This Week on perl5-porters - 6-12 April 2008
David Nicol: I have no idea what NaN(Q) is by the way; Google suggests
it is a furniture store in Japan.
Jarkko Hietaniemi: It's the strong, silent type of NaN. As opposed to
the hysteric, screaming type of NaN (aka NaNS).
-- David N. and Jarkko H., arcing tangents.
Topics of Interest
Make built-in list functions continuous
David Nicol continued the "unroll map into pushy for loops" thread
with a decent stab at an implementation using his "Macrame" macro
processing module.
Aristotle Pagaltzis stressed the usefulness of having a map block
producing n elements for each input element, where n could be a fixed
number greater than 1, an arbitrary number greater than 1, sometimes
0, and anything in between.
Some map blocks that use such behaviours lead to remarkably concise
code that would be very cumbersome to write in any other way.
http://xrl.us/bje2q
Perl @ 33536
The perl 5.8.9-tobe thread continued to attract traffic this week. Jan
Dubois reported back on the problems that the Vista platform was
causing.
It boils down to the tradeoff between keeping the binary interface
stable while preventing people from outside the core build from
reaching directly into the guts. All you have to do is define
"PERL_CORE" and you can do whatever you want.
Jan suggested that one way to do this would be to reorder the struct
members, on purpose, on each maintenance release. That would quickly
smoke out poorly-behaved modules and make them play by the rules.
Not so unsurprisingly, this met with a certain amount of favour with
the porters; at least, it wasn't dismissed out of hand.
keep your mits off
http://xrl.us/bje2u
Saving a pointer in every COP?
Nicholas Clark had a brainwave, realising that it should be feasible
to shave off a whole pointer from each COP struct. Sixteen hours
later, he delivered the patch to do just that.
He did leave us something to do, though: prove or disprove that it
made things go faster.
something to do^W^Wdone
http://xrl.us/bje2w
Jerry D. Hedden managed to spot a minor compiler warning, which
Nicholas promptly fixed.
but no benchmarks
http://xrl.us/bje2y
using "-Wall" in modules generate tons of warnings in core files
Gábor Szabó set about trying to build "Prima" with the 5.10 Strawberry
Perl distribution, and ran into grief over poorly nested comments (as
in "/* foo /* bar */").
Andy Dougherty explained that this problem also arose with fussy
compilers on Unix platforms, and there the issue was addressed by
substituting "/ *" (with a space) for the inner opening comment
marker. It was therefore a simple matter of inspection to figure out
how config.h was generated by Strawberry.
Jan Dubois took a look at win32/config_h.PL and found a substitution
operator that looked like it was to blame. Andy showed how the
"Configure" process used a "sed" one-liner to work around the damage.
http://xrl.us/bje22
TODO of the week
Last week's TODO caught James Bence's attention, so much so that he
outlined an approach he wanted to try and wondered if it would be
acceptable. Rafael Garcia-Suarez explained why it wouldn't (as it
would add to the burden of the pumpking). The primary design
consideration is that it must be able to go about its business in a
completely autonomous manner. That is, no sneaky command-line switches
to spoon-feed it what needs to be done.
figure out what it should do, then make it do it
http://xrl.us/bje24
(I should also mention that this whole TODO idea of the week is not my
own, bla^Wcredit must be given to Nicholas Clark).
Compressed man pages
Be able to install them. This would probably need a configure test to
see how the system does compressed man pages (same directory/different
directory? Same filename/different filename), as well as tweaking the
installman script to compress as necessary.
(I suspect even getting this working for two common platforms, such as
Linux, *BSD or Solaris, would be a fine start. Other local experts on
other platforms could piggy-back the work done to hit other targets).
Patches of Interest
Double magic with '\&$x'
Rafael Garcia-Suarez applied Vincent Pit's patch. Vincent had also
questioned the wisdom of allowing magic to be triggered merely by the
gathering of information to produce an error message.
Rafael commented that while in essence Vincent was correct, the
unresolved problem about what to do about code in the wild that may
depend of the precise text of an error message meant that it was
probably wiser just to let sleeping dogs lie.
so won't fix
http://xrl.us/bje26
Make "PL_AMG_names" and "PL_AMG_namelens" static
Nicholas Clark wrapped up the discussion explaining why these symbols
that had leaked out accidentally in 5.10 had to stay: some linkers
might blow up in their subsequent disappearance, even if the client
code makes no reference to them.
lazy lazy dynamic loading
http://xrl.us/bje28
"is_gv_magical" correctly checks "ISA"
Gerard Goossen, going through the code with a fine-toothed comb for
his ongoing Kurila project, spotted a mistake due to incorrect
aliasing of a C array. It could never have worked, but then again
no-one ever had problems with it. In any event, Rafael applied the
change since the test suite appears to remain satisfied.
and it does the right thing
http://xrl.us/bje3a
changes to perlsec.pod and call for removal of quicksort
John P. Lindeman offered a patch to clean some of the more glaring
documentation errors regarding sort in "perlsec", and this was applied
by Rafael. He also suggested that the sort subsystem based on the
Quicksort algorithm be removed, since these days the default mergesort
has much better worst-case characteristics for any pathological set of
data you care to throw at it.
Tom Horsley, who wrote the original implementation of quicksort for
perl, made no impassioned claim to keep it in, admitting that in fact
he thought the code had been axed years ago.
out of sorts
http://xrl.us/bje3c
New and old bugs from RT
regexp: unicode char causes a 'double free corruption' (#48156)
Niko Tyni identified the changes which had fixed this bug, but noted
that it is still present in the version of perl shipped with Debian
stable (no sniggering up in the back row please). Don Armstrong, in
the Debian bug report, believes that it may allow arbitrary code to be
executed, which if true, would be very serious indeed.
http://xrl.us/bje3r
"[[:print:]]" *versus* "\p{Print}" (#49302)
David Landgren offered a couple of tweaks to Robin Barker's prose.
Juerd Waalboer also identified a mistake.
or at least, a counter-example
http://xrl.us/bje3t
"pack 'A*'" and "pack 'a*'" untaint data in 5.10.0 (#52552)
Christopher E. Stith was surprised by the fact that in 5.10, the pack
formats 'a' and 'A' strip tainting off untrusted input, whereas 5.8
leaves it tainted. Andreas König identified change #24010 as the cause
for the change in behaviour. Alas, that particular change touched over
a dozen files (although 4 were POD files, and, surprise pp_pack.c was
touched)...
the monks have a go
http://www.perlmonks.org/?node_id=678463
I blame Unicode
http://xrl.us/bje3v
Unwanted warning from XS BOOT (#52572)
reading the fine documentation leads to resolution
http://xrl.us/bje3x
"Bizarre copy of ARRAY in sassign at Carp/Heavy.pm" (#52610)
Chris Heath reported a variation on the bizarre copy theme, seen
regularly with "Carp". Nicholas Clark thanked him for the report,
which contained a snippet of code that he was able to shrink down even
farther.
This led him to conclude that the heart of the matter is an unwanted
interaction between Perl's internal stack and lexical pads, but had no
insight as to what exactly was happening.
http://xrl.us/bje3z
Perl 5.10 regression bug in match and substitution evaluation in list
context (#52658)
Wolf-Dietrich Möller spotted a regression between 5.8 and 5.10
concerning the use of an "e" modifier on a "s///" operator. Nicholas
Clark was dismayed to discover that the bug is also in 5.8.9-tobe,
definitely not good.
Andreas König fired up his binary search bug finder, and discovered
that not one, but two separate changes (#26332 and #26334) was at
fault. The first patch dealt with accelerating "s///e" expressions by
freeing intermediate temporaries, and the second dealt with "s///e"
that die in the right hand side result in memory leaks.
we got oursel's a show-stopper
http://xrl.us/bje33
regexp failure: "(?=)" turns into OPFAIL (#52672)
L. Mai reported another failure due to the 5.10 regexp engine
overhaul: the "match always" "(?=)" idiom flipped over to mean "match
never". Fortunately Ævar Arnfjörð Bjarmason came to the rescue with a
short, sharp patch to engine to restore the previous behaviour.
we do test for this now, right?
http://xrl.us/bje35
Consistent localization between "foreach (...)" and "while (<fh>)" (#52702)
Ed Avis wanted to have warnings issued when $_ was modified, due to
differences in how "while" and "for" behave. Rafael Garcia-Suarez
thought that using "my $_" in 5.10 was a much better idea.
http://xrl.us/bje37
crash when localizing a symtab entry (#52740)
Niko Tyni reported a failure smoking mod_perl2 2.0.4rc1 on 5.10, due
to the new implementation of constant subroutines, going so far as to
identify change #29544 as being the (ir)responsible party. No takers.
http://xrl.us/bje39
Crash perl with "binmode(STDOUT, ':encoding(wildybad)')" (#52786)
Todd Olson that one can make perl crash and burn on a variety of
platforms by giving it nothing more than a bogus I/O encoding name.
http://xrl.us/bje4b
Perl5 Bug Summary
1811 (+8 -4)
http://xrl.us/bje4d
http://rt.perl.org/rt3/NoAuth/perl5/Overview.html
New Core Modules
Math::BigRat version 0.22
Tels synched blead with CPAN, applied by Rafael.
http://xrl.us/bje4f
In Brief
Nicholas Clark committed change #33657 to split out
"S_refcounted_he_new_common()" from "Perl_refcounted_he_new_common()"
after having tested it out on four different machines. Alas, it still
managed to break the build for Jerry D. Hedden. Fortunately, Nicholas
was able to commit the subsequent change #33659 that fixed up the
breakage.
http://xrl.us/bje4h
Sérgio Durigan Júnior wondered about the status of the "-m32" flag
that is planned for 5.12, which would allow a 32-bit perl to be built
on a 64-bit platform. H.Merijn Brand said that for the moment it was a
non-starter.
imagine -Duse64bitall in reverse
http://xrl.us/bje4j
The "atan2" thread kicked up far more than you would ever want to know
about NaNs.
http://xrl.us/bje4m
Kurt Starsinic fixed up a long-standing perldata.pod typo.
http://xrl.us/bje4o
Alberto Simões took a shot at benchmarking the differences between
"say" and "print".
take two
http://xrl.us/bje4q
How not to file a bug. (was: "IO::Socket::accept()" doesn't fill
"io_socket_proto" information in sockets).
http://xrl.us/bje4s
Tels wanted to know if a delivery date for perl 5.8.9 had been fixed,
as this would help him figure out what to do with warnings occurring
in "Math::BigInt". Yitzchak Scott-Thoennes suggested that it should
warn at all during normal usage.
http://xrl.us/bje4u
Vincent Pit had some "perlclib.pod" tweaks applied.
memset, strtod and friends
http://xrl.us/bje4w
He also made emacs users the world over insanely jealous with a cool
patch to make "autodoc" generate a Vim XS syntax file. He admitted it
was slightly frivolous and probably not worth taking, but Vim lovers
should be happy to know it's out there.
you also need to love XS
http://xrl.us/bje4y
Reini Urban offered an experimental work-in-progress patch for the
upcoming cygwin-1.7 release with UTF-8 path support.
oh the pain
http://xrl.us/bje42
Last week's summary
30 March-5 April 2008
http://xrl.us/bje44
About this summary
This summary was written by David Landgren. I plan to be offline next
week; there will be no summary for 13-19 April. The summary will
return the following week, with a bumper fortnight issue.
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 or attending a YAPC to help support the development of
Perl.
