This week, our p5p summary will describe a lot of little bugs, some of which were fixed, some of which weren't, in a lot of different areas of the perl interpreter.
It took a few iterations to get Jeff Pinyan's rewritten Unicode tables support in bleadperl right. Notably, some files that were named similarly were causing problems on case-insensitive filesystems.
Jeff Pinyan and Sadahiro Tomoyuki also wondered about the technical difficulty of adding support for the POSIX character classes
[.xy.] in perl's regular expressions.
Meanwhile, Sadahiro and Nicholas Clark solved bug #29149, about substr() returning an erroneous return value on an UTF-8 string on which pos() was set; he trimmed it down to a UTF-8 cache bug.
There was some discussion about the integration of the new enhanced warning Use of uninitialized value $foo in the stable branch of perl. After all, it only affects warning messages emitted by perl, and not perl's behaviour.
Meanwhile, this new warning caused some miscellaneous failures in the smoke tests, when running under some UTF-8 locales, and with input/output set to UTF-8 by default: this situation changed the way perl processed internal data, to which a defined variable name couldn't be attached anymore, and thus the warning message was modified (not reporting the variable name when it could not be figured out.)
Yitzchak Scott-Thoennes filed bug #29238, where he explores the interaction between the Existing eval via last warning and the Can't "last" outside a loop block fatal error -- the eval() in the first one masks the second, and probably shouldn't.
David Muir Sharnoff reported as bug #29019 that a read() on a non-blocking socket at eof returns undef and sets $! to EAGAIN (resource temporarily unavailable). Nicholas Clark replied that, since read() is buffered I/O, it shouldn't be used on unbuffered inputs, such as non-blocking sockets. A bit later, Ton Hospel investigated what does eof() return in this situation (bug #29277): since the system call set errno to EAGAIN, perl's eof() shouldn't return
Gisle Aas found (bug #29102) that assigning to an open lexical filehandle was crashing perl.
Ton Hospel found an amusing bug (#29127): the return value of delete() applied to an empty hash slice seems to pick up the last value on the stack.
Dan Dascalescu found a six-character way to make perl segfault (bug #28986):
perl -e 'open m'
Those three bugs (among others) were fixed by Dave Mitchell.
Marcus Holland-Moritz unveiled a bug in OpenBSD's, via failures in perl's test suite. He also implemented a workaround.
Thorvaldur Gunnlaugsson found (bug #28993) that using the function eval("") in a BEGIN block mysteriously resets the $[ variable to 0 (if it was previously set to a non-zero value.)
Stas Bekman found a bug that appears when an embedded perl interpreter is cloned after having evaluated a string that declares a new subroutine: this leaks memory. This bug (#29018) can be observed via the <Perl> sections of an apache2/mod_perl configuration file.
Chip Salzenberg found that the special DB::sub() subroutine (used to write debuggers) suffers from restrictions: roughly, it can't call XSUBs and it can't use regular expressions. Details are given:
Chia-liang Kao announced that svk, his source control software written in perl on top of Subversion, can now mirror Perforce repositories, and in particular the perl source repository (which access is restricted to a few authorized people.) John Peacock is testing it.
After a discussion on the perl-xs mailing list, Steve Hay proposed to add a couple of new macros for XS coders, to push newly created mortal values on the stack.
This summary was written by Rafael Garcia-Suarez. Weekly summaries are published on http://use.perl.org/ and posted on a mailing list, which subscription address is firstname.lastname@example.org. Comments and corrections welcome.