| PERL58DELTA(1) | Perl Programmers Reference Guide | PERL58DELTA(1) |
perl58delta - what is new for perl v5.8.0
This document describes differences between the 5.6.0 release and the 5.8.0 release.
Many of the bug fixes in 5.8.0 were already seen in the 5.6.1 maintenance release since the two releases were kept closely coordinated (while 5.8.0 was still called 5.7.something).
Changes that were integrated into the 5.6.1 release are marked "[561]". Many of these changes have been further developed since 5.6.1 was released, those are marked "[561+]".
You can see the list of changes in the 5.6.1 release (both from the 5.005_03 release and the 5.6.0 release) by reading perl561delta.
Perl 5.8 is not binary compatible with earlier releases of Perl.
You have to recompile your XS modules.
(Pure Perl modules should continue to work.)
The major reason for the discontinuity is the new IO architecture called PerlIO. PerlIO is the default configuration because without it many new features of Perl 5.8 cannot be used. In other words: you just have to recompile your modules containing XS code, sorry about that.
In future releases of Perl, non-PerlIO aware XS modules may become completely unsupported. This shouldn't be too difficult for module authors, however: PerlIO has been designed as a drop-in replacement (at the source code level) for the stdio interface.
Depending on your platform, there are also other reasons why we decided to break binary compatibility, please read on.
If your pointers are 64 bits wide, the Perl malloc is no longer being used because it does not work well with 8-byte pointers. Also, usually the system mallocs on such platforms are much better optimized for such large memory models than the Perl malloc. Some memory-hungry Perl applications like the PDL don't work well with Perl's malloc. Finally, other applications than Perl (such as mod_perl) tend to prefer the system malloc. Such platforms include Alpha and 64-bit HPPA, MIPS, PPC, and Sparc.
The AIX dynaloading now uses in AIX releases 4.3 and newer the native dlopen interface of AIX instead of the old emulated interface. This change will probably break backward compatibility with compiled modules. The change was made to make Perl more compliant with other applications like mod_perl which are using the AIX native interface.
The "my EXPR : ATTRS" syntax now applies variable attributes at run-time. (Subroutine and "our" variables still get attributes applied at compile-time.) See attributes for additional details. In particular, however, this allows variable attributes to be useful for "tie" interfaces, which was a deficiency of earlier releases. Note that the new semantics doesn't work with the Attribute::Handlers module (as of version 0.76).
The Socket extension is now dynamically loaded instead of being statically built in. This may or may not be a problem with ancient TCP/IP stacks of VMS: we do not know since we weren't able to test Perl in such configurations.
Perl now uses IEEE format (T_FLOAT) as the default internal floating point format on OpenVMS Alpha, potentially breaking binary compatibility with external libraries or existing data. G_FLOAT is still available as a configuration option. The default on VAX (D_FLOAT) has not changed.
Previously in Perl 5.6 to use Unicode one would say "use utf8" and then the operations (like string concatenation) were Unicode-aware in that lexical scope.
This was found to be an inconvenient interface, and in Perl 5.8 the Unicode model has completely changed: now the "Unicodeness" is bound to the data itself, and for most of the time "use utf8" is not needed at all. The only remaining use of "use utf8" is when the Perl script itself has been written in the UTF-8 encoding of Unicode. (UTF-8 has not been made the default since there are many Perl scripts out there that are using various national eight-bit character sets, which would be illegal in UTF-8.)
See perluniintro for the explanation of the current model, and utf8 for the current use of the utf8 pragma.
Unicode scripts are now supported. Scripts are similar to (and superior to) Unicode blocks. The difference between scripts and blocks is that scripts are the glyphs used by a language or a group of languages, while the blocks are more artificial groupings of (mostly) 256 characters based on the Unicode numbering.
In general, scripts are more inclusive, but not universally so. For example, while the script "Latin" includes all the Latin characters and their various diacritic-adorned versions, it does not include the various punctuation or digits (since they are not solely "Latin").
A number of other properties are now supported, including "\p{L&}", "\p{Any}" "\p{Assigned}", "\p{Unassigned}", "\p{Blank}" [561] and "\p{SpacePerl}" [561] (along with their "\P{...}" versions, of course). See perlunicode for details, and more additions.
The "In" or "Is" prefix to names used with the "\p{...}" and "\P{...}" are now almost always optional. The only exception is that a "In" prefix is required to signify a Unicode block when a block name conflicts with a script name. For example, "\p{Tibetan}" refers to the script, while "\p{InTibetan}" refers to the block. When there is no name conflict, you can omit the "In" from the block name (e.g. "\p{BraillePatterns}"), but to be safe, it's probably best to always use the "In").
A reference to a reference now stringifies as "REF(0x81485ec)" instead of "SCALAR(0x81485ec)" in order to be more consistent with the return value of ref().
The undocumented pack/unpack template letters D/F have been recycled for better use: now they stand for long double (if supported by the platform) and NV (Perl internal floating point type). (They used to be aliases for d/f, but you never knew that.)
The list of filenames from glob() (or <...>) is now by default sorted alphabetically to be csh-compliant (which is what happened before in most Unix platforms). (bsd_glob() does still sort platform natively, ASCII or EBCDIC, unless GLOB_ALPHASORT is specified.) [561]
Unicode in general should be now much more usable than in Perl 5.6.0 (or even in 5.6.1). Unicode can be used in hash keys, Unicode in regular expressions should work now, Unicode in tr/// should work now, Unicode in I/O should work now. See perluniintro for introduction and perlunicode for details.
See "New Unicode Properties" earlier in this document for additional information on changes with Unicode properties.
open($fh,'>:crlf :utf8', $path) || ...
or on already opened handles via extended "binmode":
binmode($fh,':encoding(iso-8859-7)');
The built-in layers are: unix (low level read/write), stdio (as in previous Perls), perlio (re-implementation of stdio buffering in a portable manner), crlf (does CRLF <=> "\n" translation as on Win32, but available on any platform). A mmap layer may be available if platform supports it (mostly Unixes).
Layers to be applied by default may be specified via the 'open' pragma.
See "Installation and Configuration Improvements" for the effects of PerlIO on your architecture name.
open KID_PS, "-|", "ps", "aux" or die $!;
forks the ps(1) command (without spawning a shell, as there are more than three arguments to open()), and reads its standard output via the "KID_PS" filehandle. See perlipc.
open($fh,">:utf8","Uni.txt");
Note for EBCDIC users: the pseudo layer ":utf8" is erroneously named for you since it's not UTF-8 what you will be getting but instead UTF-EBCDIC. See perlunicode, utf8, and http://www.unicode.org/unicode/reports/tr16/ for more information. In future releases this naming may change. See perluniintro for more information about UTF-8.
Note that after this Perl really does assume that everything is UTF-8: for example if some input handle is not, Perl will probably very soon complain about the input data like this "Malformed UTF-8 ..." since any old eight-bit data is not legal UTF-8.
Note for code authors: if you want to enable your users to use UTF-8 as their default encoding but in your code still have eight-bit I/O streams (such as images or zip files), you need to explicitly open() or binmode() with ":bytes" (see "open" in perlfunc and "binmode" in perlfunc), or you can just use "binmode(FH)" (nice for pre-5.8.0 backward compatibility).
open($fh,'>', \$variable) || ...
open($fh,"+>", undef) || ...
That is a literal undef, not an undefined value.
The new interpreter threads ("ithreads" for short) implementation of multithreading, by Arthur Bergman, replaces the old "5.005 threads" implementation. In the ithreads model any data sharing between threads must be explicit, as opposed to the model where data sharing was implicit. See threads and threads::shared, and perlthrtut.
As a part of the ithreads implementation Perl will also use any necessary and detectable reentrant libc interfaces.
A restricted hash is restricted to a certain set of keys, no keys outside the set can be added. Also individual keys can be restricted so that the key cannot be deleted and the value cannot be changed. No new syntax is involved: the Hash::Util module is the interface.
Perl used to be fragile in that signals arriving at inopportune moments could corrupt Perl's internal state. Now Perl postpones handling of signals until it's safe (between opcodes).
This change may have surprising side effects because signals no longer interrupt Perl instantly. Perl will now first finish whatever it was doing, like finishing an internal operation (like sort()) or an external operation (like an I/O operation), and only then look at any arrived signals (and before starting the next operation). No more corrupt internal state since the current operation is always finished first, but the signal may take more time to get heard. Note that breaking out from potentially blocking operations should still work, though.
In general a lot of fixing has happened in the area of Perl's understanding of numbers, both integer and floating point. Since in many systems the standard number parsing functions like "strtoul()" and "atof()" seem to have bugs, Perl tries to work around their deficiencies. This results hopefully in more accurate numbers.
Perl now tries internally to use integer values in numeric conversions and basic arithmetics (+ - * /) if the arguments are integers, and tries also to keep the results stored internally as integers. This change leads to often slightly faster and always less lossy arithmetics. (Previously Perl always preferred floating point numbers in its math.)
In double-quoted strings, arrays now interpolate, no matter what. The behavior in earlier versions of perl 5 was that arrays would interpolate into strings if the array had been mentioned before the string was compiled, and otherwise Perl would raise a fatal compile-time error. In versions 5.000 through 5.003, the error was
Literal @example now requires backslash
In versions 5.004_01 through 5.6.0, the error was
In string, @example now must be written as \@example
The idea here was to get people into the habit of writing "fred\@example.com" when they wanted a literal "@" sign, just as they have always written "Give me back my \$5" when they wanted a literal "$" sign.
Starting with 5.6.1, when Perl now sees an "@" sign in a double-quoted string, it always attempts to interpolate an array, regardless of whether or not the array has been used or declared already. The fatal error has been downgraded to an optional warning:
Possible unintended interpolation of @example in string
This warns you that "fred@example.com" is going to turn into "fred.com" if you don't backslash the "@". See http://perl.plover.com/at-error.html for more details about the history here.
Also, $Config{byteorder} is now computed dynamically--this is more robust with "fat binaries" where an executable image contains binaries for more than one binary platform, and when cross-compiling.
printf "%2\$s %1\$s\n", "foo", "bar";
will print "bar foo\n". This feature helps in writing internationalised software, and in general when the order of the parameters can vary.
Use of "/c" in substitutions, even with "/g", elicits "Use of /c modifier is meaningless in s///".
Use of "/g" with "split" elicits "Use of /g modifier is meaningless in split".
See perlmod
package MyPack;
use Attribute::Handlers;
sub Wolf :ATTR(SCALAR) { print "howl!\n" }
# later, in some package using or inheriting from MyPack...
my MyPack $Fluffy : Wolf; # the attribute handler Wolf will be called
Both variables and routines can have attribute handlers. Handlers can be specific to type (SCALAR, ARRAY, HASH, or CODE), or specific to the exact compilation phase (BEGIN, CHECK, INIT, or END). See Attribute::Handlers.
use Digest::MD5 'md5_hex';
$digest = md5_hex("Thirsty Camel");
print $digest, "\n"; # 01d19d9d2045e005c3f1b80e8b164de1
NOTE: the "MD5" backward compatibility module is deliberately not included since its further use is discouraged.
See also PerlIO::via::QuotedPrint.
Any encoding supported by Encode module is also available to the ":encoding()" layer if PerlIO is used.
# in MyFilter.pm:
package MyFilter;
use Filter::Simple sub {
while (my ($from, $to) = splice @_, 0, 2) {
s/$from/$to/g;
}
};
1;
# in user's code:
use MyFilter qr/red/ => 'green';
print "red\n"; # this code is filtered, will print "green\n"
print "bored\n"; # this code is filtered, will print "bogreen\n"
no MyFilter;
print "red\n"; # this code is not filtered, will print "red\n"
Perl installation leaves libnet unconfigured; use libnetcfg to configure it.
use Locale::Country;
$country = code2country('jp'); # $country gets 'Japan'
$code = country2code('Norway'); # $code gets 'no'
See Locale::Constants, Locale::Country, Locale::Currency, and Locale::Language.
use MIME::Base64;
$encoded = encode_base64('Aladdin:open sesame');
$decoded = decode_base64($encoded);
print $encoded, "\n"; # "QWxhZGRpbjpvcGVuIHNlc2FtZQ=="
See MIME::Base64.
use MIME::QuotedPrint;
$encoded = encode_qp("\xDE\xAD\xBE\xEF");
$decoded = decode_qp($encoded);
print $encoded, "\n"; # "=DE=AD=BE=EF\n"
print $decoded, "\n"; # "\xDE\xAD\xBE\xEF\n"
See also PerlIO::via::QuotedPrint.
use PerlIO::via::QuotedPrint;
open($fh,">:via(QuotedPrint)",$path);
This will automatically convert everything output to $fh to Quoted-Printable. See PerlIO::via and PerlIO::via::QuotedPrint.
use Switch;
you have "switch" and "case" available in Perl.
use Switch;
switch ($val) {
case 1 { print "number 1" }
case "a" { print "string a" }
case [1..10,42] { print "number in list" }
case (@array) { print "number in list" }
case /\w+/ { print "pattern" }
case qr/\w+/ { print "pattern" }
case (%hash) { print "entry in hash" }
case (\%hash) { print "entry in hash" }
case (\&sub) { print "arg to subroutine" }
else { print "previous case not true" }
}
See Switch.
use Text::Balanced 'extract_delimited';
($a, $b) = extract_delimited("'never say never', he never said", "'", '');
$a will be "'never say never'", $b will be ', he never said'.
In addition to extract_delimited(), there are also extract_bracketed(), extract_quotelike(), extract_codeblock(), extract_variable(), extract_tagged(), extract_multiple(), gen_delimited_pat(), and gen_extract_tagged(). With these, you can implement rather advanced parsing algorithms. See Text::Balanced.
use English '-no_match_vars';
(Assuming, of course, that you don't need the troublesome variables "$`", $&, or "$'".) Also, introduced @LAST_MATCH_START and @LAST_MATCH_END English aliases for "@-" and "@+".
Note that some of the Net::Ping tests are disabled when running under the Perl distribution since one cannot assume one or more of the following: enabled echo port at localhost, full Internet connectivity, or sympathetic firewalls. You can set the environment variable PERL_TEST_Net_Ping to "1" (one) before running the Perl test suite to enable all the Net::Ping tests.
The following platform-specific documents are available before the installation as README.platform, and after the installation as perlplatform:
perlaix perlamiga perlapollo perlbeos perlbs2000
perlce perlcygwin perldgux perldos perlepoc perlfreebsd perlhpux
perlhurd perlirix perlmachten perlmacos perlmint perlmpeix
perlnetware perlos2 perlos390 perlplan9 perlqnx perlsolaris
perltru64 perluts perlvmesa perlvms perlvos perlwin32
These documents usually detail one or more of the following subjects: configuring, building, testing, installing, and sometimes also using Perl on the said platform.
Eastern Asian Perl users are now welcomed in their own languages: README.jp (Japanese), README.ko (Korean), README.cn (simplified Chinese) and README.tw (traditional Chinese), which are written in normal pod but encoded in EUC-JP, EUC-KR, EUC-CN and Big5. These will get installed as
perljp perlko perlcn perltw
The story in more detail: suppose you want to serve yourself a little slice of Pi.
@digits = ( 3,1,4,1,5,9 );
A numerical sort of the digits will yield (1,1,3,4,5,9), as expected. Which 1 comes first is hard to know, since one 1 looks pretty much like any other. You can regard this as totally trivial, or somewhat profound. However, if you just want to sort the even digits ahead of the odd ones, then what will
sort { ($a % 2) <=> ($b % 2) } @digits;
yield? The only even digit, 4, will come first. But how about the odd numbers, which all compare equal? With the quicksort algorithm used to implement Perl 5.6 and earlier, the order of ties is left up to the sort. So, as you add more and more digits of Pi, the order in which the sorted even and odd digits appear will change. and, for sufficiently large slices of Pi, the quicksort algorithm in Perl 5.8 won't return the same results even if reinvoked with the same input. The justification for this rests with quicksort's worst case behavior. If you run
sort { $a <=> $b } ( 1 .. $N , 1 .. $N );
(something you might approximate if you wanted to merge two sorted arrays using sort), doubling $N doesn't just double the quicksort time, it quadruples it. Quicksort has a worst case run time that can grow like N**2, so-called quadratic behaviour, and it can happen on patterns that may well arise in normal use. You won't notice this for small arrays, but you will notice it with larger arrays, and you may not live long enough for the sort to complete on arrays of a million elements. So the 5.8 quicksort scrambles large arrays before sorting them, as a statistical defence against quadratic behaviour. But that means if you sort the same large array twice, ties may be broken in different ways.
Because of the unpredictability of tie-breaking order, and the quadratic worst-case behaviour, quicksort was almost replaced completely with a stable mergesort. Stable means that ties are broken to preserve the original order of appearance in the input array. So
sort { ($a % 2) <=> ($b % 2) } (3,1,4,1,5,9);
will yield (4,3,1,1,5,9), guaranteed. The even and odd numbers appear in the output in the same order they appeared in the input. Mergesort has worst case O(N log N) behaviour, the best value attainable. And, ironically, this mergesort does particularly well where quicksort goes quadratic: mergesort sorts (1..$N, 1..$N) in O(N) time. But quicksort was rescued at the last moment because it is faster than mergesort on certain inputs and platforms. For example, if you really don't care about the order of even and odd digits, quicksort will run in O(N) time; it's very good at sorting many repetitions of a small number of distinct elements. The quicksort divide and conquer strategy works well on platforms with relatively small, very fast, caches. Eventually, the problem gets whittled down to one that fits in the cache, from which point it benefits from the increased memory speed.
Quicksort was rescued by implementing a sort pragma to control aspects of the sort. The stable subpragma forces stable behaviour, regardless of algorithm. The _quicksort and _mergesort subpragmas are heavy-handed ways to select the underlying implementation. The leading "_" is a reminder that these subpragmas may not survive beyond 5.8. More appropriate mechanisms for selecting the implementation exist, but they wouldn't have arrived in time to save quicksort.
mkdir perl/build/directory
cd perl/build/directory
sh /path/to/perl/source/Configure -Dmksymlinks ...
This will create in perl/build/directory a tree of symbolic links pointing to files in /path/to/perl/source. The original files are left unaffected. After Configure has finished, you can just say
make all test
and Perl will be built and tested, all in perl/build/directory. [561]
Note that the 5.005 threads are unsupported and deprecated: if you have code written for the old threads you should migrate it to the new ithreads model.
make LIBPERL=libperld.a
has been removed. Use -DDEBUGGING instead.
For the list of platforms known to support Perl, see "Supported Platforms" in perlport.
Numerous memory leaks and uninitialized memory accesses have been hunted down. Most importantly, anonymous subs used to leak quite a bit. [561]
See perldebug.
use Tie::Hash;
tie my %tied_hash => 'Tie::StdHash';
...
# Used to leak memory every time local() was called;
# in a loop, this added up.
local($tied_hash{Foo}) = 1;
use Tie::Hash;
tie my %tied_hash => 'Tie::StdHash';
...
# Nothing has set the FOO element so far
{ local $tied_hash{FOO} = 'Bar' }
# This used to print, but not now.
print "exists!\n" if exists $tied_hash{FOO};
As a side effect of this fix, tied hash interfaces must define the EXISTS and DELETE methods.
Perl now works on post-4.0 BSD/OSes.
Setting $0 now works (as much as possible; see perlvar for details).
Numerous updates; currently synchronised with Cygwin 1.3.10.
EPOC now better supported. See README.epoc. [561]
Perl now works on post-3.0 FreeBSDs.
README.hpux updated; "Configure -Duse64bitall" now works; now uses HP-UX malloc instead of Perl malloc.
Numerous compilation flag and hint enhancements; accidental mixing of 32-bit and 64-bit libraries (a doomed attempt) made much harder.
Compilation of the standard Perl distribution in Mac OS Classic should now work if you have the Metrowerks development environment and the missing Mac-specific toolkit bits. Contact the macperl mailing list for details.
MPE/iX update after Perl 5.6.0. See README.mpeix. [561]
Perl now works on NetBSD/sparc.
Now works with usethreads (see INSTALL). [561]
64-bitness using the Sun Workshop compiler now works.
The native build method requires at least VOS Release 14.5.0 and GNU C++/GNU Tools 2.0.1 or later. The Perl pack function now maps overflowed values to +infinity and underflowed values to -infinity.
The operating system version letter now recorded in $Config{osvers}. Allow compiling with gcc (previously explicitly forbidden). Compiling with gcc still not recommended because buggy code results, even with gcc 2.95.2.
Fixed various alignment problems that lead into core dumps either during build or later; no longer dies on math errors at runtime; now using full quad integers (64 bits), previously was using only 46 bit integers for speed.
See "Socket Extension Dynamic in VMS" and "IEEE-format Floating Point Default on OpenVMS Alpha" for important changes not otherwise listed here.
chdir() now works better despite a CRT bug; now works with MULTIPLICITY (see INSTALL); now works with Perl's malloc.
The tainting of %ENV elements via "keys" or "values" was previously unimplemented. It now works as documented.
The "waitpid" emulation has been improved. The worst bug (now fixed) was that a pid of -1 would cause a wildcard search of all processes on the system.
POSIX-style signals are now emulated much better on VMS versions prior to 7.0.
The "system" function and backticks operator have improved functionality and better error handling. [561]
File access tests now use current process privileges rather than the user's default privileges, which could sometimes result in a mismatch between reported access and actual access. This improvement is only available on VMS v6.0 and later.
There is a new "kill" implementation based on "sys$sigprc" that allows older VMS systems (pre-7.0) to use "kill" to send signals rather than simply force exit. This implementation also allows later systems to call "kill" from within a signal handler.
Iterative logical name translations are now limited to 10 iterations in imitation of SHOW LOGICAL and other OpenVMS facilities.
ExtUtils::Embed [561]
IO::Pipe
IO::Poll
Net::Ping
Note that this means that some scripts that may have relied on earlier buggy behavior may no longer work correctly. For example, "system("nmake /nologo", @args)" will now attempt to run the file "nmake /nologo" and will fail when such a file isn't found. On the other hand, perl will now execute code such as "system("c:/Program Files/MyApp/foo.exe", @args)" correctly.
Please see perldiag for more details.
(This change was already made in 5.7.0 but bears repeating here.) (5.7.0 came out before 5.6.1: the development branch 5.7 released earlier than the maintenance branch 5.6)
A potential security vulnerability in the optional suidperl component of Perl was identified in August 2000. suidperl is neither built nor installed by default. As of November 2001 the only known vulnerable platform is Linux, most likely all Linux distributions. CERT and various vendors and distributors have been alerted about the vulnerability. See http://www.cpan.org/src/5.0/sperl-2000-08-05/sperl-2000-08-05.txt for more information.
The problem was caused by Perl trying to report a suspected security exploit attempt using an external program, /bin/mail. On Linux platforms the /bin/mail program had an undocumented feature which when combined with suidperl gave access to a root shell, resulting in a serious compromise instead of reporting the exploit attempt. If you don't have /bin/mail, or if you have 'safe setuid scripts', or if suidperl is not installed, you are safe.
The exploit attempt reporting feature has been completely removed from Perl 5.8.0 (and the maintenance release 5.6.1, and it was removed also from all the Perl 5.7 releases), so that particular vulnerability isn't there anymore. However, further security vulnerabilities are, unfortunately, always possible. The suidperl functionality is most probably going to be removed in Perl 5.10. In any case, suidperl should only be used by security experts who know exactly what they are doing and why they are using suidperl instead of some other solution such as sudo ( see http://www.courtesan.com/sudo/ ).
Several new tests have been added, especially for the lib and ext subsections. There are now about 69 000 individual tests (spread over about 700 test scripts), in the regression suite (5.6.1 has about 11 700 tests, in 258 test scripts) The exact numbers depend on the platform and Perl configuration used. Many of the new tests are of course introduced by the new modules, but still in general Perl is now more thoroughly tested.
Because of the large number of tests, running the regression suite will take considerably longer time than it used to: expect the suite to take up to 4-5 times longer to run than in perl 5.6. On a really fast machine you can hope to finish the suite in about 6-8 minutes (wallclock time).
The tests are now reported in a different order than in earlier Perls. (This happens because the test scripts from under t/lib have been moved to be closer to the library/extension they are testing.)
The compiler suite is slowly getting better but it continues to be highly experimental. Use in production environments is discouraged.
local %tied_array;
doesn't work as one would expect: the old value is restored incorrectly. This will be changed in a future release, but we don't know yet what the new semantics will exactly be. In any case, the change will break existing code that relies on the current (ill-defined) semantics, so just avoid doing this in general.
Some extensions like mod_perl are known to have issues with `largefiles', a change brought by Perl 5.6.0 in which file offsets default to 64 bits wide, where supported. Modules may fail to compile at all, or they may compile and work incorrectly. Currently, there is no good solution for the problem, but Configure now provides appropriate non-largefile ccflags, ldflags, libswanted, and libs in the %Config hash (e.g., $Config{ccflags_nolargefiles}) so the extensions that are having problems can try configuring themselves without the largefileness. This is admittedly not a clean solution, and the solution may not even work at all. One potential failure is whether one can (or, if one can, whether it's a good idea to) link together at all binaries with different ideas about file offsets; all this is platform-dependent.
for (1..5) { $_++ }
works without complaint. It shouldn't. (You should be able to modify only lvalue elements inside the loops.) You can see the correct behaviour by replacing the 1..5 with 1, 2, 3, 4, 5.
Use mod_perl 1.27 or higher.
Don't panic. Read the 'make test' section of INSTALL instead.
Use libwww-perl 5.65 or later.
Use PDL 2.3.4 or later.
You may get errors like 'Undefined symbol "Perl_get_sv"' or "can't resolve symbol 'Perl_get_sv'", or the symbol may be "Perl_sv_2pv". This probably means that you are trying to use an older shared Perl library (or extensions linked with such) with Perl 5.8.0 executable. Perl used to have such a subroutine, but that is no more the case. Check your shared library path, and any shared Perl libraries in those directories.
Sometimes this problem may also indicate a partial Perl 5.8.0 installation, see "Mac OS X dyld undefined symbols" for an example and how to deal with it.
Self-tying of arrays and hashes is broken in rather deep and hard-to-fix ways. As a stop-gap measure to avoid people from getting frustrated at the mysterious results (core dumps, most often), it is forbidden for now (you will get a fatal error even from an attempt).
A change to self-tying of globs has caused them to be recursively referenced (see: "Two-Phased Garbage Collection" in perlobj). You will now need an explicit untie to destroy a self-tied glob. This behaviour may be fixed at a later date.
Self-tying of scalars and IO thingies works.
If this test fails, it indicates that your libc (C library) is not threadsafe. This particular test stress tests the localtime() call to find out whether it is threadsafe. See perlthrtut for more information.
Note that support for 5.005-style threading is deprecated, experimental and practically unsupported. In 5.10, it is expected to be removed. You should migrate your code to ithreads.
The following tests are known to fail due to fundamental problems in the 5.005 threading implementation. These are not new failures--Perl 5.005_0x has the same bugs, but didn't have these tests.
../ext/B/t/xref.t 255 65280 14 12 85.71% 3-14
../ext/List/Util/t/first.t 255 65280 7 4 57.14% 2 5-7
../lib/English.t 2 512 54 2 3.70% 2-3
../lib/FileCache.t 5 1 20.00% 5
../lib/Filter/Simple/t/data.t 6 3 50.00% 1-3
../lib/Filter/Simple/t/filter_only. 9 3 33.33% 1-2 5
../lib/Math/BigInt/t/bare_mbf.t 1627 4 0.25% 8 11 1626-1627
../lib/Math/BigInt/t/bigfltpm.t 1629 4 0.25% 10 13 1628-
1629
../lib/Math/BigInt/t/sub_mbf.t 1633 4 0.24% 8 11 1632-1633
../lib/Math/BigInt/t/with_sub.t 1628 4 0.25% 9 12 1627-1628
../lib/Tie/File/t/31_autodefer.t 255 65280 65 32 49.23% 34-65
../lib/autouse.t 10 1 10.00% 4
op/flip.t 15 1 6.67% 15
These failures are unlikely to get fixed as 5.005-style threads are considered fundamentally broken. (Basically what happens is that competing threads can corrupt shared global state, one good example being regular expression engine's state.)
The following tests may fail intermittently because of timing problems, for example if the system is heavily loaded.
t/op/alarm.t
ext/Time/HiRes/HiRes.t
lib/Benchmark.t
lib/Memoize/t/expmod_t.t
lib/Memoize/t/speed.t
In case of failure please try running them manually, for example
./perl -Ilib ext/Time/HiRes/HiRes.t
For normal arrays "$foo = \$bar[1]" will assign "undef" to $bar[1] (assuming that it didn't exist before), but for tied/magical arrays and hashes such autovivification does not happen because there is currently no way to catch the reference creation. The same problem affects slicing over non-existent indices/keys of a tied/magical array/hash.
One can have Unicode in identifier names, but not in package/class or subroutine names. While some limited functionality towards this does exist as of Perl 5.8.0, that is more accidental than designed; use of Unicode for the said purposes is unsupported.
One reason of this unfinishedness is its (currently) inherent unportability: since both package names and subroutine names may need to be mapped to file and directory names, the Unicode capability of the filesystem becomes important-- and there unfortunately aren't portable answers.
The AIX C compiler vac version 5.0.0.0 may produce buggy code, resulting in a few random tests failing when run as part of "make test", but when the failing tests are run by hand, they succeed. We suggest upgrading to at least vac version 5.0.1.0, that has been known to compile Perl correctly. "lslpp -L|grep vac.C" will tell you the vac version. See README.aix.
"pp_sys.c", line 4651.39: 1506-280 (W) Function argument assignment between types "unsigned char*" and "const void*" is not allowed.
This is harmless; it is caused by the getnetbyaddr() and getnetbyaddr_r() having slightly different types for their first argument.
If you see op/pack, op/pat, op/regexp, or ext/Storable tests failing in a Linux/alpha or *BSD/Alpha, it's probably time to upgrade your gcc. gccs prior to 2.95.3 are definitely not good enough, and gcc 3.1 may be even better. (RedHat Linux/alpha with gcc 3.1 reported no problems, as did Linux 2.4.18 with gcc 2.95.4.) (In Tru64, it is preferable to use the bundled C compiler.)
Perl 5.8.0 doesn't build in AmigaOS. It broke at some point during the ithreads work and we could not find Amiga experts to unbreak the problems. Perl 5.6.1 still works for AmigaOS (as does the 5.7.2 development release).
The following tests fail on 5.8.0 Perl in BeOS Personal 5.03:
t/op/lfs............................FAILED at test 17 t/op/magic..........................FAILED at test 24 ext/Fcntl/t/syslfs..................FAILED at test 17 ext/File/Glob/t/basic...............FAILED at test 3 ext/POSIX/t/sigaction...............FAILED at test 13 ext/POSIX/t/waitpid.................FAILED at test 1
(Note: more information was available in README.beos until support for BeOS was removed in Perl v5.18.0)
For example when building the Tk extension for Cygwin, you may get an error message saying "unable to remap". This is known problem with Cygwin, and a workaround is detailed in here: http://sources.redhat.com/ml/cygwin/2001-12/msg00894.html
One can build but not install (or test the build of) the NDBM_File on FAT filesystems. Installation (or build) on NTFS works fine. If one attempts the test on a FAT install (or build) the following failures are expected:
../ext/NDBM_File/ndbm.t 13 3328 71 59 83.10% 1-2 4 16-71 ../ext/ODBM_File/odbm.t 255 65280 ?? ?? % ?? ../lib/AnyDBM_File.t 2 512 12 2 16.67% 1 4 ../lib/Memoize/t/errors.t 0 139 11 5 45.45% 7-11 ../lib/Memoize/t/tie_ndbm.t 13 3328 4 4 100.00% 1-4 run/fresh_perl.t 97 1 1.03% 91
NDBM_File fails and ODBM_File just coredumps.
If you intend to run only on FAT (or if using AnyDBM_File on FAT), run Configure with the -Ui_ndbm and -Ui_dbm options to prevent NDBM_File and ODBM_File being built.
t/op/stat............................FAILED at test 29 lib/File/Find/t/find.................FAILED at test 1 lib/File/Find/t/taint................FAILED at test 1 lib/h2xs.............................FAILED at test 15 lib/Pod/t/eol........................FAILED at test 1 lib/Test/Harness/t/strap-analyze.....FAILED at test 8 lib/Test/Harness/t/test-harness......FAILED at test 23 lib/Test/Simple/t/exit...............FAILED at test 1
The above failures are known as of 5.8.0 with native builds with long filenames, but there are a few more if running under dosemu because of limitations (and maybe bugs) of dosemu:
t/comp/cpp...........................FAILED at test 3 t/op/inccode.........................(crash)
and a few lib/ExtUtils tests, and several hundred Encode/t/Aliases.t failures that work fine with long filenames. So you really might prefer native builds and long filenames.
This is a known bug in FreeBSD 4.5's readdir_r(), it has been fixed in FreeBSD 4.6 (see perlfreebsd (README.freebsd)).
The ISO 8859-15 locales may fail the locale test 117 in FreeBSD. This is caused by the characters \xFF (y with diaeresis) and \xBE (Y with diaeresis) not behaving correctly when being matched case-insensitively. Apparently this problem has been fixed in the latest FreeBSD releases. ( http://www.freebsd.org/cgi/query-pr.cgi?pr=34308 )
IRIX with MIPSpro 7.3.1.2m or 7.3.1.3m compiler may fail the List::Util test ext/List/Util/t/shuffle.t by dumping core. This seems to be a compiler error since if compiled with gcc no core dump ensues, and no failures have been seen on the said test on any other platform.
Similarly, building the Digest::MD5 extension has been known to fail with "*** Termination code 139 (bu21)".
The cure is to drop optimization level (Configure -Doptimize=-O2).
If perl is configured with -Duse64bitall, the successful result of the subtest 10 of lib/posix may arrive before the successful result of the subtest 9, which confuses the test harness so much that it thinks the subtest 9 failed.
This is a known bug in the glibc 2.2.5 with long long integers. ( http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=65612 )
No known fix.
Please remember to set your environment variable LC_ALL to "C" (setenv LC_ALL C) before running "make test" to avoid a lot of warnings about the broken locales of Mac OS X.
The following tests are known to fail in Mac OS X 10.1.5 because of buggy (old) implementations of Berkeley DB included in Mac OS X:
Failed Test Stat Wstat Total Fail Failed List of Failed ------------------------------------------------------------------------- ../ext/DB_File/t/db-btree.t 0 11 ?? ?? % ?? ../ext/DB_File/t/db-recno.t 149 3 2.01% 61 63 65
If you are building on a UFS partition, you will also probably see t/op/stat.t subtest #9 fail. This is caused by Darwin's UFS not supporting inode change time.
Also the ext/POSIX/t/posix.t subtest #10 fails but it is skipped for now because the failure is Apple's fault, not Perl's (blocked signals are lost).
If you Configure with ithreads, ext/threads/t/libc.t will fail. Again, this is not Perl's fault-- the libc of Mac OS X is not threadsafe (in this particular test, the localtime() call is found to be threadunsafe.)
If after installing Perl 5.8.0 you are getting warnings about missing symbols, for example
dyld: perl Undefined symbols
_perl_sv_2pv
_perl_get_sv
you probably have an old pre-Perl-5.8.0 installation (or parts of one) in /Library/Perl (the undefined symbols used to exist in pre-5.8.0 Perls). It seems that for some reason "make install" doesn't always completely overwrite the files in /Library/Perl. You can move the old Perl shared library out of the way like this:
cd /Library/Perl/darwin/CORE
mv libperl.dylib libperlold.dylib
and then reissue "make install". Note that the above of course is extremely disruptive for anything using the /usr/local/bin/perl. If that doesn't help, you may have to try removing all the .bundle files from beneath /Library/Perl, and again "make install"-ing.
The following tests are known to fail on OS/2 (for clarity only the failures are shown, not the full error messages):
../lib/ExtUtils/t/Mkbootstrap.t 1 256 18 1 5.56% 8 ../lib/ExtUtils/t/Packlist.t 1 256 34 1 2.94% 17 ../lib/ExtUtils/t/basic.t 1 256 17 1 5.88% 14 lib/os2_process.t 2 512 227 2 0.88% 174 209 lib/os2_process_kid.t 227 2 0.88% 174 209 lib/rx_cmprt.t 255 65280 18 3 16.67% 16-18
The op/sprintf tests 91, 129, and 130 are known to fail on some platforms. Examples include any platform using sfio, and Compaq/Tandem's NonStop-UX.
Test 91 is known to fail on QNX6 (nto), because "sprintf '%e',0" incorrectly produces 0.000000e+0 instead of 0.000000e+00.
For tests 129 and 130, the failing platforms do not comply with the ANSI C Standard: lines 19ff on page 134 of ANSI X3.159 1989, to be exact. (They produce something other than "1" and "-1" when formatting 0.6 and -0.6 using the printf format "%.0f"; most often, they produce "0" and "-0".)
The socketpair tests are known to be unhappy in SCO 3.2v5.0.4:
ext/Socket/socketpair.t...............FAILED tests 15-45
In case you are still using Solaris 2.5 (aka SunOS 5.5), you may experience failures (the test core dumping) in lib/locale.t. The suggested cure is to upgrade your Solaris.
The following tests are known to fail in Solaris x86 with Perl configured to use 64 bit integers:
ext/Data/Dumper/t/dumper.............FAILED at test 268 ext/Devel/Peek/Peek..................FAILED at test 7
The following tests are known to fail on SUPER-UX:
op/64bitint...........................FAILED tests 29-30, 32-33, 35-36 op/arith..............................FAILED tests 128-130 op/pack...............................FAILED tests 25-5625 op/pow................................ op/taint..............................# msgsnd failed ../ext/IO/lib/IO/t/io_poll............FAILED tests 3-4 ../ext/IPC/SysV/ipcsysv...............FAILED tests 2, 5-6 ../ext/IPC/SysV/t/msg.................FAILED tests 2, 4-6 ../ext/Socket/socketpair..............FAILED tests 12 ../lib/IPC/SysV.......................FAILED tests 2, 5-6 ../lib/warnings.......................FAILED tests 115-116, 118-119
The op/pack failure ("Cannot compress negative numbers at op/pack.t line 126") is serious but as of yet unsolved. It points at some problems with the signedness handling of the C compiler, as do the 64bitint, arith, and pow failures. Most of the rest point at problems with SysV IPC.
Use Term::ReadKey 2.20 or later.
Guessing which symbols your C compiler and preprocessor define...
will probably fail with error messages like
CC-20 cc: ERROR File = try.c, Line = 3
The identifier "bad" is undefined.
bad switch yylook 79bad switch yylook 79bad switch yylook 79bad switch yylook 79#ifdef A29K
^
CC-65 cc: ERROR File = try.c, Line = 3
A semicolon is expected at this point.
This is caused by a bug in the awk utility of UNICOS/mk. You can ignore the error, but it does cause a slight problem: you cannot fully benefit from the h2ph utility (see h2ph) that can be used to convert C headers to Perl libraries, mainly used to be able to access from Perl the constants defined using C preprocessor, cpp. Because of the above error, parts of the converted headers will be invisible. Luckily, these days the need for h2ph is rare.
There are a few known test failures. (Note: the relevant information was available in README.uts until support for UTS was removed in Perl v5.18.0)
When Perl is built using the native build process on VOS Release 14.5.0 and GNU C++/GNU Tools 2.0.1, all attempted tests either pass or result in TODO (ignored) failures.
There should be no reported test failures with a default configuration, though there are a number of tests marked TODO that point to areas needing further debugging and/or porting work.
In multi-CPU boxes, there are some problems with the I/O buffering: some output may appear twice.
Use XML::Parser 2.31 or later.
z/OS has rather many test failures but the situation is actually much better than it was in 5.6.0; it's just that so many new modules and tests have been added.
Failed Test Stat Wstat Total Fail Failed List of Failed
---------------------------------------------------------------------------
../ext/Data/Dumper/t/dumper.t 357 8 2.24% 311 314 325 327
331 333 337 339
../ext/IO/lib/IO/t/io_unix.t 5 4 80.00% 2-5
../ext/Storable/t/downgrade.t 12 3072 169 12 7.10% 14-15 46-47 78-79
110-111 150 161
../lib/ExtUtils/t/Constant.t 121 30976 48 48 100.00% 1-48
../lib/ExtUtils/t/Embed.t 9 9 100.00% 1-9
op/pat.t 922 7 0.76% 665 776 785 832-
834 845
op/sprintf.t 224 3 1.34% 98 100 136
op/tr.t 97 5 5.15% 63 71-74
uni/fold.t 780 6 0.77% 61 169 196 661
710-711
The failures in dumper.t and downgrade.t are problems in the tests, those in io_unix and sprintf are problems in the USS (UDP sockets and printf formats). The pat, tr, and fold failures are genuine Perl problems caused by EBCDIC (and in the pat and fold cases, combining that with Unicode). The Constant and Embed are probably problems in the tests (since they test Perl's ability to build extensions, and that seems to be working reasonably well.)
Though mostly working, Unicode support still has problem spots on EBCDIC platforms. One such known spot are the "\p{}" and "\P{}" regular expression constructs for code points less than 256: the "pP" are testing for Unicode code points, not knowing about EBCDIC.
"Time::Piece" (previously known as "Time::Object") was removed because it was felt that it didn't have enough value in it to be a core module. It is still a useful module, though, and is available from the CPAN.
Perl 5.8 unfortunately does not build anymore on AmigaOS; this broke accidentally at some point. Since there are not that many Amiga developers available, we could not get this fixed and tested in time for 5.8.0. Perl 5.6.1 still works for AmigaOS (as does the 5.7.2 development release).
The "PerlIO::Scalar" and "PerlIO::Via" (capitalised) were renamed as "PerlIO::scalar" and "PerlIO::via" (all lowercase) just before 5.8.0. The main rationale was to have all core PerlIO layers to have all lowercase names. The "plugins" are named as usual, for example "PerlIO::via::QuotedPrint".
The "threads::shared::queue" and "threads::shared::semaphore" were renamed as "Thread::Queue" and "Thread::Semaphore" just before 5.8.0. The main rationale was to have thread modules to obey normal naming, "Thread::" (the "threads" and "threads::shared" themselves are more pragma-like, they affect compile-time, so they stay lowercase).
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.
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.
Written by Jarkko Hietaniemi <jhi@iki.fi>.
| 2023-11-25 | perl v5.32.1 |