This Week on perl5-porters - 27 November-3 December 2006
"So I don't really have an answer for you. When faced with a similar
situation, I punted and did a PerlIO_fdopen(fileno(stdio)) instead of
a PerlIO_importFILE(stdio). This way you know you are creating a
unixio layer which gets ref counted properly." -- Craig A. Berry,
somewhere in the sprawling PerlIO layers saga.
Topics of Interest
"PerlIO::Scalar" buffering
H.Merijn Brand wondered if he had uncovered a problem with performing
IO to a scalar, since at the end of a series of operations where the
scalar should have been emptied, a lone newline character remained.
http://xrl.us/tqdv
Whither 5.8.9
Jerry D. Hedden wondered if there was going to be a 5.8.9 release, and
if so would arrive before 5.10.0. H.Merijn Brand confirmed that a
5.8.9 release is going to happen some time, and it's likely to escape
from the lab before 5.10. No dates have been set for either event as
yet.
when it's ready
http://xrl.us/tqdw
Old "dmake"s for building perl on Win32...
Steve Hay noticed that recent changes to the Makefiles for Windows are
no longer understood by the older make programs floating around in the
Windows world. He felt that the best approach was to update the README
to explain the fix, in the unlikely event that someone encounters it,
and to move on.
time to go
http://xrl.us/tqdx
... and archaic Visual C++ compilers
Steve also noted that the compiler writers at Microsoft have played
the compiler command line switch fandango in recent versions, and now
warnings are emitted when certain deprecated switches are used. To get
things into shape means drawing the line at VC++ 5.0. This equates to
roughly a 10 year old compiler. The Makefile states that things should
be compatible all the way back to 2.0, but Steve has his doubts.
Nicholas Clark pointed out that if people really need to use a pre-5.0
compiler, all they have to do is change the new-name switch back to
old-name, and things should work. Steve updated README.win32 to note
this point.
http://xrl.us/tqdy
Trouble with PERLIO_INIT/PERLIO_TERM changes
Steve Hay and Jarkko Hietaniemi tried to make sense of the ongoing
PerlIO problems, in that reference counts were going too far negative
and bad things were happening as a result. And the code was a bit too
twisty for people to understand exactly what was happening. My eyes
sort of glazed over on this thread.
It eventually wound up being filed as a bug, see below.
http://xrl.us/tqdz
Also, in another thread, Steve also wondered whether
PerlIO_importFILE() needed to do a "refcnt_inc". Craig A. Berry showed
him how to test the hypothesis. One thing that came out of it is that
Steve noticed that the code was leaking file descriptors.
http://xrl.us/tqd2
Similarly, Yves Orton found a segfault on Windows due to one of the
many changes that were attempted this week to come to grips with these
issues. Rafaël was pretty sure about which patch was the problem, and
when Yves reverted it, his build succeeded in starting the test suite.
http://xrl.us/tqd3
"PathTools" now at svk.perl.org
Ken Williams announced that he'd thrown up "File::Spec" and "Cwd", on
a Subversion repository, and asked if there were any volunteers who
wanted to poke around at them. Philippe Bruhat bravely stepped
forward.
http://xrl.us/tqd4
Change #26513 breaks "Devel::SmallProf"
Nicholas Clark found that the addition of an assertion to ensure an AV
is always non-null made "Devel::SmallProf" curl up and die, and didn't
what to do about it.
http://xrl.us/tqd5
In the new month, the thread continued: Gisle Aas observed that the
problem went away of the build was configured with "usethreads".
Rafaël Garcia-Suarez realised that the heart of the problem lay in
toke.c, and made the appropriate change there.
http://xrl.us/tqd6
Minor patch for "Benchmark.pm"
Christophe Grosjean patched what he thought was a minor problem in
"Benchmark", but was really quite a significant bug when you think
about it: "cmpthese" used to ignore child process times, which led to
the fascinating outcome that a process that spawned an external
command appeared to be faster than something done in native Perl.
Highly illogical
http://xrl.us/tqd7
Patches of Interest
Abstract mempool header testing
Jarkko did a bit of work on memory pools, (that approximates the
conditions of Win32 on Unix), where memory allocation (especially in
threads) is a much trickier issue. Nicholas Clark gave a clear
explanation of how the mechanisms present in the source code fitted
together, and Jan Dubois completed the picture.
Based especially upon Jan's information, Jarkko produced a patch to
attempt to clear up some PerlIO problems. Jan, however, thought that
the code in question traded one set of problems (the mempool
allocations) for another (leaking stuff from the opcode tree instead),
and explained how to fix that problem
http://xrl.us/tqd8
As the month rolled over into December, Yves Orton took a stab at the
matter, pointing to a couple of spots in the code that he felt relied
on flawed assumptions. The patches went back and forth, and I lost
track of which patch undid what, and which ones were reverted.
Jan Dubois suggested one strategy for getting things straightened out,
and Jarkko coded it up as a patch. This seemed to fix things
sufficiently well that Steve Hay was able to unrevert a previously
reverted patch with everything still behaving itself.
Yves was most impressed, but also confused, and asked if there was
some documentation that explained The Big Picture. Steve Hay did his
best to illuminate the part of the canvas that he understood, hoping
that Jan Dubois would chip in and complete as necessary. And so he
did. Jan gave a thorough best exposé on the issue of mempools here:
http://xrl.us/tqd9
Now all we need is for someone to draw a diagram of the myriad ways in
which one may allocate memory in perl. The thread in the December,
from the top:
http://xrl.us/tqea
Building 5.8.8 on OS/2
Ilya Zakharevich delivered a patch to get 5.8.8 compiling again on
OS/2. There are still some tests that fail, and Ilya promised to
address them in a future patch.
http://xrl.us/tqeb
"perl5db" on "miniperl"
Similarly, Ilya noticed that perl5db.pl no longer worked with
"miniperl", due to a reliance on "IO::Handle", which was not yet fully
operational at that stage of the build process. So he tweaked things
to handle this situation gracefully.
http://xrl.us/tqec
"Test::Class" and blead
A test in "Test::Class" tested for a bogus regular expression, and was
supposed to die. The funny part was that thanks to Yves Orton's
tireless efforts, what used to be a meaningless string of characters
is now a legal pattern, so the pattern matches and the test fails.
So Nicholas Clark made the pattern even more bogus, to force the
pattern to fail again. Which judging by Yves's prodigious output,
should see the test safe for a least three weeks.
NO CARRIER
http://xrl.us/tqed
MacOSX debugger fork support
perldb.pl received some more patching goodness from Bo Lindbergh to
allow the debugger to be fired up in a separate terminal window on Mac
OS/X.
http://xrl.us/tqee
Make "B::Lint" use "Module::Pluggable"
Joshua ben Jore was one of the first people to take advantage of
"Module::Pluggable"'s new promotion to core, and rewrote "B::Lint" to
make use of it. He was immensely pleased, since it meant that he was
able to rip out the previous clumsy efforts.
The street finds its own uses for things
http://xrl.us/tqef
No code specified for -.
Jarkko discovered that an 'e' was dropped in the message "No code
specified for -e" if in fact there was no code.
dash before e except after p
http://xrl.us/tqeg
Perl regexp structures bifurcation
Yves Orton continued to deliver a stream of changes to introduce a
higher level of abstraction to the regular expression engine. This is
part of his ongoing work to permit pluggable engines.
http://xrl.us/tqeh
"\R" means something else in POSIX
Yves discovered that not only does the Unicode consortium care about
regular expressions, they even published a specification on the
matter. Yves read the spec, and learnt that they recommend using "\R"
as a best-practice sequence for specifying a generic line-break
marker.
This unfortunately clashes with "\R" as a relative back-reference
assertion in Perl's regular expression, but since we are still not out
of beta, Yves was quite happy to cast around for a different unused
character. After looking around at what was left of the alphabet, he
settled in "\g" (as in group). He implemented a generalised syntax
with curly braces, so that now "\g1" or "\g{1}" is the same as the
existing "\1", and conversely, that "\g-1" and "\g{-1}" as a
replacement for the now-defunct "\R1" syntax. (Yves, if you're reading
this, doesn't that mean you could drop "\k<foo>" and subsume it into
"\g<foo>"? You have 15 minutes to write the patch).
Yves also thought that implement "\R" shouldn't be too difficult
either, noting in passing that Philip Hazel will be delivering it in
the next release of PCRE.
14 minutes and counting
http://xrl.us/tqei
New and old bugs from RT
Segfault due to a semicolon inside an array reference (#40995)
Shlomi Fish discovered that a semi-colon, such as in "[ $i->func2();
]" gets the tokeniser in a tizz. Rafaël noted that this misbehaviour
was fixed in blead.
http://xrl.us/tqej
Setting $0 invalidates environment shown by "ps" (#41008)
Alexander Bluhm noted that assigning to $0 causes unwanted
side-effects on OpenBSD, and suggested a patch to fix things up.
Rafaël took it and applied it to "blead".
http://xrl.us/tqek
"(?(COND))" in pattern matching not working properly (#41010)
Karl Williamson discovered a bug where an alternation ("(a|b)") did
not match the same thing as a character class ("[ab]"). What was worse
was that the original program, when it didn't crash, the values of
other assorted variables in the program mysteriously changed. Yves
Orton figured out what the problem was, and developed a patch to fix
things up.
Applied
http://xrl.us/tqem
"backslash-G" does not appear to work in a while loop (#41011)
Jim Woodworth filed what he thought was a bug, but Dr. Ruud suspected
that he was a victim of crypto-context.
http://xrl.us/tqen
Changing $0 on darwin leads to excessive padding in "ps" (#38868)
The week continued in a $0 vein, with Darwin also causing trouble.
Chris Dolan sorted things out with the help of the C preprocessor.
You were expecting \0?
http://xrl.us/tqeo
Regexp optimizer loses its hopes too soon (#39096)
Yves Orton explored the reasons why the optimiser was ignoring obvious
shortcuts and pointed out why a naive solution wouldn't work, and the
problems that a generalised solution would have to consider.
It used to be simple
http://xrl.us/tqep
"LD_LIBRARY_PATH" changes not exported to Dynaloader,Perl 5.8.8 On
Solaris 8 (#40655)
A mysterious person reported that this bug was resolved.
Private and Confidential
http://xrl.us/tqeq
SIGSEGV with "pos()" in regexp (#40989)
Yves Orton noted that this bug no longer existed in "blead". This is
no doubt a pleasing side-effect of all the work that the engine has
received in recent times, although he was hard pressed to say which
change would have been responsible. By extension, he didn't expect the
fix to be backported to "maint".
One more reason
http://xrl.us/tqer
Yves noted that bug #39893 was probably another facet of the same
problem.
http://xrl.us/tqes
"File::Glob" braces with character classes erratic behaviour (#41027)
Kolano discovered some inconsistencies with globs, curly braces and
character classes.
http://xrl.us/tqet
Nested closures result in scoping related segmentation fault (#41028)
... and they are fixed in blead.
http://xrl.us/tqeu
"PerlIOUnix_refcnt_dec()" croaks in bleadperl since #29065 (#41030)
Steve Hay reported that since change #29065, his
"Win32-SharedFileOpen" croaked with a strange error from somewhere
within PerlIO. He traced the problem to a couple of patches, thinking
that while they were not incorrect, perhaps there was an underlying
bug which had only now been exposed as a result.
http://xrl.us/tqev
MSWin32: can't disable crlf trans. on STDxxx using :raw or binmode (#41034)
Vaclav Ovsik reported a problem with end-of-line markers on Windows.
He was not able to disable end-of-line translations on files opened by
default (such as "STDOUT" and "STDERR", but files opened afterwards
were just fine.
http://xrl.us/tqew
Perl5 Bug Summary
We almost crept under the 1500 barrier this week.
Once more into the breech
http://xrl.us/tqex
http://rt.perl.org/rt3/NoAuth/perl5/Overview.html
New Core Modules
* "IPC::Cmd" achieved core status this week. Steve Hay tweaked it
slightly to fix a test problem on Windows.
http://xrl.us/tqey
* "threads" moved onto version 1.53. This corrected some of the
problems that Steve Hay was experiencing with his smoke machine,
but now he saw that the test suite was hanging on a different
test.
http://xrl.us/tqez
* "File::Fetch" made it to the core. Johan Vromans wondered why this
functionality was not added to "LWP::UserAgent". The reason is
that "File::Fetch" is a wrapper for any number of command-line
tools and protocols than can be used to fetch a file (such as
"rsync", "LWP" itself, "curl", "wget" and the like).
http://xrl.us/tqe2
In Brief
Nicholas Clark taught the Aho-Corasick regexp patch how to behave in
the presence of threads.
http://xrl.us/tqe3
H.Merijn Brand discovered that "PerlIO::Scalar" ignores "format" and
"write", preferring instead to send the output to "STDOUT", instead of
the associated variable.
The last user of formats in the world
http://xrl.us/tqe4
"Time::Piece" is back from the future.
http://xrl.us/tqe5
David Landgren discovered that a literal before "\k<foo>" (a named
back reference) can really ruin a conference talk. Yves Orton resolved
the problem in 20 minutes, give or take a few.
http://xrl.us/tqe6
The 5.8.3 "sprintf" patch made its way into the Mac OS X Security
Update 2006-007.
http://xrl.us/tqe7
Paul Marquess independently rediscovered the joy of testing
filehandles with threaded 5.6.x perls, and decided to give it a wide
berth.
http://xrl.us/tqe8
http://xrl.us/tqe9
Bo Lindbergh proposed adding lexical scope to "SelfLoader", complete
with patch, but elicited no comments.
http://xrl.us/tqfa
Gabor Szabo and Abe Timmerman talked about the broken pages on
db.test-smoke.org.
In November
http://xrl.us/tqfb
and in December
http://xrl.us/tqfc
Yves Orton had some unhappy results with "Time::HiRes". After an
incredibly long thread, everything was sorted out.
http://xrl.us/tqfd
Johan Vromans was worried that adding "Module::Load::Conditional" to
the core would be like unleashing Pandora's box.
http://xrl.us/tqfe
Yves Orton wanted to know whether 'twas better to merge bugs or set up
dependencies.
Whatever lowers the count most
http://xrl.us/tqff
Jarkko tried in vain to sort out the UTF-8 -related failures that have
been occurring in the smokes since August. Nicholas Clark explained
the difficulties.
http://xrl.us/tqfg
Jarkko added some things to think about in "perltodo".
Nick's movie
http://xrl.us/tqfh
Jim Cromie noticed that the "install" target misses
"Compress/IO/{Base,Zlib}". The ideal solution involves using a more
orthodox directory layout for the modules: "Configure" gets too
confused otherwise.
http://xrl.us/tqfi
Rob "Sisyphus" puzzled over integer exponentiation under
"-Duse64bitint" on Cygwin, wondering why the same result could be
stored as an NV (a float) or an IV (an integer). Nicholas Clark could
no longer remember why, only that it was a conscious design decision
not to do integer exponentiation.
http://xrl.us/tqfj
About this summary
This summary was written by David Landgren.
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.
--
"It's overkill of course, but you can never have too much overkill."