This Week on perl5-porters - 30 January-5 February 2006
Perl 5.8.8 was released this week.
perlvar.pod: $^X isn't necessarily "argv[0]"
Yitzchak Scott-Thoennes patched perlvar.pod to point out that $^X
could contain information from /proc/self/exe, on platforms that
supported it. But even on platforms that do support it, such as Linux,
you can still run into grief with "chroot"ed environments.
Nevertheless, Rafael Garcia-Suarez liked the patch enough to accept
it. Nicholas Clark pointed out that the name is different on FreeBSD
and Solaris.
Alan Burlison showed how to find the name of executable on Solaris
without having to rely on access to /proc, which interested Rafael.
Gisle Aas was unable to compile Alan's snippet, and it turned out that
the technique Alan was relying on was only fairly recently made
available on Solaris. So he coded another approach.
An improvement to $^X
http://xrl.us/jxs7
An earlier, similar attempt
http://xrl.us/jxs8
$^E and Borland on Win32
Steve Hay wondered about the differences between
# borland
perl -e "open F, 'nothere' or die $^E"
Died at -e line 1.
# MSVC
perl -e "open F, 'nothere' or die $^E"
The system cannot find the file specified at -e line 1.
so he wrote some code to find out what was going on. And afterwards he
still couldn't figure out where and when the Windows "GetLastError"
was being called. Jan Dubois clarified the issue concerning whether
what "GetLastError" returns is relevant or not. And I think the
discussion stopped there.
arenaroots patch
Jim Cromie sent in a consolidated patch that implements putting all
the arenas into a singly-linked list. He then sent in a patch to
implement meta-arenas, which allow perl to separate the
meta-information about the arena from the arena itself, which results
in some small savings of memory.
Consolidation
http://xrl.us/jxs9
Separation of meta-data
http://xrl.us/jxta
http://xrl.us/jxtb
http://xrl.us/jxtc
Segmentation fault when matching regexps
Lukasz Debowski filed bug #38379 that showed how to make perl dump
core when trying to match a string against a particularly arduous
regular expression. Yves Orton explained that the problem came from
excessive back-tracking that the pattern forced the regexp engine to
undertake, and offered a couple of ideas about how to rewrite the
pattern in question to make it more efficient.
Lukasz thought that it should be possible to get the regexp engine to
bail out gracefully under such circumstances and provide a more
informative error message about the problem.
Patches, as they say, are welcome.
http://xrl.us/jxtd
Typo in perlfunc.pod
Eagle-eyed "p.boven" spotted a documentation error in "perlfunc.pod",
filed as bug #38380. Trying to find the error by reading the patch
alone is just about impossible, fortunately a description of the
problem was included in the report. Nicholas applied the patch.
Try and spot the difference by reading the patch
http://xrl.us/jxte
Exhausting memory when printing large, deep hashes
Paul Boutros filed bug #38384, wherein traversing a large hash of
hashes uses a phenomenal amount of memory. Gisle Aas realised that the
problem was merely an issue of numbers being interpreted as strings,
thus causing the hashes' "SvIV" structures to be upgraded to the
larger "SvPV" structures.
(The trick is to use "0+$foo{bar}" to force numeric context).
http://xrl.us/jxtf
5.8.8
With the lyrics from Bein' Green, by Joe Raposo, Nicholas Clark
released the long-awaited Perl version 5.8.8 to the world.
It's out there
http://xrl.us/jxtg
Alternate lyrics
http://members.tripod.com/Tiny_Dancer/green.html
Camelpacks and Activestate
There is an issue or two surrounding Camelpack, which is bundle of
Activestate and MingW. One issue being that it might be violating
Activestate's license terms, and/or the O'Reilly camel trademark, and
even possibly R. J. Reynolds as well.
http://xrl.us/jxth
XS::Connect, or making C/Perl thunk faster
Chase Venters has problems with DateTime, in that it is too slow. He
thought about recasting some of the code into C, but learnt that the
authors had already tried that, and noticed an overall slowdown,
because of the cost of thunking (crossing between Perl and C). And
wondered whether there was a way of reducing the cost, and put forward
a couple of ideas. Glenn Linderman came up with a pretty good
refinement.
Tony Cook mentioned something called "Imager" that already lets one do
something along those lines (or at least, the next release will).
Nick Ing-Simmons suggested Chase take a look at both "Tk" and and
"DBI", which have been refining this concept for a decade. "Tk", in
particular, uses Perl to generate (print) C code, which helps cut down
on typos.
http://xrl.us/jxti
"autouse" noisy with 5.8.8
Rafael noticed that "autouse" may now produce warnings of the form
Prototype mismatch: sub main::bar: none vs ($) at
/usr/lib/perl5/5.8.8/autouse.pm line 57
and fixed it with change #27034. Nicholas wondered whether Rafael's
change was a fix, or just sweeping a problem under the carpet. Because
autouse is pretty hairy anyway, Rafael figured that it was all right.
http://xrl.us/jxtj
$, is undefined by default
In bug #38398, Roderick Schertler raised the point that it would be
nice if $, (the separator when printing in list context), was an empty
string rather than "undef", since that would allow
print join $,, @foo;
rather than
print join defined $, ? $, : '', @foo;
Abigail agreed, but thought that the proper way to fix the problem was
with a pragma:
no warnings;
print join $,, @_;
Jim Cromie suggested
local $, = '';
... but that only works if $, hasn't already been set. H.Merijn Brand
(implicitly) suggested recompiling Perl with his "dor" patch:
print join $,//'', @foo
http://xrl.us/jxtk
No more "Null(av|ch|cv|hv|sv)"
Steven Schubiger grepped through the code base and pinpointed a number
of places where "Nullav", "Nullsv" and the like continue to lurk, and
wondered whether they could be replaced by "NULL", and whether
"perlapi.pod" should note that they are deprecated.
Andy Lester pointed out some even more dubious code constructed that
need to be NULLified, and Nicholas admitted to being happy to see them
all go.
So Steven tackled "Nullav" and "Nullch" first, except that he also
deleted the definitions themselves, which would destroy backwards
compatibility, so he re-did the patch to keep them.
And then followed up with "Nullcv", "Nullsv" and the rest of the gang.
There followed a discussion about the possibly apocryphal hardware
whose null pointer bits are not all zero.
http://xrl.us/jxtm
Doing proper UTF-9 hash lookups
Nicholas proposed a patch to make the tokeniser itself UTF-8 aware.
This makes two TODO tests pass, but alas another test starts to fail.
A code audit is needed to make sure UTF-8-flagged keys are dealt with
correctly everywhere. And then the same thing needs to be done to the
pad code (the code that deals with lexicals).
http://xrl.us/jxtn
Unicode slowdown 5.8.6 "->" 5.8.7
Nicholas also revisited the Unicode slowdown and proposed an API
enhancement in order to minimise data copying. Sadahiro Tomoyuki (who
is the expert in these matters) came up with a better solution.
Nicholas implemented Tomoyuki's idea, and saw a fourfold improvement
in speed. Phil Pennock took the code for a spin, and sadly noted that
it shaved only two seconds off a forty-odd seconds run time.
Unicode melts my brain, again
http://xrl.us/jxto
Make parallel "make test" work
Nicholas Clark continued on a roll, trying to track down parallel
testing failures, and isolated a race condition between "List::Util"
and "Math::BigInt::Fastcalc". It looks as if the Makefile.PL for
"Math::BigInt::FastCalc" is being executed at precisely the same
moment that "List::Util" is being copied to ".../lib". Which is of
course extremely difficult to reproduce.
Steve Peters was able to identify that the real culprit was in fact
"overload.pm" and not "Math::BigInt::FastCalc".
http://xrl.us/jxtp
Nicholas found another parallel build race failure, which prompted
Alan Burlison to reminisce on problems he had encountered with
parallel tests at Sun.
http://xrl.us/jxtq
Trouble ahead for "nmake" on Vista?
Adam Kennedy heard that "nmake" (the aging version that can be legally
distributed) no longer runs on Vista (or at least, a current beta),
and wondered what the solution was. Apparently, there's a "dmake"
floating around that seems to have everyone's favour.
http://xrl.us/jxtr
Files in use in Windows and Cwd.pm
There was a really long thread to do with Window,
"ExtUtils::MakeMaker" and "Cwd.pm", but it came across badly threaded
on Xray and I had a hard time following it all. The pointers are:
Someone's MUA is playing silly buggers here
http://xrl.us/jxts
http://xrl.us/jxtt
http://xrl.us/jxtu
blead @ 27028 OpenVMS Alpha 8.2 - All tests pass
John E. Malmberg reported a perfect compile on VMS -- yay!
http://xrl.us/jxtv
5.8.8-RC1 on Cray
H.Merijn Brand gave perl a smoke test on Cray hardware. Oddly enough,
it appears to have about the same performance as a TRS-80. Just
running "Configure" alone took over 24 hours...
http://xrl.us/jxtw
Label optimised away when it shouldn't
Gisle noticed that
$ perl -le 'goto foo; if (0) { foo: print "hi" }'
produces
Can't find label foo at -e line 1.
Not that anyone should want to code like that, nor should it be fixed,
but that maybe the behaviour be noted somewhere in case someone comes
to grief over it. Paul Johnson pointed out the relevant part of
"perlfunc" that does in fact explain it.
http://xrl.us/jxtx
A good p5p mailing list archive
Steve Hay was looking for a good archive of the p5p mailing list. It
seems that xray had been down for a while, and so he looked at the
alternatives.
http://xrl.us/jxty
Enhancing "PERL_TRACK_MEMPOOL"
Nicholas took Jan Dubois's "PERL_TRACK_MEMPOOL" code and added some
functionality to it to improve the feedback when running under
"valgrind", and also added a "PERL_POISON" switch to improve the
chances of dying a quick death when buggy code tries to access memory
it no longer owns.
http://xrl.us/jxtz
Perl5 Bug Summary
1543 open issues.
http://xrl.us/jxt2
New Core Modules
CPAN-1.83_64
Andreas Koenig uploaded CPAN-1.83_64, which, if nothing goes wrong,
will become the 1.84 official release soonish.
http://xrl.us/jxt3
About this summary
This summary was written by David Landgren. I wasn't able to start
this until Thursday night, due to various constraints in Real Life
(such as drinking beer with the local Perl mongers last night). As a
result (of the late start, not the beer) the threshold for
interesting-enough-to-summarise threads was higher than usual. Also, I
didn't have the time for the witty URL titles. Sorry about that, but a
brief summary is better than no summary at all.
Information concerning bugs referenced in this summary (as #nnnnn) may
be viewed at http://rt.perl.org/rt3/Ticket/Display.html?id=nnnnn
Information concerning patches to maint or blead referenced in this
summary (as #nnnnn) may be viewed at
http://public.activestate.com/cgi-bin/perlbrowse?patch=nnnnn
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: [EMAIL PROTECTED]). The
archive is at http://dev.perl.org/perl5/list-summaries/. Corrections
and comments are welcome.
If you found this summary useful or enjoyable, please consider
contributing to the Perl Foundation to help support the development of
Perl.
--
"It's overkill of course, but you can never have too much overkill."
