All the Perl that's Practical to Extract and Report

This Week on perl5-porters (6-12 January 2003)

rafael on 2003-01-13T22:36:29

The porters were busy, and this week's report features a large number of different subjects, from portability and compilation to the proper semantics of method dispatch, not forgetting the usual amount of strange bugs. Read below about the latest potential evolutions of Perl 5.

Warning on UTF8 locales ?

In bug report #19743, Jarkko Hietaniemi summarized one of the main problems with UTF8 locales. Under an UTF8 locale, perl 5.8.0 implicitely turns on the UTF8-ness of input filehandles. Also, reading illegal UTF-8 data does not trigger any warnings; but later, trying to use the resulting data does. This is annoying.

The proposed solutions are a combination of the following points :

  • Don't turn on silently UTF8-ness of all filehandles; issue a mandatory message (when the developer hasn't made clear he accepts UTF8);
  • Don't turn on automagically UTF8-ness of all filehandles; require the script to have use()'d the locale pragma (or even a possible use locale qw(utf8)) to do so; (Ton Hospel later questions whether the lexical scoping behavior of this pragma is well-suited to solving this problem);
  • Reading illegal UTF8 data from a filehandle marked as UTF8 should issue a warning (but this is hard to achieve without slowing down all input).

Pathnames and Portability

There was quite a long thread, crossposted to perl5-porters and to the Cygwin developers list, about the proper way to handle path names on Cygwin (how it should deal with Windows-style and Unix-style pathnames, and how the conversions should be done). This is related to definiting the proper behavior of File::Spec::Cygwin. Among the points are : what to do with pathologic cases (files names legal on Windows but not on Cygwin), what exactly is an absolute path (it turns out that UNC paths and paths that start with a single backslash are considered absolute on Windows), etc. Then, the thread becomes verbose and off-topic.

share() empties hashes

Barrie Slaymaker noticed (last month) that marking a hash reference as shared clears the hash silently. Nick Ing-Simmons explains that it was implemented that way by laziness and for speed : copying the hash contents when sharing it means writing a internal routine to do this, and that should be safe regarding potential reference loops and already-shared internal data structures. Benjamin Goldberg suggests to introduce a separate rshare() function which does this recursive copying.

is-a basic type

Piers Cawley asked why method dispatch doesn't work with basic types; more precisely, while a blessed hashref is-a HASH (according to UNIVERSAL::isa()), it's not possible to call methods from package HASH on this blessed hashref. The discussion following this difficult question spawns the territories of Perl 6.

Profiler segfaults

Blair Zajac continued to investigate cases of segmentation fault within the Perl profiler Devel::DProf. Apparently, increasing a fudge factor (a stack size) solves his problems. (Ilya Zakharevich called this voodoo programming but Benjamin Goldberg provided a more logical explanation for it.)

Latest Berkeley DB

Bug #19884, reported by Martin Mokrejs, is about building perl 5.8.0 with Berkeley DB 4.1.25. This doesn't work, because (quoting Paul Marquess) Perl 5.8.0 was released before Berkeley DB 4.1 came out. There was a change in the Berkeley DB API that wasn't backward compatible. The recommended solution is to build perl without DB_File (by giving the -Ui_db command-line option to Configure) and to install later the latest DB_File from the CPAN.

Old bug reopened

Six months ago bug #15479 was fixed (this was about the simple statement %::="" leading to a core dump when warnings are enabled). John Peacock remarks that the fix isn't good enough, because it doesn't work when some operators are overloaded. John also makes a good job at trimming down the cause of this bug.

Sun's upcoming compiler

Alan Burlison (from Sun) told the Perl 5 porters about the upcoming version of the Sun C compiler. This future version is expected to break the hints file for Solaris in various ways (firstly, the detection of the compiler version, secondly, checking the presence of the libsunmath library). He volunteers to provide a patch.

In brief

Rafael Garcia-Suarez proposes to ship a simple RPM spec file with perl. Axel Thimm notes that this hides nasty portability pitfalls. He adds that it might be interesting to try an LSB (Linux Standard Base) packaged perl (see http://www.linuxbase.org/appbat/ ).

Rafael also changed the value of the internal variable ${^TAINT}, so it's now 0 without taint checks, 1 with -T, and -1 with -t or -TU. Previously it was only 0 or 1.

Ernest Lergon found bug #19767 pertaining to use of \p{} escapes within character classes in precompiled regular expressions.

Peter Scott reported bug #19815, about an amusing syntax error that confuses perl's lexer enough to make it unable to count lines properly.

Mark-Jason Dominus notices that fsync(2) is missing from the POSIX module (bug #19860). However the functionality is provided by IO::Handle::sync().

Brendan O'Dea sent a load of patches.

There are some smoke escaping from the Cygwin builds. Apparently Merijn now installs bleading edge snapshots of Cygwin on the machine he runs the some tests on.

About this summary

This summary brought to you by Rafael Garcia-Suarez. Summaries are available on http://use.perl.org/ and/or via a mailing list, which subscription address is perl5-summary-subscribe@perl.org . Comments and corrections are welcome.