| perldelta - what is new for perl v5.8.1 |
perldelta - what is new for perl v5.8.1
This document describes differences between the 5.8.0 release and the 5.8.1 release.
If you are upgrading from an earlier release like 5.6.1, read first the the perl58delta manpage, which describes differences between 5.6.0 and 5.8.0.
Mainly due to security reasons, the ``random ordering'' of hashes
has been made even more random. Previously while the order of hash
elements from keys(), values(), and each() was essentially random,
it was still repeatable. Now, however, the order varies between
different runs of Perl.
Perl has never guaranteed any ordering of the hash keys, and the ordering has already changed several times during the lifetime of Perl 5. Also, the ordering of hash keys has always been, and continues to be, affected by the insertion order.
The added randomness may affect applications.
One possible scenario is when output of an application has included
hash data. For example, if you have used the Data::Dumper module to
dump data into different files, and then compared the files to see
whether the data has changed, now you will have false positives since
the order in which hashes are dumped will vary. In general the cure
is to sort the keys (or the values); in particular for Data::Dumper to
use the Sortkeys option. If some particular order is really
important, use tied hashes: for example the Tie::IxHash module
which by default preserves the order in which the hash elements
were added.
More subtle problem is reliance on the order of ``global destruction''. That is what happens at the end of execution: Perl destroys all data structures, including user data. If your destructors (the DESTROY subroutines) have assumed any particular ordering to the global destruction, there might be problems ahead. For example, in a destructor of one object you cannot assume that objects of any other class are still available, unless you hold a reference to them. If the environment variable PERL_DESTRUCT_LEVEL is set to a non-zero value, or if Perl is exiting a spawned thread, it will also destruct the ordinary references and the symbol tables that are no longer in use. You can't call a class method or an ordinary function on a class that has been collected that way.
The hash randomisation is certain to reveal hidden assumptions about some particular ordering of hash elements, and outright bugs: it revealed a few bugs in the Perl core and core modules.
To disable the hash randomisation in runtime, set the environment
variable PERL_HASH_SEED to 0 (zero) before running Perl (for more
information see PERL_HASH_SEED in the perlrun manpage), or to disable the feature
completely in compile time, compile with -DNO_HASH_SEED (see INSTALL).
See Algorithmic Complexity Attacks in the perlsec manpage for the original rationale behind this change.
In Perl 5.8.0 all filehandles, including the standard filehandles, were implicitly set to be in Unicode UTF-8 if the locale settings indicated the use of UTF-8. This feature caused too many problems, so the feature was turned off and redesigned: see Core Enhancements.
The version strings or v-strings (see Version Strings in the perldata manpage) feature introduced in Perl 5.6.0 has been a source of some confusion-- especially when the user did not want to use it, but Perl thought it knew better. Especially troublesome has been the feature that before a ``=>'' a version string (a ``v'' followed by digits) has been interpreted as a v-string instead of a string literal. In other words:
%h = ( v65 => 42 );
has meant since Perl 5.6.0
%h = ( 'A' => 42 );
(at least in platforms of ASCII progeny) Perl 5.8.1 restores the more natural interpretation
%h = ( 'v65' => 42 );
The multi-number v-strings like v65.66 and 65.66.67 still continue to be v-strings in Perl 5.8. Note, however, that v-strings as a whole are now deprecated, see Deprecation Warnings.
The -C switch has changed in an incompatible way. The old semantics of this switch only made sense in Win32 and only in the ``use utf8'' universe in 5.6.x releases, and do not make sense for the Unicode implementation in 5.8.0. Since this switch could not have been used by anyone, it has been repurposed. The behavior that this switch enabled in 5.6.x releases may be supported in a transparent, data-dependent fashion in a future release.
For the new life of this switch, see UTF-8 no longer default under UTF-8 locales, and -C in the perlrun manpage.
Perl 5.8.1 uses the /d switch when running the cmd.exe shell
internally for system(), backticks, and when opening pipes to external
programs. The extra switch disables the execution of AutoRun commands
from the registry, which is generally considered undesirable when
running external programs. If you wish to retain compatibility with
the older behavior, set PERL5SHELL in your environment to cmd /x/c.
In Perl 5.8.0 many Unicode features were introduced. One of them was found to be of more nuisance than benefit: the automagic (and silent) ``UTF-8-ification'' of filehandles, including the standard filehandles, if the user's locale settings indicated use of UTF-8.
For example, if you had en_US.UTF-8 as your locale, your STDIN and
STDOUT were automatically ``UTF-8'', in other words an implicit
binmode(..., ``:utf8'') was made. This meant that trying to print, say,
chr(0xff), ended up printing the bytes 0xc3 0xbf. Hardly what you had
in mind unless you were aware of this feature of Perl 5.8.0. The problem
is that the vast majority of people weren't: for example in RedHat releases
8 and 9 the default locale setting is UTF-8, so all RedHat users got
UTF-8 filehandles, whether they wanted it or not. The pain was
intensified by the Unicode implementation of Perl 5.8.0 (still) having nasty
bugs, especially related to the use of s/// and tr///.
(Bugs that have been fixed in 5.8.1)
Therefore a decision was made to backtrack the feature and change it
from implicit silent default to explicit conscious option. The new
Perl command line option -C and its counterpart environment
variable PERL_UNICODE can now be used to control how Perl and Unicode
interact at interfaces like I/O and for example the command line
arguments. See the perlrun manpage for more information.
You can also now use safe signals with POSIX::SigAction. See POSIX::SigAction in the POSIX manpage.
In Perl 5.8.0 the so-called ``safe signals'' were introduced. This means that Perl no longer handles signals immediately but instead ``between opcodes'', when it is safe to do so. The earlier immediate handling easily could corrupt the internal state of Perl, resulting in mysterious crashes.
However, the new safer model has its problems too. Because now an opcode, a basic unit of Perl execution, is never interrupted but instead let to run to completion, certain operations that can take a long time now really do take a long time. For example, certain network operations have their own blocking and timeout mechanisms, and being able to interrupt them immediately would be nice.
Therefore perl 5.8.1 introduces a ``backdoor'' to restore the pre-5.8.0
(pre-5.7.3, really) signal behaviour. Just set the environment variable
PERL_SIGNALS to unsafe, and the old immediate (and unsafe)
signal handling behaviour returns.
Formerly, the indices passed to FETCH, STORE, EXISTS, and
DELETE methods in tied array class were always non-negative. If
the actual argument was negative, Perl would call FETCHSIZE implicitly
and add the result to the index before passing the result to the tied
array method. This behaviour is now optional. If the tied array class
contains a package variable named $NEGATIVE_INDICES which is set to
a true value, negative values will be passed to FETCH, STORE,
EXISTS, and DELETE unchanged.
The syntaxes
local ${$x}
local @{$x}
local %{$x}
now do localise variables, given that the $x is a valid variable name.
The copy of the Unicode Character Database included in Perl 5.8 has been updated to 4.0.0 from 3.2.0. This means for example that the Unicode character properties are as in Unicode 4.0.0.
There is one new feature deprecation. Perl 5.8.0 forgot to add some deprecation warnings, these warnings have now been added. Finally, a reminder of an impending feature removal.
Version Strings (v-strings) have been deprecated and they will be
removed in some future release of Perl. Each v-string will trigger
the warning v-strings are deprecated. The marginal benefits of
v-strings were greatly outweighed by the potential for Surprise and
Confusion. If you really want to continue using v-strings but not to
see the deprecation warnings, use:
no warnings 'deprecated';
Pseudo-hashes were deprecated in Perl 5.8.0 and will be removed in
Perl 5.10.0, see the perl58delta manpage for details. Each attempt to access
pseudo-hashes will trigger the warning Pseudo-hashes are deprecated.
If you really want to continue using pseudo-hashes but not to see the
deprecation warnings, use:
no warnings 'deprecated';
Or you can continue to use the the fields manpage pragma, but please don't expect the data structures to be pseudohashes any more.
5.005-style threads (activated by use Thread;) were deprecated in
Perl 5.8.0 and will be removed after Perl 5.8, see the perl58delta manpage for
details. Each 5.005-style thread creation will trigger the warning
5.005 threads are deprecated. If you really want to continue
using the 5.005 threads but not to see the deprecation warnings, use:
no warnings 'deprecated';
The $* variable controlling multi-line matching has been deprecated
and will be removed after 5.8. The variable has been deprecated for a
long time, and a deprecation warning Use of $* is deprecated is given,
now the variable will just finally be removed. The functionality has
been supplanted by the /s and /m modifiers on pattern matching.
If you really want to continue using the $*-variable but not to see
the deprecation warnings, use:
no warnings 'deprecated';
PerlIO::get_layers(FH) returns the names of the PerlIO layers active on a filehandle.
utf8::is_utf8() has been added as a quick way to test whether a scalar is encoded internally in UTF-8 (Unicode).
B::Concise
B::Deparse
Benchmark - an optional feature, :hireswallclock, now allows
for high resolution wall clock times (uses Time::HiRes).
CGI
charnames - one can now have custom character name aliases.
CPAN - there is now a simple command line frontend to the CPAN.pm module called cpan.
Data::Dumper - a new option, Pair, allows choosing the separator between hash keys and values.
DB_File
Devel::PPPort
Digest::MD5
Encode - significant updates on the encoding pragma functionality (tr/// and the DATA filehandle, formats). The ISO 8859-6 conversion table has been corrected (the 0x30..0x39 erroneously mapped to U+0660..U+0669, instead of U+0030..U+0039), the UTF7 encoding has been added.
libnet
Math::BigInt
MIME::Base64
NEXT - diamond inheritance now works
Net::Ping
podlators
Pod::LaTeX
PodParsers
Pod::Perldoc - complete rewrite
Scalar::Util - new utilities: refaddr, isvstring, looks_like_number, set_prototype.
Storable - can now store code references (via B::Deparse, so not foolproof).
Term::ANSIcolor
Test::Harness - now much more picky about extra or missing output from test scripts.
Test::More
Test::Simple
Text::Balanced
Time::HiRes - use of nanosleep(), if available, allows mixing subsecond sleeps with alarms.
threads - several fixes, for example for join() problems and memory
leaks. In some platforms (like Linux) that use glibc the memory
footprint of one ithread has been reduced by several hundred kilobytes.
threads::shared - many memory leaks have been fixed.
Unicode::Collate
Unicode::Normalize
Win32::GetFolderPath
Win32::GetOSVersion - now returns extra information
The Perl debugger (lib/perl5db.pl) has now been extensively documented and bugs found while documenting have been fixed.
perldoc has been rewritten from scratch to be more robust and featureful.
perl573delta has been added to list the differences between the (now quite obsolete) development releases 5.7.2 and 5.7.3.
perl58delta has been added: it is the perldelta of 5.8.0, detailing the differences between 5.6.0 and 5.8.0.
perlartistic has been added: it is the Artistic License in pod format, making it easier for modules to refer to it.
perlgpl has been added: it is the GNU General Public License in pod format, making it easier for modules to refer to it.
perlmacosx has been added to tell about the installation and use of Perl in Mac OS X.
perlos400 has been added to tell about the installation and use of Perl in OS/400 PASE.
The UNIX standard Perl location, /usr/bin/perl, is no longer overwritten by default if it exists. This change was very prudent because so many UNIX vendors already provide a /usr/bin/perl, but simultaneously many system utilities may depend on that exact version of Perl, so better not to overwrite it.
One can now specify installation directories for site and vendor man and HTML pages, and site and vendor scripts. See INSTALL.
One can now specify a destination directory for Perl installation
by specifying the DESTDIR variable for make install. (This feature
is slightly different from the previous Configure -Dinstallprefix=....)
See INSTALL.
gcc versions 3.x introduced a new warning that caused a lot of noise
during Perl compilation: gcc -Ialreadyknowndirectory (warning:
changing search order). This warning has now been avoided by
Configure weeding out such directories before the compilation.
One can now build subsets of Perl core modules by using the
Configure flags -Dnoextensions=... and -Donlyextensions=...,
see INSTALL.
In Cygwin Perl can now be built with threads (Configure -Duseithreads).
In newer FreeBSD releases Perl 5.8.0 compilation failed because of trying to use malloc.h, which in FreeBSD is just a dummy file, and a fatal error to even try to use. Now malloc.h is not used.
Perl is now known to build also in Hitachi HI-UXMPP.
Mac OS X now installs with Perl version number embedded in
installation directory names for easier upgrading of user-compiled
Perl, and the installation directories in general are more standard.
In other words, the default installation no longer breaks the
Apple-provided Perl. On the other hand, with Configure -Dprefix=/usr
you can now really replace the Apple-supplied Perl (please be careful).
Mac OS X now builds Perl statically by default. This change was done
mainly for faster startup times. The Apple-provided Perl is still
dynamically linked and shared, and you can enable the sharedness for
your own Perl builds by Configure -Duseshrplib.
Perl has been ported to IBM's OS/400 PASE environment. The best way to build a Perl for PASE is to use an AIX host as a cross-compilation environment. See README.os400.
Yet another cross-compilation option has been added: now Perl builds on OpenZaurus. See the Cross subdirectory.
Tru64 gcc 3.2.1 -O3 toke.c dropped to -O2 because of gigantic
memory use otherwise.
Tru64 can now build Perl with the newer Berkeley DBs.
Building Perl on WinCE has been much enhanced, see the wince subdirectory.
There have been many fixes in the area of anonymous subs, lexicals and closures. Although this means that Perl is now more ``correct'', it is possible that some existing code will break that happens to rely on the faulty behaviour. In practice this is unlikely unless your code contains a very complex nesting of anonymous subs, evals and lexicals.
binmode(SOCKET, ``:utf8'') only worked on the input side, not on the output side of the socket. Now it works both ways.
For threaded Perls certain system database functions like getpwent()
and getgrent() now grow their result buffer dynamically, instead of
failing. This means that at sites with lots of users and groups the
functions no longer fail by returning only partial results.
Perl 5.8.0 had accidentally broken the capability for users to define their own uppercase<->lowercase Unicode mappings (as advertised by the Camel). This feature has been fixed and is also documented better.
In 5.8.0 this
$some_unicode .= <FH>;
didn't work correctly but instead corrupted the data. This has now been fixed.
FETCH etc may now safely access tied values (ie resulting in a recursive call to FETCH etc).
At startup Perl blocks the SIGFPE signal away since there isn't much Perl can do about it. Previously this blocking was in effect also for programs executed from within Perl. Now Perl restores the original SIGFPE handling routine, whatever it was, before running external programs.
Linenumbers in Perl scripts may now be greater than 65536, or 2**16. (Perl scripts have always been able to be larger than that, it's just that the linenumber for reported errors and warnings have ``wrapped around''.) While scripts that large usually indicate a need to rethink your code a bit, such Perl scripts do exist, for example as results from generated code. Now linenumbers can go all the way to 4294967296, or 2**32.
Encode: if a filehandle has been marked as to have an encoding, unmappable characters are detected already during input, not later (when the corrupted data is being used).
PerlIO::scalar; reading from non-string scalars (like the special variables, see the perlvar manpage) now works.
Linux
HP-UX
VMS
poll(), and IO::Poll
now uses the vendor-supplied function if detected.
A rare access violation at Perl start-up could occur if the Perl image was
installed with privileges or if there was an identifier with the
subsystem attribute set in the process's rightslist. Either of these
circumstances triggered tainting code that contained a pointer bug.
The faulty pointer arithmetic has been fixed.
The length limit on values (not keys) in the %ENV hash has been raised
from 255 bytes to 32640 bytes (except when the PERL_ENV_TABLES setting
overrides the default use of logical names for %ENV). If it is
necessary to access these long values from outside Perl, be aware that
they are implemented using search list logical names that store the
value in pieces, each 255-byte piece (up to 128 of them) being an
element in the search list. When doing a lookup in %ENV from within
Perl, the elements are combined into a single value. The existing
VMS-specific ability to access individual elements of a search list
logical name via the $ENV{'foo;N'} syntax (where N is the search list
index) is unimpaired.
The piping implementation now uses local rather than global DCL
symbols for inter-process communication.
File::Find could become confused when navigating to a relative
directory whose name collided with a logical name. This problem has
been corrected by adding directory syntax to relative path names, thus
preventing logical name translation.
Win32
fork() emulation has been fixed.
The return value of the ioctl() built-in function was accidentally
broken in 5.8.0. This has been corrected.
The internal message loop executed by perl during blocking operations
sometimes interfered with messages that were external to Perl.
This often resulted in blocking operations terminating prematurely or
returning incorrect results, when Perl was executing under environments
that could generate Windows messages. This has been corrected.
Pipes and sockets are now automatically in binary mode.
The four-argument form of select() did not preserve $! (errno) properly
when there were errors in the underlying call. This is now fixed.
All the warnings related to pack() and unpack() were made more
informative and consistent.
The old version
A thread exited while %d other threads were still running
was misleading because the ``other'' included also the thread giving the warning.
It is not illegal to clear a restricted hash, so the warning was removed.
You must specify the block of code for sub.
The old version
Invalid [] range "%s" in transliteration operator
was simply wrong because there are no ``[] ranges'' in tr///.
Self-explanatory.
The padding spaces would appear after the newline, which is probably not what you had in mind.
If you think this
$x & $y == 0
tests whether the bitwise AND of $x and $y is zero, you will like this warning.
This warning should have been already in 5.8.0, since they are.
You cannot read() (or sysread()) from a closed or unopened filehandle.
This warning should have been already in 5.8.0, since they are.
Something pulled the plug on a live free variable, Perl plays safe by bailing out.
An illegal user-defined Unicode casemapping was specified.
Something modified the values being iterated over. This is not good.
The marginal benefits of v-strings were greatly outweighed by the potential for Surprise and Confusion.
These news matter to you only if you either write XS code or like to
know about or hack Perl internals (using Devel::Peek or any of the
B:: modules counts), or like to run Perl with the -D option.
The embedding examples of the perlembed manpage have been reviewed to be
uptodate and consistent: for example, the correct use of
PERL_SYS_INIT3() and PERL_SYS_TERM().
Extensive reworking of the pad code has been conducted by Dave Mitchell.
Extensive work on the v-strings by John Peacock.
UTF-8 length and position cache: to speed up the handling of Unicode (UTF-8) scalars, a cache was introduced. Potential problems if an extension bypasses the official APIs and directly modifies the PV of an SV: the UTF-8 cache does not get cleared as it should.
APIs obsoleted in Perl 5.8.0, like sv_2pv, sv_catpvn, sv_catsv, sv_setsv, are again available.
Certain Perl core C APIs like cxinc and regatom are no longer available. This is intentional. They never should have been available with the shorter names, and if you application depends on them, you should (be ashamed and) contact perl5-porters to discuss what are the proper APIs.
Certain APIs like Perl_list are no longer available without their
Perl_ prefix. If your XS module stops working because some
functions cannot be found, in many cases a simple fix is to add the
Perl_ prefix to the function and the thread context aTHX_ as the
first argument of the function call. This is also how it should
always have been done: letting the Perl_-less forms to leak from the
core was an accident. For cleaner embedding you can also force this
for all APIs by defining at compile time the cpp define PERL_NO_SHORT_NAMES.
Perl_save_bool() has been added.
Regexp objects (those created with qr) now have S-magic rather than
R-magic. This fixed regexps of the form /...(??{...;$x})/ to no
longer ignore changes made to $x. The S-magic avoids avoid dropping
the caching optimization and making (??{...}) constructs obscenely
slow (and consequently useless). See also Magic Variables in the perlguts manpage.
Regexp::Copy was affected by this change.
-DL removed (the leaktest had been broken and unsupported for years,
use alternative debugging mallocs or tools like valgrind and Purify).
Verbose modifier v added for -DXv and -Dsv, see the perlrun manpage.
In Perl 5.8.0 there were about 69000 separate tests in about 700 test files, in Perl 5.8.1 there are about 75000 separate tests in about 750 test files. The exact numbers depend on the Perl configuration and on the operating system platform.
The hash randomisation mentioned in Incompatible Changes is definitely problematic: it will wake dormant bugs and shake out bad assumptions.
If you want to use mod_perl 2.x with Perl 5.8.1, you will need mod_perl-1.99_10 or higher. Earlier versions of mod_perl 2.x do not work with the randomised hashes. (mod_perl 1.x works fine.) You will also need Apache::Test 1.04 or higher.
Many of the rarer platforms that worked 100% or pretty close to it with perl 5.8.0 have been left a little bit untended since their maintainers have been otherwise busy lately, and therefore there will be more failures on those platforms. Such platforms include Mac OS Classic, IBM z/OS (and other EBCDIC platforms), and NetWare. The most common Perl platforms (Unix and Unix-like, Microsoft platforms, and VMS) have large enough testing and expert population that they are doing well.
Tied hashes do not currently return anything useful in scalar context, for example when used as boolean tests:
if (%tied_hash) { ... }
The current nonsensical behaviour is always to return false, regardless of whether the hash is empty or has elements.
The root cause is that there is no interface for the implementors of tied hashes to implement the behaviour of a hash in scalar context.
The subtest 2 of lib/Net/Ping/t/510_ping_udp.t might fail if you have an unusual networking setup. The test is trying to send a UDP ping to the IP address 127.0.0.1.
IBM z/OS and other EBCDIC platforms continue to be problematic regarding Unicode support.
The choice of malloc (the C-level memory management interface) when building Perl is problematic in FreeBSD.
Using FreeBSD's system malloc for Perl was found to be very slow: in some cases that was 200 times slower than using the Perl malloc. One such case is file input: for example
# slurping the whole compressed Perl source code into $a
if (open F,"perl-5.8.1.tar.gz") { local $/; $a=<F> }
is about 200-250 times slower with the system malloc than with the Perl malloc.
One could use Perl's malloc (Configure -Dusemymalloc), but that was found to cause random core dumps in FreeBSD with multithreaded programs. No such problems were found in other platforms, however.
A decision was made to stick with the system malloc, regardless of the performance problems.
With certain HP C compiler releases (e.g. B.11.11.02) you will get many warnings like this:
cc: "/usr/include/sys/socket.h", line 504: warning 562: Redeclaration of "sendfile" with a different storage class specifier: "sendfile" will have internal linkage.
This warning, however, is not serious and can be ignored.
The test t/uni/tr_7jis.t is known to report failure under 'make test' or the test harness with certain releases of IRIX (at least IRIX 6.5 and MIPSpro Compilers Version 7.3.1.1m), but if run manually the test fully passes.
In the latest Tru64 releases (e.g. v5.1B) the gcc (3.3) cannot be used to compile a threaded Perl (-Duseithreads) because the system <pthread.h> file doesn't know about gcc.
As of the 5.8.0 release, sysopen()/sysread()/syswrite() do not behave
like they used to in 5.6.1 and earlier with respect to ``text'' mode.
These built-ins now always operate in ``binary'' mode (even if sysopen()
was passed the O_TEXT flag, or if binmode() was used on the file
handle). Note that this issue should only make a difference for disk
files, as sockets and pipes have always been in ``binary'' mode in the
Windows port. As this behavior is currently considered a bug,
compatible behavior may be re-introduced in a future release. Until
then, the use of sysopen(), sysread() and syswrite() is not supported
for ``text'' mode operations.
The following things might happen in future. The first publicly available releases having these characteristics will be the developer releases Perl 5.9.x, culminating in the Perl 5.10.0 release. These are our best guesses at the moment: we reserve the right to rethink.
-A option.
A new operator // (defined-or) will be available. This means that
one will be able to say
$a // $b
instead of
defined $a ? $a : $b
and
$c //= $d;
instead of
$c = $d unless defined $c;
The operator will have the same precedence and associativity as ||.
A source code patch against the Perl 5.8.1 sources will be available
in CPAN as authors/id/H/HM/HMBRAND/dor-5.8.1.diff.
unpack() will default to unpacking the $_.
Various Copy-On-Write techniques will be investigated in hopes
of speeding up Perl.
CPANPLUS, Inline, and Module::Build will become core modules.
5.005 Threads Will Be Removed
V-strings (Version Strings) Will Be Removed
The $* Variable Will Be Removed
Pseudohashes Will Be Removed
If you find what you think is a bug, you might check the articles recently posted to the comp.lang.perl.misc newsgroup and the perl bug database at http://bugs.perl.org/ . There may also be information at http://www.perl.com/ , the Perl Home Page.
If you believe you have an unreported bug, please run the perlbug
program included with your release. Be sure to trim your bug down
to a tiny but sufficient test case. Your bug report, along with the
output of perl -V, will be sent off to perlbug@perl.org to be
analysed by the Perl porting team. You can browse and search
the Perl 5 bugs at http://bugs.perl.org/
The Changes file for exhaustive details on what changed.
The INSTALL file for how to build Perl.
The README file for general stuff.
The Artistic and Copying files for copyright information.
| perldelta - what is new for perl v5.8.1 |