NAME

perl5380delta - what is new for perl v5.38.0

DESCRIPTION

This document describes differences between the 5.36.0 release and the 5.38.0 release.

Core Enhancements

New class Feature

A new experimental syntax is now available for defining object classes, where per-instance data is stored in "field" variables that behave like lexicals.

use feature 'class';
class Point
{
field $x;
field $y;
method zero { $x = $y = 0; }
}

This is described in more detail in perlclass. Notes on the internals of its implementation and other related details can be found in perlclassguts.

This remains a new and experimental feature, and is very much still under development. It will be the subject of much further addition, refinement and alteration in future releases. As it is experimental, it yields warnings in the experimental::class category. These can be silenced by a no warnings statement.

use feature 'class';
no warnings 'experimental::class';

Unicode 15.0 is supported

See https://www.unicode.org/versions/Unicode15.0.0/ for details.

Deprecation warnings now have specific subcategories

All deprecation warnings now have their own specific deprecation category which can be disabled individually. You can see a list of all deprecated features in perldeprecation, and in warnings. The following list is from warnings:

+- deprecated ----+
| |
| +- deprecated::apostrophe_as_package_separator
| |
| +- deprecated::delimiter_will_be_paired
| |
| +- deprecated::dot_in_inc
| |
| +- deprecated::goto_construct
| |
| +- deprecated::smartmatch
| |
| +- deprecated::unicode_property_name
| |
| +- deprecated::version_downgrade

It is still possible to disable all deprecation warnings in a single statement with

no warnings 'deprecated';

but now is possible to have a finer grained control. As has historically been the case these warnings are automatically enabled with

%{^HOOK} API introduced

For various reasons it can be difficult to create subroutine wrappers for some of perl's keywords. Any keyword which has an undefined prototype simply cannot be wrapped with a subroutine, and some keywords which perl permits to be wrapped are in practice very tricky to wrap. For example require is tricky to wrap, it is possible but doing so changes the stack depth, and the standard methods of exporting assume that they will be exporting to a package at certain stack depth up the stack, and the wrapper will thus change where functions are exported to unless implemented with a great deal of care. This can be very awkward to deal with.

Accordingly we have introduced a new hash called %{^HOOK} which is intended to facilitate such cases. When a keyword supports any kind of special hook then the hook will live in this new hash. Hooks in this hash will be named after the function they are called by, followed by two underbars and then the phase they are executed in, currently either before or after the keyword is executed.

In this initial release we support two hooks require__before and require__after. These are provided to make it easier to perform tasks before and after a require statement.

See perlvar for more details.

PERL_RAND_SEED

Added a new environment variable PERL_RAND_SEED which can be used to cause a perl program which uses rand without using srand() explicitly or which uses srand() with no arguments to be repeatable. See perlrun. This feature can be disabled at compile time by passing

-Accflags=-DNO_PERL_RAND_SEED

to Configure during the build process.

Defined-or and logical-or assignment default expressions in signatures

The default expression for a subroutine signature parameter can now be assigned using the //= or ||= operators, to apply the defaults whenever the caller provided an undefined or false value (respectively), rather than simply when the parameter is missing entirely. For more detail see the documentation in perlsub.

@INC Hook Enhancements and $INC and INCDIR

The internals for @INC hooks have been hardened to handle various edge cases and should no longer segfault or throw assert failures when hooks modify @INC during a require operation. As part of this we now ensure that any given hook is executed at most once during a require call, and that any duplicate directories do not trigger additional directory probes.

To provide developers more control over dynamic module lookup, a new hook method INCDIR is now supported. An object supporting this method may be injected into the @INC array, and when it is encountered in the module search process it will be executed, just like how INC hooks are executed, and its return value used as a list of directories to search for the module. Returning an empty list acts as a no-op. Note that since any references returned by this hook will be stringified and used as strings, you may not return a hook to be executed later via this API.

When an @INC hook (either INC or INCDIR) is called during require, the $INC variable will be localized to be the value of the index of @INC that the hook came from. If the hook wishes to override what the "next" index in @INC should be it may update $INC to be one less than the desired index (undef is equivalent to -1). This allows an @INC hook to completely rewrite the @INC array and have perl restart its directory probes from the beginning of @INC.

Blessed CODE references in @INC that do not support the INC or INCDIR methods will no longer trigger an exception, and instead will be treated the same as unblessed coderefs are, and executed as though they were an INC hook.

Forbidden control flow out of defer or finally now detected at compile-time

It is forbidden to attempt to leave a defer or finally block by means of control flow such as return or goto. Previous versions of perl could only detect this when actually attempted at runtime.

This version of perl adds compile-time detection for many cases that can be statically determined. This may mean that code which compiled successfully on a previous version of perl is now reported as a compile-time error with this one. This only happens in cases where it would have been an error to actually execute the code anyway; the error simply happens at an earlier time.

Optimistic Eval in Patterns

The use of (?{ ... }) and (??{ ... }) in a pattern disables various optimisations globally in that pattern. This may or may not be desired by the programmer. This release adds the (*{ ... }) equivalent. The only difference is that it does not and will never disable any optimisations in the regex engine. This may make it more unstable in the sense that it may be called more or less times in the future, however the number of times it executes will truly match how the regex engine functions. For example, certain types of optimisation are disabled when (?{ ... }) is included in a pattern, so that patterns which are O(N) in normal use become O(N*N) with a (?{ ... }) pattern in them. Switching to (*{ ... }) means the pattern will stay O(N).

REG_INF has been raised from 65,536 to 2,147,483,647

Many regex quantifiers used to be limited to U16_MAX in the past, but are now limited to I32_MAX, thus it is now possible to write /(?:word){1000000}/ for example. Note that doing so may cause the regex engine to run longer and use more memory.

New API functions optimize_optree and finalize_optree

There are two new API functions for operating on optree fragments, ensuring you can invoke the required parts of the optree-generation process that might otherwise not get invoked (e.g. when creating a custom LOGOP). To get access to these functions, you first need to set a #define to opt-in to using these functions.

#define PERL_USE_VOLATILE_API

These functions are closely tied to the internals of how the interpreter works, and could be altered or removed at any time if other internal changes make that necessary.

Some gotos are now permitted in defer and finally blocks

Perl version 5.36.0 added defer blocks and permitted the finally keyword to also add similar behaviour to try/catch syntax. These did not permit any goto expression within the body, as it could have caused control flow to jump out of the block. Now, some goto expressions are allowed, if they have a constant target label, and that label is found within the block.

use feature 'defer';
defer {
goto LABEL;
print "This does not execute\n";
LABEL: print "This does\n";
}

New regexp variable ${^LAST_SUCCESSFUL_PATTERN}

This allows access to the last succesful pattern that matched in the current scope. Many aspects of the regex engine refer to the "last successful pattern". The empty pattern reuses it, and all of the magic regex vars relate to it. This allows access to its pattern. The following code

if (m/foo/ || m/bar/) {
s//PQR/;
}

can be rewritten as follows

if (m/foo/ || m/bar/) {
s/${^LAST_SUCCESSFUL_PATTERN}/PQR/;
}

and it will do the exactly same thing.

Locale category LC_NAME now supported on participating platforms

On platforms that have the GNU extension LC_NAME category, you may now use it as the category parameter to "setlocale" in POSIX to set and query its locale.

Incompatible Changes

readline() no longer clears the stream error and eof flags

readline(), also spelled <>, would clear the handle's error and eof flags after an error occurred on the stream.

In nearly all cases this clear is no longer done, so the error and eof flags now properly reflect the status of the stream after readline().

Since the error flag is no longer cleared calling close() on the stream may fail and if the stream was not explicitly closed, the implicit close of the stream may produce a warning.

This has resulted in two main types of problems in downstream CPAN modules, and these may also occur in your code:

  • If your code reads to end of file, and then rebinds the handle to a new file descriptor, previously since the eof flag wasn't set you could continue to read from the stream. You now need to clear the eof flag yourself with $handle->clearerr() to continue reading.

  • If your code encounters an error on the stream while reading with readline() you will need to call $handle->clearerr to continue reading. The one case this occurred the underlying file descriptor was marked non-blocking, so the read() system call was failing with EAGAIN, which resulted in the error flag being set on the stream.

The only case where error and eof flags continue to cleared on error is when reading from the child process for glob() in miniperl. This allows it to correctly report errors from the child process on close(). This is unlikely to be an issue during normal perl development.

[GH #20060]

INIT blocks no longer run after an exit() in BEGIN

INIT blocks will no longer run after an exit() performed inside of a BEGIN. This means that the combination of the -v option and the -c option no longer executes a compile check as well as showing the perl version. The -v option executes an exit(0) after printing the version information inside of a BEGIN block, and the -c check is implemented by using INIT hooks, resulting in the -v option taking precedence.

[GH #1537] [GH #20181]

Syntax errors no longer produce "phantom error messages"

Generally perl will continue parsing the source code even after encountering a compile error. In many cases this is helpful, for instance with misspelled variable names it is helpful to show as many examples of the error as possible. But in the case of syntax errors continuing often produces bizarre error messages and may even cause segmentation faults during the compile process. In this release the compiler will halt at the first syntax error encountered. This means that any code expecting to see the specific error messages we used to produce will be broken. The error that is emitted will be one of the diagnostics that used to be produced, but in some cases some messages that used to be produced will no longer be displayed.

See "Changes to Existing Diagnostics" for more details.

utf8::upgrade()

Starting in this release, if the input string is undef, it remains undef. Previously it would be changed into a defined, zero-length string.

Changes to "thread-safe" locales

Perl 5.28 introduced "thread-safe" locales on systems that supported them, namely modern Windows, and systems supporting POSIX 2008 locale operations. These systems accomplish this by having per-thread locales, while continuing to support the older global locale operations for code that doesn't take the steps necessary to use the newer per-thread ones.

It turns out that some POSIX 2008 platforms have or have had buggy implementations, which forced perl to not use them. The ${^SAFE_LOCALES} scalar variable contains 0 or 1 to indicate whether or not the current platform is considered by perl to have a working thread-safe implementation. Some implementations have been fixed already, but FreeBSD and Cygwin have been newly discovered to be sufficiently buggy that the thread-safe operations are no longer used by perl, starting in this release. Hence, ${^SAFE_LOCALES} is now 0 for them. Older versions of perl can be configured to avoid these buggy implementations by adding the Configure option -DNO_POSIX_2008_LOCALE.

And v5.38 fixes a bug in all previous perls that led to locales not being fully thread-safe. The first thread that finishes caused the main thread (named thread0) to revert to the global locale in effect at startup, discarding whatever the thread's locale had been previously set to. If any other thread had switched to the global locale by calling switch_to_global_locale() in XS code, those threads would all share the global locale, and thread0 would not be thread-safe.

Deprecations

Use of ' as a package name separator is deprecated

Using ' as package separator in a variable named in a double-quoted string has warned since 5.28. It is now deprecated in both string interpolation and non-interpolated contexts, and will be removed in Perl 5.42.

Switch and Smart Match operator

The "switch" feature and the smartmatch operator, ~~, were introduced in v5.10. Their behavior was significantly changed in v5.10.1. When the "experiment" system was added in v5.18.0, switch and smartmatch were retroactively declared experimental. Over the years, proposals to fix or supplement the features have come and gone.

In v5.38.0, we are declaring the experiment a failure. Some future system may take the conceptual place of smartmatch, but it has not yet been designed or built.

These features will be entirely removed from perl in v5.42.0.

Performance Enhancements

  • Additional optree optimizations for common OP patterns. For example, multiple simple OPs replaced by a single streamlined OP, so as to be more efficient at runtime. [GH #19943], [GH #20063], [GH #20077].

  • Creating an anonymous sub no longer generates an srefgen op, the reference generation is now done in the anoncode or anonconst op, saving runtime. [GH #20290]

Modules and Pragmata

Updated Modules and Pragmata

  • Added the is_tainted() builtin function. [GH #19854]

  • Added the export_lexically() builtin function as per PPC 0020. [GH #19895]

  • Support for PPC 0018, use feature "module_true"; has been added to the default feature bundle for v5.38 and later. It may also be used explicitly. When enabled inside of a module the module does not need to return true explicitly, and in fact the return will be forced to a simple true value regardless of what it originally was.

  • Attribute::Handlers has been upgraded from version 1.02 to 1.03.

  • attributes has been upgraded from version 0.34 to 0.35.

  • autodie has been upgraded from version 2.34 to 2.36.

  • B has been upgraded from version 1.83 to 1.88.

  • B::Concise has been upgraded from version 1.006 to 1.007.

  • B::Deparse has been upgraded from version 1.64 to 1.74.

  • Benchmark has been upgraded from version 1.23 to 1.24.

  • bignum has been upgraded from version 0.65 to 0.66.

  • Carp has been upgraded from version 1.52 to 1.54.

  • Class::Struct has been upgraded from version 0.66 to 0.68.

  • Compress::Raw::Bzip2 has been upgraded from version 2.103 to 2.204_001.

  • Compress::Raw::Zlib has been upgraded from version 2.105 to 2.204_001.

  • Config::Perl::V has been upgraded from version 0.33 to 0.36.

  • CPAN has been upgraded from version 2.33 to 2.36.

  • Data::Dumper has been upgraded from version 2.184 to 2.188.

  • DB_File has been upgraded from version 1.857 to 1.858.

  • Devel::Peek has been upgraded from version 1.32 to 1.33.

  • Devel::PPPort has been upgraded from version 3.68 to 3.71.

  • Digest::MD5 has been upgraded from version 2.58 to 2.58_01.

  • Digest::SHA has been upgraded from version 6.02 to 6.04.

  • DynaLoader has been upgraded from version 1.52 to 1.54.

  • Encode has been upgraded from version 3.17 to 3.19.

  • encoding::warnings has been upgraded from version 0.13 to 0.14.

  • Env has been upgraded from version 1.05 to 1.06.

  • Errno has been upgraded from version 1.36 to 1.37.

  • experimental has been upgraded from version 0.028 to 0.031.

  • ExtUtils::CBuilder has been upgraded from version 0.280236 to 0.280238.

  • ExtUtils::Install has been upgraded from version 2.20 to 2.22.

  • ExtUtils::MakeMaker has been upgraded from version 7.64 to 7.70.

  • ExtUtils::Miniperl has been upgraded from version 1.11 to 1.13.

  • ExtUtils::ParseXS has been upgraded from version 3.45 to 3.51.

  • ExtUtils::PL2Bat has been upgraded from version 0.004 to 0.005.

  • ExtUtils::Typemaps has been upgraded from version 3.45 to 3.51.

  • feature has been upgraded from version 1.72 to 1.82.

  • File::Basename has been upgraded from version 2.85 to 2.86.

  • File::Copy has been upgraded from version 2.39 to 2.41.

  • File::Find has been upgraded from version 1.40 to 1.43.

  • File::Glob has been upgraded from version 1.37 to 1.40.

  • File::Spec has been upgraded from version 3.84 to 3.89.

  • File::stat has been upgraded from version 1.12 to 1.13.

  • FileHandle has been upgraded from version 2.03 to 2.05.

  • Filter::Util::Call has been upgraded from version 1.60 to 1.64.

  • GDBM_File has been upgraded from version 1.23 to 1.24.

  • Getopt::Long has been upgraded from version 2.52 to 2.54.

  • Hash::Util has been upgraded from version 0.28 to 0.30.

  • HTTP::Tiny has been upgraded from version 0.080 to 0.083.

  • I18N::Langinfo has been upgraded from version 0.21 to 0.22.

  • IO has been upgraded from version 1.50 to 1.52.

  • IO-Compress has been upgraded from version 2.106 to 2.204.

  • IO::Socket::IP has been upgraded from version 0.41 to 0.41_01.

    On DragonflyBSD, detect setsockopt() not actually clearing IPV6_V6ONLY even when setsockopt() returns success. [cpan #148293]

  • IO::Zlib has been upgraded from version 1.11 to 1.14.

  • JSON::PP has been upgraded from version 4.07 to 4.16.

  • libnet has been upgraded from version 3.14 to 3.15.

  • Locale::Maketext has been upgraded from version 1.31 to 1.33.

  • Math::BigInt has been upgraded from version 1.999830 to 1.999837.

  • Math::BigInt::FastCalc has been upgraded from version 0.5012 to 0.5013.

  • Math::BigRat has been upgraded from version 0.2621 to 0.2624.

  • Math::Complex has been upgraded from version 1.5902 to 1.62.

  • Memoize has been upgraded from version 1.03_01 to 1.16.

  • MIME::Base64 has been upgraded from version 3.16 to 3.16_01.

  • Module::CoreList has been upgraded from version 5.20220520 to 5.20230520.

  • mro has been upgraded from version 1.26 to 1.28.

  • NDBM_File has been upgraded from version 1.15 to 1.16.

  • Net::Ping has been upgraded from version 2.74 to 2.76.

  • ODBM_File has been upgraded from version 1.17 to 1.18.

  • Opcode has been upgraded from version 1.57 to 1.64.

  • overload has been upgraded from version 1.35 to 1.37.

  • parent has been upgraded from version 0.238 to 0.241.

  • PerlIO::via::QuotedPrint has been upgraded from version 0.09 to 0.10.

  • Pod::Checker has been upgraded from version 1.74 to 1.75.

  • Pod::Html has been upgraded from version 1.33 to 1.34.

  • Pod::Usage has been upgraded from version 2.01 to 2.03.

  • podlators has been upgraded from version 4.14 to 5.01.

  • POSIX has been upgraded from version 2.03 to 2.13.

  • re has been upgraded from version 0.43 to 0.44.

  • Safe has been upgraded from version 2.43 to 2.44.

  • Scalar::Util has been upgraded from version 1.62 to 1.63.

  • SDBM_File has been upgraded from version 1.15 to 1.17.

  • Socket has been upgraded from version 2.033 to 2.036.

  • Storable has been upgraded from version 3.26 to 3.32.

  • Sys::Hostname has been upgraded from version 1.24 to 1.25.

  • Term::Cap has been upgraded from version 1.17 to 1.18.

  • Test::Simple has been upgraded from version 1.302190 to 1.302194.

  • Text::Balanced has been upgraded from version 2.04 to 2.06.

  • threads has been upgraded from version 2.27 to 2.36.

  • threads::shared has been upgraded from version 1.64 to 1.68.

  • Tie::File has been upgraded from version 1.06 to 1.07.

  • Time::HiRes has been upgraded from version 1.9770 to 1.9775.

  • Time::Piece has been upgraded from version 1.3401 to 1.3401_01.

  • Unicode::Normalize has been upgraded from version 1.31 to 1.32.

  • UNIVERSAL has been upgraded from version 1.14 to 1.15.

  • User::grent has been upgraded from version 1.03 to 1.04.

  • User::pwent has been upgraded from version 1.01 to 1.02.

  • utf8 has been upgraded from version 1.24 to 1.25.

  • warnings has been upgraded from version 1.58 to 1.65.

  • XS::APItest has been upgraded from version 1.22 to 1.32.

  • XSLoader has been upgraded from version 0.31 to 0.32.

Documentation

New Documentation

perlclass

Describes the new class feature.

perlclassguts

Describes the internals of the new class feature.

Changes to Existing Documentation

We have attempted to update the documentation to reflect the changes listed in this document. If you find any we have missed, open an issue at https://github.com/Perl/perl5/issues.

Additionally, the following selected changes have been made:

perlapi

perldeprecation

  • Added information about unscheduled deprecations and their categories.

  • Added category information for existing scheduled deprecations.

  • Added smartmatch and apostrophe as a package separator deprecation data.

perlintern

  • Documented save_pushptr

  • Documented save_scalar_at

  • Entries have been added to perlguts for the new newAV_alloc_x, newAV_alloc_xz and *_simple functions.

  • References to the now-defunct PrePAN service have been removed from perlcommunity and perlmodstyle.

  • A section on symbol naming has been added to perlhacktips.

  • perlexperiment has been edited to properly reference the warning categories for the defer block modifier and extra paired delimiters for quote-like operators.

perlexperiment

  • Smartmatch has been moved from experimental status to deprecated status. Unfortunately the experiment did not work out.

perlfunc

  • Some wording improvements have been made for the ucfirst, push, unshift and bless functions, as well as additional examples added.

perlhacktips

perlop

  • Document the behavior of matching the empty pattern better and specify its relationship to the new ${^LAST_SUCCESSFUL_PATTERN} properly.

perlvar

  • Added a section on "Scoping Rules of Regex Variables", and other wording improvements made throughout.

  • Added information on the new %{^HOOK} interface, and the new require__before and require__after hooks which it exposes.

  • Correct information on the regex variables ${^PREMATCH}, ${^MATCH} and ${^POSTMATCH}, all of which were incorrectly documented due to an oversight. Specifically they only work properly after a regex operation that used the /p modifier to enable them.

  • Added information on the new regex variable ${^LAST_SUCCESSFUL_PATTERN}, which represents the pattern of the last successful regex match in scope.

Diagnostics

The following additions or changes have been made to diagnostic output, including warnings and fatal error messages. For the complete list of diagnostic messages, see perldiag.

New Diagnostics

New Errors

New Warnings

Changes to Existing Diagnostics

  • The compiler will now stop parsing on the first syntax error it encounters. Historically the compiler would attempt to "skip past" the error and continue parsing so that it could list multiple errors. For things like undeclared variables under strict this makes sense. For syntax errors however it has been found that continuing tends to result in a storm of unrelated or bizarre errors that mostly just obscure the true error. In extreme cases it can even lead to segfaults and other incorrect behavior.

    Therefore we have reformed the continuation logic so that the parse will stop after the first seen syntax error. Semantic errors like undeclared variables will not stop the parse, so you may still see multiple errors when compiling code. However if there is a syntax error it will be the last error message reported by perl and all of the errors that you see will be something that actually needs to be fixed.

  • Error messages that output class or package names have been modified to output double quoted strings with various characters escaped so as to make the exact value clear to a reader. The exact rules on which characters are escaped may change over time but currently are that printable ASCII codepoints, with the exception of " and \, and unicode word characters whose codepoint is over 255 are output raw, and any other symbols are escaped much as Data::Dumper might escape them, using \n for newline and \" for double quotes, etc. Codepoints in the range 128-255 are always escaped as they can cause trouble on unicode terminals when output raw.

    In older versions of perl the one liner

    $ perl -le'"thing\n"->foo()'

    would output the following error message exactly as shown here, with text spread over multiple lines because the "\n" would be emitted as a raw newline character:

    Can't locate object method "foo" via package "thing
    " (perhaps you forgot to load "thing
    "?) at -e line 1.

    As of this release we would output this instead (as one line):

    Can't locate object method "foo" via package "thing\n"
    (perhaps you forgot to load "thing\n"?) at -e line 1.

    Notice the newline in the package name has been quoted and escaped, and thus the error message is a single line. The text is shown here wrapped to two lines only for readability.

  • When package or class names in errors are very large the middle excess portion will be elided from the message. As of this release error messages will show only up to the first 128 characters and the last 128 characters in a package or class name in error messages. For example

    $ perl -le'("Foo" x 1000)->new()'

    will output the following as one line:

    Can't locate object method "new" via package "FooFooFooFooFooFooFoo
    FooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFoo
    FooFooFooFooFooFooFooFooFooFooFooFooFooFo"..."oFooFooFooFooFooFooFoo
    FooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFoo
    FooFooFooFooFooFooFooFooFooFooFooFooFoo" (perhaps you forgot to load
    "FooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFoo
    FooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFo"...
    "oFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFoo
    FooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFooFoo"?)
    at -e line 1.

    Notice the "prefix"..."suffix" form of the package name in this case. In previous versions of perl the complete string would have been shown making the error message over 6k long and there was no upper limit on the length of the error message at all. If you accidentally used a 1MB string as a class name then the error message would be over 2MB long. In this perl the upper limit should be around 2k when eliding and escaping are taken into account.

  • Removed Complex regular subexpression recursion limit (%d) exceeded

    The regular expresion engine has not used recursion in some time. This warning no longer makes sense.

    See [GH #19636].

  • Various warnings that used to produce parenthesized hints underneath the main warning message and after its "location data" were chanaged to put the hint inline with the main message. For instance:

    Bareword found where operator expected at -e line 1, near "foo bar"
    (Do you need to predeclare foo?)

    will now look like this but as one line:

    Bareword found where operator expected (Do you need to predeclare
    foo?) at -e line 1, near "foo bar"

    as a result such warnings will no longer trigger $SIG{__WARN__} twice, and the hint will be visible when fatal warnings is in effect.

  • The error message that is produced when a require or use statement fails has been changed. It used to contain the words @INC contains:, and it used to show the state of @INC *after* the require had completed and failed. The error message has been changed to say @INC entries checked: and to reflect the actual directories or hooks that were executed during the require statement. For example:

    perl -e'push @INC, sub {@INC=()}; eval "require Frobnitz"
    or die $@'
    Can't locate Frobnitz.pm in @INC (you may need to install the
    Frobnitz module) (@INC contains:) at (eval 1) line 1.

    Will change to (with some output elided for clarity):

    perl -e'push @INC, sub {@INC=()}; eval "require Frobnitz"
    or die $@'
    Can't locate Frobnitz.pm in @INC (you may need to install the
    Frobnitz module) (@INC entries checked:
    .../site_perl/5.38.0/x86_64-linux .../site_perl/5.38.0
    .../5.38.0/x86_64-linux .../5.38.0 CODE(0x562745e684b8))
    at (eval 1) line 1.

    thus showing the actual directories checked. Code that checks for @INC contains: in error messages should be hardened against any future wording changes between the @INC and :, for instance use qr/\@INC[ \w]+:/ instead of using qr/\@INC contains:/ or qr/\@INC entries checked:/ in tests as this will ensure both forward and backward compatibility.

  • Old package separator used in string

    This diagnostic is now also part of the deprecated category.

  • given is deprecated replaces given is experimental.

  • when is deprecated replaces when is experimental.

  • Smartmatch is deprecated replaces Smartmatch is experimental.

Configuration and Compilation

  • make -j6 minitest could fail due to a build conflict in building $(MINIPERL_EXE) between the main make process and a child process. [GH #19829]

  • Properly populate osvers on Dragonfly BSD when the hostname isn't set.

  • Fix typos for C99 macro name PRIX64.

  • Remove ancient and broken GCC for VMS support

  • Remove vestigial reference to /VAXC qualifier

  • Remove sharedperl option on VMS

  • VMS now has mkostemp

  • Configure now properly handles quoted elements outputted by gcc. [GH #20606]

  • Configure probed for the return type of malloc() and free() by testing whether declarations for those functions produced a function type mismatch with the implementation. On Solaris, with a C++ compiler, this check always failed, since Solaris instead imports malloc() and free() from std:: with using for C++ builds. Since the return types of malloc() and free() are well defined by the C standard, skip probing for them. Configure command-line arguments and hints can still override these type in the unlikely case that is needed. [GH #20806]

Testing

Tests were added and changed to reflect the other additions and changes in this release. Furthermore, these significant changes were made:

  • Unicode normalization tests have been added.

  • t/test.pl: Add ability to cancel an watchdog timer

Platform Support

Discontinued Platforms

Ultrix

Support code for DEC Ultrix has been removed. Ultrix was the native Unix-like operating system for various Digital Equipment Corporation machines. Its final release was in 1995.

Platform-Specific Notes

DragonflyBSD

Skip tests to workaround an apparent bug in setproctitle(). [GH #19894]

FreeBSD

FreeBSD no longer uses thread-safe locale operations, to avoid a bug in FreeBSD

Replace the first part of archname with uname -p [GH #19791]

Solaris

Avoid some compiler and compilation issues on NetBSD/Solaris from regexec.c and regcomp.c.

Synology

Update Synology Readme for DSM 7.

Windows

Fix win32 memory alignment needed for gcc-12 from vmem.h.

utimes() on Win32 would print a message to stderr if it failed to convert a supplied time_t to a FILETIME. [GH #19668]

In some cases, timestamps returned by stat() and lstat() failed to take daylight saving time into account. [GH #20018] [GH #20061]

stat() now works on AF_UNIX socket files. [GH #20204]

readlink() now returns the PrintName from a symbolic link reparse point instead of the SubstituteName, which should make it better match the name the link was created with. [GH #20271]

lstat() on Windows now returns the length of the link target as the size of the file, as it does on POSIX systems. [GH #20476]

symlink() on Windows now replaces any / in the target with \, since Windows does not recognise / in symbolic links. The reverse translation is not done by readlink(). [GH #20506]

symlink() where the target was an absolute path to a directory was incorrectly created as a file symbolic link. [GH #20533]

POSIX::dup2 no longer creates broken sockets. [GH #20920]

Closing a busy pipe could cause Perl to hang. [GH #19963]

Internal Changes

  • Removed many deprecated C functions.

    These have been deprecated for a long time. See https://github.com/perl/perl5/commit/7008caa915ad99e650acf2aea40612b5e48b7ba2 for a full list.

  • get_op_descs, get_op_names, get_opargs, get_no_modify and get_ppaddr have been marked deprecated.

  • hv_free_ent has been marked as internal API.

  • save_pushptr, save_pushptrptr, and save_pushi32ptr have been marked as internal API.

  • New bool related functions and macros have been added to complement the new bool type introduced in 5.36:

    The functions are:

    newSVbool(const bool bool_val)
    newSV_true()
    newSV_false()
    sv_set_true(SV *sv)
    sv_set_false(SV *sv)
    sv_set_bool(SV *sv, const bool bool_val)

    The macros are:

    SvIandPOK(sv)
    SvIandPOK_off(sv)
    SvIandPOK_on
  • Perl is no longer manipulating the environ array directly. The variable PL_use_safe_putenv has been removed and PERL_USE_SAFE_PUTENV is always defined. This means XS modules can now call setenv and putenv without causing segfaults. [perl #19399]

  • Internal C API functions are now hidden with __attribute__((hidden)) on the platforms that support it. This means they are no longer callable from XS modules on those platforms.

    It should be noted that those functions have always been hidden on Windows. This change merely brings that to the other platforms. [perl #19655]

  • New formatting symbols were added for printing values declared as U32 or I32:

    I32df -- Like %d
    U32of -- Like %o
    U32uf -- Like %u
    U32xf -- Like %x
    U32Xf -- Like %X

    These are used in the same way already existing similar symbols, such as IVdf, are used. See "I/O Formats" in perlapi.

  • new 'HvHasAUX' macro

  • regexec.c: Add some branch predictions reorder conds

  • locale: Change macro name to be C conformant

  • Rename the PADNAMEt_* constants to PADNAMEf_*

  • Changes all the API macros that retrieve a PV into a call to an inline function so as to evaluate the parameter just once.

  • regexec.c: multiple code refactor to make the code more readable

  • perl.h: Change macro name to be C conformant (remove leading _ from NOT_IN_NUMERIC macros)

  • regcomp.h: add new BITMAP_BIT macro in addition to the existing BITMAP_BYTE and BITMAP_TEST ones.

  • Create new regnode type ANYOFH. populate_ANYOF_from_invlist was renamed to populate_bitmap_from_invlist

  • regex: Refactor bitmap vs non-bitmap of qr/[]/

  • regcomp.c: add new functions to convert from an inversion list to a bitmap (and vice versa) populate_bitmap_from_invlist and populate_invlist_from_bitmap.

  • Add newAVav() to create an AV from an existing AV. Add newAVhv() to create an AV using keys and values from an existing HV.

  • Fix definition of Perl_atof.

  • Fix undefined behavior with overflow related OPTIMIZE_INFTY and delta in regcomp.c.

  • Fix regnode pointer alignment issue in regcomp.h.

  • The CVf_METHOD CV flag and associated CvMETHOD macro has been renamed to CVf_NOWARN_AMBIGUOUS and CvNOWARN_AMBIGUOUS. This closer reflects its actual behaviour (it suppresses a warning that would otherwise be generated about ambiguous names), in order to be less confusing with CvIsMETHOD, which indicates that a CV is a method subroutine relating to the class feature.

  • The OPf_SPECIAL flag is no longer set on the OP_ENTERSUB op constructed to call the VERSION, import and unimport methods as part of a use statement and attribute application, nor when assigning to an :lvalue subroutine.

  • A new CV flag CVf_REFCOUNTED_ANYSV has been added, which indicates that the CV is an XSUB and stores an SV pointer in the CvXSUBANY.any_sv union field. Perl core operations such as cloning or destroying the CV will maintain the reference count of the pointed-to SV, destroying it when required.

  • A new API function "Perl_localeconv" in perlapi is added. This is the same as POSIX::localeconv (returning a hash of the localeconv() fields), but directly callable from XS code.

  • A new API function, "Perl_langinfo8" in perlapi is added. This is the same as plain "Perl_langinfo" in perlapi, but with an extra parameter that allows the caller to simply and reliably know if the returned string is UTF-8.

  • We have introduced a limit on the number of nested eval EXPR/BEGIN blocks and require/BEGIN (and thus use statements as well) to prevent C stack overflows. This variable can also be used to forbid BEGIN blocks from executing during eval EXPR compilation. The limit defaults to 1000 but can be overridden by setting the ${^MAX_NESTED_EVAL_BEGIN_BLOCKS} variable. The default itself can be changed at compile time with

    -Accflags='-DPERL_MAX_NESTED_EVAL_BEGIN_BLOCKS_DEFAULT=12345'

    Note that this value relates to the size of your C stack and if you choose an inappropriately large value Perl may segfault, be conservative about what you choose.

  • A new magic type PERL_MAGIC_extvalue has been added. This is available for use like PERL_MAGIC_ext, but is a value magic: upon localization the new value will not be magical.

  • The SSNEW(), SSNEWt(), SSNEWa() and SSNEWat() APIs now return a SSize_t value. The SSPTR() and SSPTRt() macros now expect a SSize_t parameter, and enforce that on debugging builds. [GH #20411]

  • Filenames in cops are now refcounted under threads. Under threads we were copying the filenames into each opcode. This is because in theory opcodes created in one thread can be destroyed in another. The change adds a new struct/type RCPV, which is a refcounted string using shared memory. This is implemented in such a way that code that previously used a char * can continue to do so, as the refcounting data is located a specific offset before the char * pointer itself.

  • Added HvNAMEf and HvNAMEf_QUOTEDPREFIX special formats. They take an HV * as an argument and use HvNAME() and related macros to determine the string, its length, and whether it is utf8.

  • The underlying Perl_dowantarray function implementing the long-deprecated GIMME macro has been marked as deprecated, so that use of the macro emits a compile-time warning. GIMME has been documented as deprecated in favour of GIMME_V since Perl v5.6.0, but had not previously issued a warning.

  • The API function "utf8_length" in perlapi is now more efficient.

  • Added SAVERCPV() and SAVEFREERCPV() for better support for working with RCPV (reference counted string/pointer value) structures which currently are used in opcodes to share filename and warning bit data in a memory efficient manner.

  • Added MORTALSVFUNC_SV() and MORTALDESTRUCTOR_SV() macros, which make it possible to create a destructor which is fired at the end of the current statement. This uses the PERL_MAGIC_destruct magic to use "free" magic to trigger an action when a variable is freed. The action can be specified as a C function or as a Perl code reference.

  • Added the %{^HOOK} api and related PERL_MAGIC_hook and PERL_MAGIC_hookelem for providing ways to hook selected perl functions which for one reason or another are problematic to wrap with a customized subroutine.

  • Added support for ${^HOOK}{require__before} which can be used to rewrite the filename that require will try to load, and also to block require from loading a specific module, even via fully qualified filename. The hook can also be used to perform "pre-require" and "post-require" actions.

  • Added support for ${^HOOK}{require__after} which can be used to track what modules have been required after the fact.

  • Regular expression opcodes (regops) now use a standardized structure layout that uses unions to expose data in different format. This means it should be much easier to extend or modify regops to use more memory. This has been used to make a number of regops track how many parens they contain.

Selected Bug Fixes

  • Avoid recursion and stack overflow parsing 'pack' template

    [GH #16319]

  • An eval() as the last statement in a regex code block could trigger an interpreter panic; e.g.

    /(?{ ...; eval {....}; })/

    [GH #19680]

  • Disabling the bareword_filehandles feature no longer treats print Class->method as an error. [GH #19704]

  • When a Perl subroutine tail-calls an XS subroutine using goto &xs_sub, the XS subroutine can now correctly determine its calling context. Previously it was always reported as scalar.

    In addition, where the Perl subroutine is freed at the same time:

    sub foo { *foo = sub {}; goto &xs_sub }

    this formerly could lead to crashes if the XS subroutine tried to use the value of PL_op, since this was being set to NULL. This is now fixed.

    [GH #19936]

  • setsockopt() now uses the mechanism added in 5.36 to better distinguish between numeric and string values supplied as the OPTVAL parameter. [GH #18761]

  • 4-argument select() now rejects strings with code points above 255. Additionally, for code points 128-255, this operator will now always give the corresponding octet to the OS, regardless of how Perl stores such code points in memory. (Previously Perl leaked its internal string storage to the OS.) [GH #19882]

  • Fix panic issue from val {} inside /(?{...})/ [GH #19390]

  • Fix multiple compiler warnings from regexp.c, locale.c [GH #19915]

  • Fix a bug with querying locales on platforms that don't have LC_NUMERIC [GH #19890]

  • Prevent undefined behaviour in S_maybe_multideref().

  • Avoid signed integer overflow in use integer ops.

  • Avoid adding an offset to a NULL pointer in hv_delete_common.

  • PerlIO::get_layers will now accept IO references too

    Previously it would only take glob references or names of globs. Now it will also accept IO references.

  • Fixes to memory handling for PL_splitstr:

    • If a thread was created the allocated string would be freed twice.

    • If two -F switches were supplied the memory allocated for the first switch wouldn't be freed.

  • Correctly handle OP_ANONCODE ops generated by CPAN modules that don't include the OPf_REF flag when propagating lvalue context. [GH #20532]

  • POSIX::strxfrm now uses the LC_CTYPE locale category to specify its collation, ignoring any differing LC_COLLATE. It doesn't make sense for a string to be encoded in one locale (say, ISO-8859-6, Arabic) and to collate it based on another (like ISO-8859-7, Greek). Perl assumes that the current LC_CTYPE locale correctly represents the encoding, and collates accordingly.

    Also, embedded NUL characters are now allowed in the input.

    If locale collation is not enabled on the platform (LC_COLLATE), the input is returned unchanged.

  • Double FETCH during stringification of tied scalars returning an overloaded object have been fixed. The FETCH method should only be called once, but prior to this release was actually called twice. [GH #20574]

  • Writing to a magic variables associated with the selected output handle, $^, $~, $=, $- and $%, no longer crashes perl if the IO object has been cleared from the selected output handle. [GH #20733]

  • Redefining a use constant list constant with use constant now properly warns. This changes the behaviour of use constant but is a core change, not a change to constant.pm. [GH #20742]

  • Redefining a use constant list constant with an empty prototype constant sub would result in an assertion failure. [GH #20742]

  • Fixed a regression where the INC method for objects in @INC would not be resolved by AUTOLOAD, while it was in 5.36. The INCDIR method for objects in @INC cannot be resolved by AUTOLOAD as INC would have been resolved first. [GH #20665]

  • $SIG{__DIE__} will now be called from eval when the code dies during compilation regardless of how it dies. This means that code expecting to be able to upgrade $@ into an object will be called consistently. In earlier versions of perl $SIG{__DIE__} would not be called for certain compilation errors, for instance undeclared variables. For other errors it might be called if there were more than a certain number of errors, but not if there were less. Now you can expect that it will be called in every case.

  • Compilation of code with errors used to inconsistently stop depending on the count and type of errors encountered. The intent was that after 10 errors compilation would halt, but bugs in this logic meant that certain types of error would be counted, but would not trigger the threshold check to stop compilation. Other errors would. With this release after at most 10 errors compilation will terminate, regardless of what type of error they were.

    Note that you can change the maximum count by defining PERL_STOP_PARSING_AFTER_N_ERRORS to be something else during the configuration process. For instance

    ./Configure ... -Accflags='-DPERL_STOP_PARSING_AFTER_N_ERRORS=100'

    would allow up to 100 errors.

  • The API function "my_snprintf" in perlapi now prints a non-dot decimal point if the perl code it ultimately is called from is in the scope of use locale and the locale in effect calls for that.

  • A number of bugs related to capture groups in quantified groups in regular expression have been fixed, especially in alternations. For example in a pattern like:

    "foobazfoobar" =~ /((foo)baz|foo(bar))+/

    the regex variable $2 will not be "foo" as it once was, it will be undef.

  • Bugs with regex backreference operators that are inside of a capture group have been fixed. For instance:

    "xa=xaaa" =~ /^(xa|=?\1a){2}\z/

    will now correctly not match. [GH #10073]

  • SSGROW() and SSCHECK() have been reworked to ensure that the requested space is actually allocated. SSCHECK() is now an alias for SSGROW().

Acknowledgements

Perl 5.38.0 represents approximately 12 months of development since Perl 5.36.0 and contains approximately 290,000 lines of changes across 1,500 files from 100 authors.

Excluding auto-generated files, documentation and release tools, there were approximately 190,000 lines of changes to 970 .pm, .t, .c and .h files.

Perl continues to flourish into its fourth decade thanks to a vibrant community of users and developers. The following people are known to have contributed the improvements that became Perl 5.38.0:

Alex, Alexander Nikolov, Alex Davies, Andreas König, Andrew Fresh, Andrew Ruthven, Andy Lester, Aristotle Pagaltzis, Arne Johannessen, A. Sinan Unur, Bartosz Jarzyna, Bart Van Assche, Benjamin Smith, Bram, Branislav Zahradník, Brian Greenfield, Bruce Gray, Chad Granum, Chris 'BinGOs' Williams, chromatic, Clemens Wasser, Craig A. Berry, Dagfinn Ilmari Mannsåker, Dan Book, danielnachun, Dan Jacobson, Dan Kogai, David Cantrell, David Golden, David Mitchell, E. Choroba, Ed J, Ed Sabol, Elvin Aslanov, Eric Herman, Felipe Gasper, Ferenc Erki, Firas Khalil Khana, Florian Weimer, Graham Knop, Håkon Hægland, Harald Jörg, H.Merijn Brand, Hugo van der Sanden, James E Keenan, James Raspass, jkahrman, Joe McMahon, Johan Vromans, Jonathan Stowe, Jon Gentle, Karen Etheridge, Karl Williamson, Kenichi Ishigaki, Kenneth Ölwing, Kurt Fitzner, Leon Timmermans, Li Linjie, Loren Merritt, Lukas Mai, Marcel Telka, Mark Jason Dominus, Mark Shelor, Matthew Horsfall, Matthew O. Persico, Mattia Barbon, Max Maischein, Mohammad S Anwar, Nathan Mills, Neil Bowers, Nicholas Clark, Nicolas Mendoza, Nicolas R, Paul Evans, Paul Marquess, Peter John Acklam, Peter Levine, Philippe Bruhat (BooK), Reini Urban, Renee Baecker, Ricardo Signes, Richard Leach, Russ Allbery, Scott Baker, Sevan Janiyan, Sidney Markowitz, Sisyphus, Steve Hay, TAKAI Kousuke, Todd Rinaldo, Tomasz Konojacki, Tom Stellard, Tony Cook, Tsuyoshi Watanabe, Unicode Consortium, vsfos, Yves Orton, Zakariyya Mughal, Zefram, 小鸡.

The list above is almost certainly incomplete as it is automatically generated from version control history. In particular, it does not include the names of the (very much appreciated) contributors who reported issues to the Perl bug tracker.

Many of the changes included in this version originated in the CPAN modules included in Perl's core. We're grateful to the entire CPAN community for helping Perl to flourish.

For a more complete list of all of Perl's historical contributors, please see the AUTHORS file in the Perl source distribution.

Reporting Bugs

If you find what you think is a bug, you might check the perl bug database at https://github.com/Perl/perl5/issues. There may also be information at http://www.perl.org/, the Perl Home Page.

If you believe you have an unreported bug, please open an issue at https://github.com/Perl/perl5/issues. Be sure to trim your bug down to a tiny but sufficient test case.

If the bug you are reporting has security implications which make it inappropriate to send to a public issue tracker, then see "SECURITY VULNERABILITY CONTACT INFORMATION" in perlsec for details of how to report the issue.

Give Thanks

If you wish to thank the Perl 5 Porters for the work we had done in Perl 5, you can do so by running the perlthanks program:

perlthanks

This will send an email to the Perl 5 Porters list with your show of thanks.

SEE ALSO

The Changes file for an explanation of how to view 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.