perl5005delta - what's new for perl5.005
This document describes differences between the 5.004 release and this one.
Perl is now developed on two tracks: a maintenance track that makes small, safe updates to released production versions with emphasis on compatibility; and a development track that pursues more aggressive evolution. Maintenance releases (which should be considered production quality) have subversion numbers that run from 1 to 49, and development releases (which should be considered "alpha" quality) run from 50 to 99.
Perl 5.005 is the combined product of the new dual-track development scheme.
Starting with Perl 5.004_50 there were many deep and far-reaching changes to the language internals. If you have dynamically loaded extensions that you built under perl 5.003 or 5.004, you can continue to use them with 5.004, but you will need to rebuild and reinstall those extensions to use them 5.005. See INSTALL for detailed instructions on how to upgrade.
The new Configure defaults are designed to allow a smooth upgrade from 5.004 to 5.005, but you should read INSTALL for a detailed discussion of the changes in order to adapt them to your system.
When none of the experimental features are enabled, there should be very few user-visible Perl source compatibility issues.
If threads are enabled, then some caveats apply. @_ and $_ become lexical variables. The effect of this should be largely transparent to the user, but there are some boundary conditions under which user will need to be aware of the issues. For example, local(@_) results in a "Can't localize lexical variable @_ ..." message. This may be enabled in a future version.
Some new keywords have been introduced. These are generally expected to have very little impact on compatibility. See "New INIT keyword", "New lock keyword", and "New qr// operator".
Certain barewords are now reserved. Use of these will provoke a warning if you have asked for them with the -w switch. See "our is now a reserved word".
There have been a large number of changes in the internals to support the new features in this release.
Core sources now require ANSI C compiler
An ANSI C compiler is now required to build perl. See INSTALL.
All Perl global variables must now be referenced with an explicit prefix
All Perl global variables that are visible for use by extensions now have a PL_ prefix. New extensions should not refer to perl globals by their unqualified names. To preserve sanity, we provide limited backward compatibility for globals that are being widely used like sv_undef and na (which should now be written as PL_sv_undef, PL_na etc.)
If you find that your XS extension does not compile anymore because a perl global is not visible, try adding a PL_ prefix to the global and rebuild.
It is strongly recommended that all functions in the Perl API that don't begin with perl be referenced with a Perl_ prefix. The bare function names without the Perl_ prefix are supported with macros, but this support may cease in a future release.
See perlapi.
Enabling threads has source compatibility issues
Perl built with threading enabled requires extensions to use the new dTHR macro to initialize the handle to access per-thread data. If you see a compiler error that talks about the variable thr not being declared (when building a module that has XS code), you need to add dTHR; at the beginning of the block that elicited the error.
The API function perl_get_sv("@",GV_ADD) should be used instead of directly accessing perl globals as GvSV(errgv). The API call is backward compatible with existing perls and provides source compatibility with threading is enabled.
See "C Source Compatibility" for more information.
This version is NOT binary compatible with older versions. All extensions will need to be recompiled. Further binaries built with threads enabled are incompatible with binaries built without. This should largely be transparent to the user, as all binary incompatible configurations have their own unique architecture name, and extension binaries get installed at unique locations. This allows coexistence of several configurations in the same directory hierarchy. See INSTALL.
A few taint leaks and taint omissions have been corrected. This may lead to "failure" of scripts that used to work with older versions. Compiling with -DINCOMPLETE_TAINTS provides a perl with minimal amounts of changes to the tainting behavior. But note that the resulting perl will have known insecurities.
Oneliners with the -e switch do not create temporary files anymore.
Many new warnings that were introduced in 5.004 have been made optional. Some of these warnings are still present, but perl's new features make them less often a problem. See "New Diagnostics".
Perl has a new Social Contract for contributors. See Porting/Contract.
The license included in much of the Perl documentation has changed. Most of the Perl documentation was previously under the implicit GNU General Public License or the Artistic License (at the user's choice). Now much of the documentation unambiguously states the terms under which it may be distributed. Those terms are in general much less restrictive than the GNU GPL. See perl and the individual perl manpages listed therein.
WARNING: Threading is considered an experimental feature. Details of the implementation may change without notice. There are known limitations and some bugs. These are expected to be fixed in future versions.
See README.threads.
WARNING: The Compiler and related tools are considered experimental. Features may change without notice, and there are known limitations and bugs. Since the compiler is fully external to perl, the default configuration will build and install it.
The Compiler produces three different types of transformations of a perl program. The C backend generates C code that captures perl's state just before execution begins. It eliminates the compile-time overheads of the regular perl interpreter, but the run-time performance remains comparatively the same. The CC backend generates optimized C code equivalent to the code path at run-time. The CC backend has greater potential for big optimizations, but only a few optimizations are implemented currently. The Bytecode backend generates a platform independent bytecode representation of the interpreter's state just before execution. Thus, the Bytecode back end also eliminates much of the compilation overhead of the interpreter.
The compiler comes with several valuable utilities.
B::Lint is an experimental module to detect and warn about suspicious code, especially the cases that the -w switch does not detect.
B::Deparse can be used to demystify perl code, and understand how perl optimizes certain constructs.
B::Xref generates cross reference reports of all definition and use of variables, subroutines and formats in a program.
B::Showlex show the lexical variables used by a subroutine or file at a glance.
perlcc is a simple frontend for compiling perl.
See ext/B/README, B, and the respective compiler modules.
Perl's regular expression engine has been seriously overhauled, and many new constructs are supported. Several bugs have been fixed.
Here is an itemized summary:
Changes in the RE engine:
Unneeded nodes removed;
Substrings merged together;
New types of nodes to process (SUBEXPR)* and similar expressions
quickly, used if the SUBEXPR has no side effects and matches
strings of the same length;
Better optimizations by lookup for constant substrings;
Better search for constants substrings anchored by $ ;Changes in Perl code using RE engine:
More optimizations to s/longer/short/;
study() was not working;
/blah/ may be optimized to an analogue of index() if $& $` $' not seen;
Unneeded copying of matched-against string removed;
Only matched part of the string is copying if $` $' were not seen;Note that only the major bug fixes are listed here. See Changes for others.
Backtracking might not restore start of $3.
No feedback if max count for * or + on "complex" subexpression
was reached, similarly (but at compile time) for {3,34567}
Primitive restrictions on max count introduced to decrease a
possibility of a segfault;
(ZERO-LENGTH)* could segfault;
(ZERO-LENGTH)* was prohibited;
Long REs were not allowed;
/RE/g could skip matches at the same position after a
zero-length match;The following new syntax elements are supported:
(?<=RE)
(?<!RE)
(?{ CODE })
(?i-x)
(?i:RE)
(?(COND)YES_RE|NO_RE)
(?>RE)
\zSee "New qr// operator".
Better debugging output (possibly with colors),
even from non-debugging Perl;
RE engine code now looks like C, not like assembler;
Behaviour of RE modifiable by `use re' directive;
Improved documentation;
Test suite significantly extended;
Syntax [:^upper:] etc., reserved inside character classes;(?i) localized inside enclosing group;
$( is not interpolated into RE any more;
/RE/g may match at the same position (with non-zero length)
after a zero-length match (bug fix).See banner at the beginning of malloc.c for details.
Perl now contains its own highly optimized qsort() routine. The new qsort() is resistant to inconsistent comparison functions, so Perl's sort() will not provoke coredumps any more when given poorly written sort subroutines. (Some C library qsort()s that were being used before used to have this problem.) In our testing, the new qsort() required the minimal number of pair-wise compares on average, among all known qsort() implementations.
See perlfunc/sort.
Perl's signal handling is susceptible to random crashes, because signals arrive asynchronously, and the Perl runtime is not reentrant at arbitrary times.
However, one experimental implementation of reliable signals is available when threads are enabled. See Thread::Signal. Also see INSTALL for how to build a Perl capable of threads.
The internals now reallocate the perl stack only at predictable times. In particular, magic calls never trigger reallocations of the stack, because all reentrancy of the runtime is handled using a "stack of stacks". This should improve reliability of cached stack pointers in the internals and in XSUBs.
Perl used to complain if it encountered literal carriage returns in scripts. Now they are mostly treated like whitespace within program text. Inside string literals and here documents, literal carriage returns are ignored if they occur paired with linefeeds, or get interpreted as whitespace if they stand alone. This behavior means that literal carriage returns in files should be avoided. You can get the older, more compatible (but less generous) behavior by defining the preprocessor symbol PERL_STRICT_CR when building perl. Of course, all this has nothing whatever to do with how escapes like \r are handled within strings.
Note that this doesn't somehow magically allow you to keep all text files in DOS format. The generous treatment only applies to files that perl itself parses. If your C compiler doesn't allow carriage returns in files, you may still be unable to build modules that need a C compiler.
substr, pos and vec don't leak memory anymore when used in lvalue context. Many small leaks that impacted applications that embed multiple interpreters have been fixed.
The build-time option -DMULTIPLICITY has had many of the details reworked. Some previously global variables that should have been per-interpreter now are. With care, this allows interpreters to call each other. See the PerlInterp extension on CPAN.
See "Temporary Values via local()" in perlsub.
%! is transparently tied to the Errno moduleSee perlref.
EXPR foreach EXPR is supportedSee perlsyn.
See perlsub.
$^E is meaningful on Win32See perlvar.
foreach (1..1000000) optimizedforeach (1..1000000) is now optimized into a counting loop. It does not try to allocate a 1000000-size list anymore.
Foo:: can be used as implicitly quoted package nameBarewords caused unintuitive behavior when a subroutine with the same name as a package happened to be defined. Thus, new Foo @args, use the result of the call to Foo() instead of Foo being treated as a literal. The recommended way to write barewords in the indirect object slot is new Foo:: @args. Note that the method new() is called with a first argument of Foo, not Foo:: when you do that.
exists $Foo::{Bar::} tests existence of a packageIt was impossible to test for the existence of a package without actually creating it before. Now exists $Foo::{Bar::} can be used to test if the Foo::Bar namespace has been created.
See perllocale.
Perl5 has always had 64-bit support on systems with 64-bit longs. Starting with 5.005, the beginnings of experimental support for systems with 32-bit long and 64-bit 'long long' integers has been added. If you add -DUSE_LONG_LONG to your ccflags in config.sh (or manually define it in perl.h) then perl will be built with 'long long' support. There will be many compiler warnings, and the resultant perl may not work on all systems. There are many other issues related to third-party extensions and libraries. This option exists to allow people to work on those issues.
die() now accepts a reference value, and $@ gets set to that value in exception traps. This makes it possible to propagate exception objects. This is an undocumented experimental feature.
printf format conversions are handled internallySee "printf" in perlfunc.
INIT keywordINIT subs are like BEGIN and END, but they get run just before the perl runtime begins execution. e.g., the Perl Compiler makes use of INIT blocks to initialize and resolve pointers to XSUBs.
lock keywordThe lock keyword is the fundamental synchronization primitive in threaded perl. When threads are not enabled, it is currently a noop.
To minimize impact on source compatibility this keyword is "weak", i.e., any user-defined subroutine of the same name overrides it, unless a use Thread has been seen.
qr// operatorThe qr// operator, which is syntactically similar to the other quote-like operators, is used to create precompiled regular expressions. This compiled form can now be explicitly passed around in variables, and interpolated in other regular expressions. See perlop.
our is now a reserved wordCalling a subroutine with the name our will now provoke a warning when using the -w switch.
See Tie::Array.
Several missing hooks have been added. There is also a new base class for TIEARRAY implementations. See Tie::Array.
substr() can now both return and replace in one operation. The optional 4th argument is the replacement string. See "substr" in perlfunc.
splice() with a negative LENGTH argument now work similar to what the LENGTH did for substr(). Previously a negative LENGTH was treated as 0. See "splice" in perlfunc.
When you say something like substr($x, 5) = "hi", the scalar returned by substr() is special, in that any modifications to it affect $x. (This is called a 'magic lvalue' because an 'lvalue' is something on the left side of an assignment.) Normally, this is exactly what you would expect to happen, but Perl uses the same magic if you use substr(), pos(), or vec() in a context where they might be modified, like taking a reference with \ or as an argument to a sub that modifies @_. In previous versions, this 'magic' only went one way, but now changes to the scalar the magic refers to ($x in the above example) affect the magic lvalue too. For instance, this code now acts differently:
$x = "hello";
sub printit {
$x = "g'bye";
print $_[0], "\n";
}
printit(substr($x, 0, 5));In previous versions, this would print "hello", but it now prints "g'bye".
If $/ is a reference to an integer, or a scalar that holds an integer, <> will read in records instead of lines. For more info, see "$/" in perlvar.
Configure has many incremental improvements. Site-wide policy for building perl can now be made persistent, via Policy.sh. Configure also records the command-line arguments used in config.sh.
BeOS is now supported. See README.beos.
DOS is now supported under the DJGPP tools. See README.dos (installed as perldos on some systems).
MiNT is now supported. See README.mint.
MPE/iX is now supported. See README.mpeix.
MVS (aka OS390, aka Open Edition) is now supported. See README.os390 (installed as perlos390 on some systems).
Stratus VOS is now supported. See README.vos.
Win32 support has been vastly enhanced. Support for Perl Object, a C++ encapsulation of Perl. GCC and EGCS are now supported on Win32. See README.win32, aka perlwin32.
VMS configuration system has been rewritten. See README.vms (installed as README_vms on some systems).
The hints files for most Unix platforms have seen incremental improvements.
Perl compiler and tools. See B.
A module to pretty print Perl data. See Data::Dumper.
A module to dump perl values to the screen. See Dumpvalue.
A module to look up errors more conveniently. See Errno.
A portable API for file operations.
Query and manage installed modules.
Manipulate .packlist files.
Make functions/builtins succeed or die.
Constants and other support infrastructure for System V IPC operations in perl.
A framework for writing test suites.
Base class for tied arrays.
Base class for tied handles.
Perl thread creation, manipulation, and support.
Set subroutine attributes.
Compile-time class fields.
Various pragmata to control behavior of regular expressions.
You can now run tests for x seconds instead of guessing the right number of tests to run.
Keeps better time.
Carp has a new function cluck(). cluck() warns, like carp(), but also adds a stack backtrace to the error message, like confess().
CGI has been updated to version 2.42.
More Fcntl constants added: F_SETLK64, F_SETLKW64, O_LARGEFILE for large (more than 4G) file access (the 64-bit support is not yet working, though, so no need to get overly excited), Free/Net/OpenBSD locking behaviour flags F_FLOCK, F_POSIX, Linux F_SHLCK, and O_ACCMODE: the mask of O_RDONLY, O_WRONLY, and O_RDWR.
The accessors methods Re, Im, arg, abs, rho, theta, methods can ($z->Re()) now also act as mutators ($z->Re(3)).
A little bit of radial trigonometry (cylindrical and spherical) added, for example the great circle distance.
POSIX now has its own platform-specific hints files.
DB_File supports version 2.x of Berkeley DB. See ext/DB_File/Changes.
MakeMaker now supports writing empty makefiles, provides a way to specify that site umask() policy should be honored. There is also better support for manipulation of .packlist files, and getting information about installed modules.
Extensions that have both architecture-dependent and architecture-independent files are now always installed completely in the architecture-dependent locations. Previously, the shareable parts were shared both across architectures and across perl versions and were therefore liable to be overwritten with newer versions that might have subtle incompatibilities.
See perlmodinstall and CPAN.
Cwd::cwd is faster on most platforms.
h2ph and related utilities have been vastly overhauled.
perlcc, a new experimental front end for the compiler is available.
The crude GNU configure emulator is now called configure.gnu to avoid trampling on Configure under case-insensitive filesystems.
perldoc used to be rather slow. The slower features are now optional. In particular, case-insensitive searches need the -i switch, and recursive searches need -r. You can set these switches in the PERLDOC environment variable to get the old behavior.
Config.pm now has a glossary of variables.
Porting/patching.pod has detailed instructions on how to create and submit patches for perl.
perlport specifies guidelines on how to write portably.
perlmodinstall describes how to fetch and install modules from CPAN sites.
Some more Perl traps are documented now. See perltrap.
perlopentut gives a tutorial on using open().