Peters wanted to see how current Intel C++ and Sun Studio fare, since they usually leave gcc in the dust.
the return of the compiler wars http://xrl.us/b7fqw
Devel::Cover In a side remark on the Summer of code mentor applications starting
(and ending) next week thread, Paul Johnson, on the subject of
modules that maintain a rather incestuous relationship with the
core, said that for better or worse, Devel::Cover was destined
to remain a module that remained outside of core.
He admitted that while it would be nice to have someone step in and fix up its breakage following significant changes to the core, he was sufficiently lucid to realise that the resources available to develop the core are thin, and that in releasing such a module, he was responsible for expending the appropriate effort to keep it working. Joshua ben Jore admitted that his love-hate relationship with the core veered strongly to the hate side.
all raise your glasses to Paul http://xrl.us/b7fqy
??{ } closure semantics Yuval Kogman also ran into some trouble in some hairy patterns using
??{ } constructs, and assumed it was a bug. Rafaël confirmed the
issue, saying that it was indeed a bug, Dave Mitchell knew about it,
and wanted to fix it. David Nicol asked if it was more difficult that
just incrementing the refcount on lexicals encountered within
??{...} constructs and maintaining a list of same to decrement
when the regexp goes out of scope, and Dave said yes.
it is *more* difficult than that http://xrl.us/b7fq2
H.Merijn Brand idly wondered how we were supposed to smoke from a git repository, since a nice monotonic sequence, a feature that endears us to Perforce, is replaced by a SHA-1 digest that only an idiot savant could love. Rafael sidestepped the issue by saying that the 'HEAD' tag gets you what you need.
git describe http://xrl.us/b7fq4
Chia-liang Kao noticed that 5.10.0 spews out "used only once: possible typo" where before silence used to reign. He wondered if it was possible for the core to suppress the warning if it emanated from an XS boot block.
Nicholas Clark thought that something needed to be done to address the problem, so Chia-liang will file a bug next week.
direct from the time machine http://xrl.us/b7fq6
Write portable self destruct code for tests to stop them burning CPU in infinite loops. This needs to avoid using alarm, as some of the tests are testing alarm/sleep or timers.
(This TODO is perhaps too terse. Interested parties are invited to ask the porters for an idea of what really needs to be done. Even TODOing the TODO would be helpful at this point).
Archive::Tar was building corrupt archives on VMS John E. Malmberg finally figured out exactly what the problem was with
Archive::Tar on VMS. So he fixed it.
now to get it applied http://xrl.us/b7fq8
Steffen Müller was embarrassed by a very silly mistake in his work on lib.pm. He corrected it, and Rafaël applied the correction.
http://xrl.us/b7fra
File::Copy & permission bits Following Craig A. Berry's belated follow-up on the change in File::Copy's
behaviour, Charles Bailey, made an even more belated follow-up, and it was
full of good stuff. Charles explained the original rationale behind copy
and syscopy and how things have drifted over the years since the original
implementation was drafted.
He also agreed that the new changes were likely to cause trouble on non POSIXish platforms. Abigail expressed surprise that the patch was even accepted in the first place, given the provisos surrounding it (that it was non-backwards compatible and that it was unclear what it would do on non-POSIX platforms).
While no clear concensus emerged as to what should be done, most people that commented suggesting that backwards compatibility was paramount, and in any event the code needs a thorough overhaul.
up on the blocks http://xrl.us/b7frc
stat() and -X for tied handles Robert Norris, who goes for tied file handles in a big way, delved into
the guts in order to allow stat and -X (where X is f, d, b, ...)
to do something sane. Rafaël Garcia-Suarez looked upon it favourably,
but waited to see if other people had any opinions, as he doesn't really
do tied handles at all.
http://xrl.us/b7fre
A correction for a bug in Pod::Simple meant that a work-around in
the test suite was no longer required. This made it into both blead
and the podlators kit.
http://xrl.us/b7frg
perldoc -q ... searches index entries (X<>) Renée Bäcker noticed a slew of X<> going to waste
in the perl FAQs, and therefore tweaked perldoc -q to look at them
when asked to perform a search.
http://xrl.us/b7fri
Steve Peters noted that Cygwin is now reliable, although the problem persists with Windows.
http://xrl.us/b7frk
CGI::Util, expire_calc (#33047) Tom Hospel wondered why his bug had been resolved, since he thought the initial problems were still present. Bram demonstrated why he thought otherwise, and invited Tom to point out where his reasoning was incorrect.
In any event, the bug should probably be addressed via Lincoln Stein, the module's author.
and we will synch http://xrl.us/b7frn
\n after $ in regex (#35877) Bram thought that a nice warning for this problem would be a warning
that said "Possible unintended interpolation of $\ in regular
expression', assuming it's actually possible. Yitzchak Scott-Thoennes
observed that there is an elegant work-around to silence the warning,
by way of ${\}.
assuming that's also possible http://xrl.us/b7frp
File::Temp and unsafe shell characters (#50146) Ed Avis failed to see why anyone still used of system when
IPC::Run was so much nicer.
'coz it ain't core? http://xrl.us/b7frr
pod2man needs to translate some ASCII characters (#50950) Steve Peters committed Mike Ward's patch that addressed the rendering of backticks. There was also the issue of a =head2 with two C<> blocks (#53974)
podlators are go http://xrl.us/b7frt http://xrl.us/b7frv
Warren Dodge reported a problem in ClearCase::CtCmd puking on a free().
Steve Peters asked if he was able to coax a full stack trace out of his
perl binary, but Warren wasn't sure how to do that and asked for guidance.
help me-e-e-e-e http://xrl.us/b7ftv
Steve Peters noted that thanks to change #33401 there's one less assumption in the code.
Posixly robust http://xrl.us/b7frx
Steve Peters verified that this problem is solved in 5.10. The next thing to do will be to see what a 5.8.8 debugging build thinks of it.
SMOD http://xrl.us/b7frz
Bram thought that the /n modifier for a non-interpolating regular
expression was a bad choice since it has a purpose in .NET patterns.
Ed Avis thought that /d would do just as well; the letter is not
important, the functionality is. Abigail could see the point in
having a generalised quoting mechanism for non-interpolating patterns,
à la q() and qq(), so long as it didn't break any existing
code. Ed pointed out that m/./d was currently a syntax error.
doomed episodic demigoddesses http://xrl.us/b7fr3
Rafael applied Goro Fiju's patch, who then realised that the second part of it was redundant since the first part of the patch ensured a level of consistency that obviated the need for a subsequent sanity check.
http://xrl.us/b7fr5
After some further discussion with Niko Tyni concerning the eventual security implications leaving this bug unchecked could have, Rafaël applied the patch.
http://xrl.us/b7fr7
Ian Goodacre supplied a goodly-sized chunk of documentation to explain hash subscripts more clearly. The patch was left on the table, perhaps because Ian did not supply another patch that removed the equivalent budget of words plus from elsewhere in the documentation.
documentation karma http://xrl.us/b7fr9
Reini Urban suggested a fix and Rafaël applied it.
http://xrl.us/b7fsb
File::Spec Cygwin/Win32 case_tolerant improvement (#55162) Reini Urban attempted to improve File::Spec to make it behave better
on Windows and produced several iterations of a patch that still wasn't
accepted by midweek.
http://xrl.us/b7fsd
Kevin Ryde discovered that using an encoding on a file handle will
interfere with any subsequent attempts to change the flushing
behaviour (as in $|).
http://xrl.us/b7fsf
John Gardiner Myers found some more arguable behaviour in the handling of pattern matching in UTF-8 strings. Rafaël agreed with John, and stated that some of the misbehaviour will definitely need to be smacked down. Bram wrote some TODO tests.
http://xrl.us/b7fsh
Reini Urban discovered a long-standing bug in B::CC, which has been
there more or less since its inception. Steve Peters applied the patch.
and you are the first person to notice... oh wait http://xrl.us/b7fsj
Paul Fenwick was surprised to learn that using the diamond operator on an
file handle in void context will warn if the file was opened for output,
but not if it was opened for append. So he took a look at pp_hot.c to
cook up a patch, and did so, but thought it was too easy. Especially since
more exotic file opening directives like +>> should not warn.
Rafaël thought that fixing that would require a new IoTYPE.
http://xrl.us/b7fsm
With Renée and Steve P getting their hands dirty along with Bram, the total bug count continues its downward trend. Hooray.
274 new + 1286 open = 1560 (+15 -60) http://xrl.us/b7fso http://rt.perl.org/rt3/NoAuth/perl5/Overview.html
The official announcement from Russ Allbery.
http://xrl.us/b7fsq
Steve Peters reported that Pod::Simple was slightly broken for a
while since it initiated an undeclared dependency on HTML::Parser
in order to scoop out decode_entites. But everything was straightened
out in the end.
a simple matter of programming http://xrl.us/b7fss
On the other hand, when Nicholas Clark integrated the upgrade into 5.8.9 he discovered that a new test was failing in UTF-8 locales.
http://xrl.us/b7fsu
Sébastien Aperghis-Tramoni realised that 0.25 had not made it into core, which was a pity, since it contained fixes to all sorts of issues.
http://xrl.us/b7fsw
Nicholas went out for a smoke, and discovered lots of errors on
blead. Sébastien was able to reduce the problem to a tiny test case
involving constant. After some more detective work, he confirmed
that there was a discrepancy between ExtUtils::Constant in blead
and on CPAN. And further that the discrepancy was an issue of proxy
subroutines.
This in turn led to the discovery that there wasn't any documentation
for ExtUtils::Constant::ProxySubs, which, at the very least, makes
it difficult for anyone to realise that it could be of any use.
Working together, Sébastien and Nicholas ironed out all the bugs that were deemed necessary to fix.
some 5.6.0 breakage observed http://xrl.us/b7fsy
Paul Fenwick reported on his autodie.pm updates.
now with 5 minute video tutorial! http://xrl.us/b7fs2
Dave Mitchell observed that change #32881, to make Data::Dumper
handle blessed regexes properly caused 5.10-maint to dump core.
http://xrl.us/b7fs4
Bram spent most of the week helping Jochen Luig come to grips with some perl 5.10.0 installation issues on CentOS 5 but success eluded them.
http://xrl.us/b7fs6
John E. Malmberg proposed some correct CGI/t/upload*.t fixes for VMS, but learnt that Craig A. Berry had already beaten him to it.
http://xrl.us/b7fs8
Re: Change 33987: Upgrade to podlators-2.1.0
http://xrl.us/b7fsu
Some of the porters helped out with some datapoints to help Torsten
Foertsch understand some Linux::Smaps kernel/perlio dependencies.
try this at home http://xrl.us/b7fta
Rafaël Garcia-Suarez took Renée Bäcker's patch to make
perldoc -q slurp do something useful.
no more bug #8806 http://xrl.us/b7ftc
Steve Peters idly wondered if there was a VAX Ultrix box alive somewhere. If so, it would help solve bug #17146.
stalled until further notice http://xrl.us/b7fte
Renée Bäcker found a series of tickets in RT that could be closed. But when I went to look, Rafaël had already beaten me to it.
i feel inferior today http://xrl.us/b7ftg
Renée also noticed, regarding bug #19388, that the behaviour changed in Perl 5.10.
but is it any better? http://xrl.us/b7fti
Shlomi Fish announced that he was revamping the Perl FAQs. Rafaël
directed him to the perlfaq-workers mailing list.
http://xrl.us/b7ftk
This Week on perl5-porters - 25-31 May 2008 http://xrl.us/b7ftn
This summary was written by David Landgren.
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 or attending a YAPC to help support the development of Perl.