This Week on perl5-porters - 1-10 September 2006
"Most version control systems have flexible commit hooks to
automatically perform actions" -- Michael G Schwern
"Obi Wan Perforce: This isn't the version control system you were
looking for. Move along, move along..." -- John Peacock
Topics of Interest
CPAN-ification of core modules - dual life modules
Gabor Szabo wanted to know if the dual-lifing of core modules on CPAN
was still on the agenda, and if so, was there a list of suitable
candidates, and/or modules that should never become dual-lifed. Rafael
Garcia-Suarez said that the process was done on an as-needed basis.
Dual-lifing is when a module is bundled with the core distribution,
but also available separately on CPAN. For instance, "Time::HiRes", a
module that for a long time was only available on CPAN, went into core
for 5.8.
It stood at version 1.48 when 5.8.0 was released, 1.66 when 5.8.3 was
released, and the current version on CPAN is around 1.90. If it wasn't
dual-lifed, you would have to wait until a new version of Perl came
out to get the upgrade. As it is, any of those versions of perl can be
upgraded to use the current version of Time::HiRes from CPAN, with all
the attendant bug fixes and improvements that the maintainer has made
to it over the past four years.
Dual-lifing the core modules, then, is simply giving this same
flexibility to modules that *only* come with the perl distribution,
and do not have a distinct distribution on CPAN.
Michael Schwern explained a couple of the tricks needed to test core
modules, said that back-porting the module to earlier perls would be a
bonus, and suggested "Exporter" and "Class::Struct" as two possible
immediate candidates.
Moving out today
http://xrl.us/rqwb
Dual life for "Exporter"
So Adriano Ferreira took Schwern's suggestion seriously and set about
dual-lifing the venerable "Exporter" module. Andy Lester noticed this,
and wondered what scratch it itched.
Michael explained that he thought cleaning up cruftiness was a good
enough reason. Giving people the chance of upgrading just "Exporter"
in order to take advantage of some new feature was a Good Thing,
especially if this would otherwise mean upgrading to the latest
version of Perl.
After this was explained to Andy, he wholeheartedly agreed, and
suggested that Adriano hook up with Ricardo Signes, who apparently has
a number of ideas on the subject.
Following that, H.Merijn Brand played devil's advocate and explained
what was bad about the idea. The porters decided that the current
method of doing things should be able to handle the points H.Merijn
raised.
Exporting Exporter
http://xrl.us/rqwc
Boolean Operator Evil
David Wheeler was writing a mini-language in Perl, but discovered that
the lazy evaluation semantics of "or" makes life complicated when you
try and overload it. Fergal Daly mentioned Tangram's trick of
co-opting "&" and "|", which works for most people most of the time.
Joshua ben Jore thought that this approach would be less pain than
grovelling around in "B::" space to deparse the op-tree (which David
had mooted as a possible approach).
chromatic mentioned that it is within the realms of the feasible (and
a little XS code) to override any op in the op-tree. Mark-Jason
Dominus suggested David get in touch with Audrey Tang, who does this
sort of mind-bending syntax overloading all the time.
Covert operations
http://xrl.us/rqwd
Trimming OPs
Jim Cromie posted a 90KB patch to combine the "op_next" and
"op_sibling" fields into a union, which could shave up to a fifth of
the size off the the total of a full op-tree. This might provide a
benefit to "mod_perl"-like environments (and hey! it might improve L2
cache behaviour).
As things stand, it nearly all works, and Jim asked for clarification
on a couple of points. After three days of deafening silence, Leon
Brocard wondered out loud whether the added complexity was worth it.
Now with less fat
http://xrl.us/rqwe
Patches of Interest
Use direct object constructor calls
chromatic landed a rather large documentation patch. The idea was to
replace all indirect constructor calls ("my $f = new Foo") by direct
calls ("my $f = Foo->new"), the reason being that the documentation
should encourage unambiguous and reliable programming practices.
Surprisingly enough, the patch elicited next to no comment (chromatic
and I discussed this at YAPC::Europe last week, imagining that it
would cause a meltdown on the list). Rafael applied it.
http://xrl.us/rqwf
He then did the same in lib/*, but ran into problems with
"IO::Compress" and "File::Temp" (at least), since they are both
dual-lifed. This will require a lot more coordination with the various
distribution maintainers and thus may fall within the realms of
fiddling with the bloody grammar[tm].
http://xrl.us/rqwg
Correction to POD for "FETCH_*_ATTRIBUTES"
Jerry D. Hedden was astonished to find a glaring seven year old
documentation error in attributes.pm. Which may mean that either
no-one uses them or no-one reads documentation. Whatever the reason,
Rafael applied Jerry's correction.
http://xrl.us/rqwh
Trie jumping
After a false start last month, Yves Orton managed to beat his latest
regular expression engine optimisation into submission.
Now, "/foo[xy]+|bar[tu]*|baz+/" goes really fast. That is,
fixed-string prefixes followed by metacharacters in alternations reap
the benefit of Yves' earlier trie work, whereas before, only
"/foo|bar|baz/" would have seen an improvement.
Jumping out of tries
http://xrl.us/rqwi
Unicode Character Database (UCD) 5.0.0 released
Jarkko Hietaniemi urged all porters to fire up their clay tablets and
drag out their styluses (ahem, *styli*), because the latest Unicode
definitions now include... cuneiform.
Think of the possibilities! And to bring those dreams to reality,
Jarkko provided a patch to update Perl with this information.
If it dumps core, will it shatter?
http://xrl.us/rqwj
C++ status update
Jarkko also phoned in with the latest reports about compiling the perl
source with a C++ compiler. He had managed to get quite far with
Tru64's C++ compiler, because it is reasonably lenient about array
declarations and disambiguating "delete" as a keyword. Other compilers
are more problematic.
Dominic Dunlop took it for a spin on Mac OS X with g++ 4.0.1 but ran
into problems right away with "Configure" getting a bit confused about
compiler capabilities. He then delivered a patch that cleared up many
of the issues.
http://xrl.us/rqwk
New and old bugs from RT
5.005_62: "make distclean" problem (#1647)
A bug opened in 1999 was fixed some time in the past seven years.
http://xrl.us/rqwm
"PerlIO::scalar" doesn't know how to truncate (#40241)
Bo Lindbergh and Nick Ing-Simmons continued to explore the issue of
what file descriptor functionality needs to be implemented to make
"PerlIO::Scalar" behave as much as possible like a file handle. "-s"
(for file size) would be useful, and last-modified-time would probably
have to be kluged in, but "chown" and "chmod" could probably be
no-ops. And so forth.
http://xrl.us/rqwn
"Math::BigInt" silently drops lib argument if not available (#40242)
Tels and Andreas König continued to look at this problem, and finally
agreed on the right approach to take. Now all that has to happen is
for Tels to implement it.
http://xrl.us/rqwo
"kill( 0, ... )" does not return alive status of child process (#40262)
"M" noted a discrepancy between the documentation and reality of a
"kill" call, to determine the status of a child. David Nicol came up
with a suitable tweak to the documentation, which was applied by
Rafael.
http://xrl.us/rqwp
Bug with "bignum" (#40263)
Leif Pedersen encountered a strange problem with "bignum", whereby
"$char = ord($char) - ord('A') + 10" appeared to have no effect. Tels
twisted the example slightly, and got it to produce a "Modification of
a read-only value" error, which led him to conclude that it was an
aliasing error.
http://xrl.us/rqwq
"xsubpp" ignores "PREFIX" when wrapping function declared with INTERFACE
(#40264)
Salvador Fandiño uncovered a nasty interaction in "xsubpp" between
"PREFIX" and "INTERFACE", and proposed a fix. Rafael agreed that the
change was necessary, even if it was not backwards compatible. He
suggested that the issue be raised with Ken Williams, who maintains
"ExtUtils::ParseXS" (since "xsubpp" is merely a wrapper around that).
http://xrl.us/rqwr
"PerlIO::scalar" doesn't respect readonly-ness (#40267)
Bo Lindbergh uncovered another beauty in "PerlIO::scalar". Mike Guy
hoped that a couple of variations on the theme would be also be fixed
by whoever fixed it. Rafael fixed it with change #28798, so let's hope
it all gets bedded down in the test suite.
http://xrl.us/rqws
subroutine call with "&" in "perlop" example (#40272)
Sebastian Schmidt suggested function call in an "s///e" operation
should lose its "&" sigil. Rafael agreed, so now it doesn't.
http://xrl.us/rqwt
Perl coredumps on nested "s///e" (#40274)
Jean-Louis Leroy showed how perl goes all pear-shaped when one tries
to nest "s///e" operations. In perl's defence, the code does look
interestingly abstruse.
http://xrl.us/rqwu
"Configure" problem (#40286)
Howard Shubs had a problem configuring perl on a path containing a
space. Andy Dougherty came up with a crude patch to backslash-escape
spaces in paths, but suspected that a number of other places may need
similar treatment. Kids, this is definitely something you can try at
home.
http://xrl.us/rqwv
Perl5 Bug Summary
10 more thrown onto the pile, for a total of 1531.
http://xrl.us/rqww
Close one today!
http://rt.perl.org/rt3/NoAuth/perl5/Overview.html
New Core Modules
* Michael G. Schwern rolled out "ExtUtils::MakeMaker" version
6.30_02, as a release candidate for 6.31. Craig A. Berry reported
all green lights on VMS.
http://xrl.us/rqwx
* Michael also push "Test::Builder/More/Simple" version 0.64_01 out
the door. Two important changes that people should be aware of:
5.8.0 threaded builds are no longer supported (because the
implementation at the time was too buggy) and the report output
has changed. This later point will be of great importance to any
module that scrapes test output.
http://xrl.us/rqwy
In Brief
One of the reasons it is worth pushing for C++-compilable sources, is
that it gets us things like inlined functions, IEEE floats and other
goodnesses. Things that probably exist in a current C++ compiler, but
won't, if ever come to C compilers.
http://xrl.us/rqwz
Yves Orton responded to a problem of "^C" segfaulting on Win32 by
pointing out that ^C is weird on Win32, as some of the implementation
is in fact running in a separate thread.
http://xrl.us/rqw2
Ask Bjørn Hansen nixed the idea of "security@" address for the
reporting of security vulnerabilities in perl, since past experience
has shown its purpose to be little more than a spam magnet.
http://xrl.us/rqw3
chromatic added handy index numbering to the "caller()" documentation
in "perlfunc". Useful stuff, applied.
http://xrl.us/rqw4
Nick Ing-Simmons gave further reasons why it was worth making the
codebase compile with a C++ compiler.
http://xrl.us/rqw5
About this summary
This summary was written by David Landgren. Yes, this summary covers
ten days worth of traffic. You lucky people.
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, please consider contributing to the
Perl Foundation to help support the development of Perl.
--
Much of the propaganda that passes for news in our own society is given
to immobilising and pacifying people and diverting them from the idea
that they can confront power. -- John Pilger
