/ 
perl-5.22.0-RC1
River stage five • 12010 direct dependents • 33600 total dependents
⭐ Starred 2013 GitHub stars
/ perldelta

NAME

perldelta - what is new for perl v5.22.0

DESCRIPTION

This document describes differences between the 5.22.0 release and the 5.20.0 release.

If you are upgrading from an earlier release such as 5.18.0, first read perl5200delta, which describes differences between 5.18.0 and 5.20.0.

Core Enhancements

New bitwise operators

A new experimental facility has been added that makes the four standard bitwise operators (& | ^ ~) treat their operands consistently as numbers, and introduces four new dotted operators (&. |. ^. ~.) that treat their operands consistently as strings. The same applies to the assignment variants (&= |= ^= &.= |.= ^.=).

To use this, enable the "bitwise" feature and disable the "experimental::bitwise" warnings category. See "Bitwise String Operators" in perlop for details. [rt.perl.org #123466]

New double-diamond operator

<<>> is like <> but uses three-argument open to open each file in @ARGV. So each element of @ARGV is an actual file name, and "|foo" won't be treated as a pipe open.

New \b boundaries in regular expressions

qr/\b{gcb}/

gcb stands for Grapheme Cluster Boundary. It is a Unicode property that finds the boundary between sequences of characters that look like a single character to a native speaker of a language. Perl has long had the ability to deal with these through the \X regular escape sequence. Now, there is an alternative way of handling these. See "\b{}, \b, \B{}, \B" in perlrebackslash for details.

qr/\b{wb}/

wb stands for Word Boundary. It is a Unicode property that finds the boundary between words. This is similar to the plain \b (without braces) but is more suitable for natural language processing. It knows, for example that apostrophes can occur in the middle of words. See "\b{}, \b, \B{}, \B" in perlrebackslash for details.

qr/\b{sb}/

sb stands for Sentence Boundary. It is a Unicode property to aid in parsing natural language sentences. See "\b{}, \b, \B{}, \B" in perlrebackslash for details.

no re covers more and is lexical

Previously running no re would only turn off a few things. Now it turns off all the enabled things. For example, previously, you couldn't turn off debugging, once enabled, inside the same block.

Non-Capturing Regular Expression Flag

Regular expressions now support a /n flag that disables capturing and filling in $1, $2, etc... inside of groups:

"hello" =~ /(hi|hello)/n; # $1 is not set

This is equivalent to putting ?: at the beginning of every capturing group.

See "n" in perlre for more information.

use re 'strict'

This applies stricter syntax rules to regular expression patterns compiled within its scope, which hopefully will alert you to typos and other unintentional behavior that backwards-compatibility issues prevent us from doing in normal regular expression compilations. Because the behavior of this is subject to change in future Perl releases as we gain experience, using this pragma will raise a category experimental::re_strict warning. See 'strict' in re.

qr/foo/x now ignores any Unicode pattern white space

The /x regular expression modifier allows the pattern to contain white space and comments, both of which are ignored, for improved readability. Until now, not all the white space characters that Unicode designates for this purpose were handled. The additional ones now recognized are U+0085 NEXT LINE, U+200E LEFT-TO-RIGHT MARK, U+200F RIGHT-TO-LEFT MARK, U+2028 LINE SEPARATOR, and U+2029 PARAGRAPH SEPARATOR.

Unicode 7.0 is now supported

For details on what is in this release, see http://www.unicode.org/versions/Unicode7.0.0/.

use locale can restrict which locale categories are affected

It is now possible to pass a parameter to use locale to specify a subset of locale categories to be locale-aware, with the remaining ones unaffected. See "The "use locale" pragma" in perllocale for details.

Perl now supports POSIX 2008 locale currency additions.

On platforms that are able to handle POSIX.1-2008, the hash returned by POSIX::localeconv() includes the international currency fields added by that version of the POSIX standard. These are int_n_cs_precedes, int_n_sep_by_space, int_n_sign_posn, int_p_cs_precedes, int_p_sep_by_space, and int_p_sign_posn.

Better heuristics on older platforms for determining locale UTF8ness

On platforms that implement neither the C99 standard nor the POSIX 2001 standard, determining if the current locale is UTF8 or not depends on heuristics. These are improved in this release.

Aliasing via reference

Variables and subroutines can now be aliased by assigning to a reference:

\$c = \$d;
\&x = \&y;

Or by using a backslash before a foreach iterator variable, which is perhaps the most useful idiom this feature provides:

foreach \%hash (@array_of_hash_refs) { ... }

This feature is experimental and must be enabled via use feature 'refaliasing'. It will warn unless the experimental::refaliasing warnings category is disabled.

See "Assigning to References" in perlref

prototype with no arguments

prototype() with no arguments now infers $_. [perl #123514]

New "const" subroutine attribute

The "const" attribute can be applied to an anonymous subroutine. It causes it to be executed immediately when it is cloned. Its value is captured and used to create a new constant subroutine that is returned. This feature is experimental. See "Constant Functions" in perlsub.

fileno now works on directory handles

When the relevant support is available in the operating system, the fileno builtin now works on directory handles, yielding the underlying file descriptor in the same way as for filehandles. On operating systems without such support, fileno on a directory handle continues to return the undefined value, as before, but also sets $! to indicate that the operation is not supported.

Currently, this uses either a dd_fd member in the OS DIR structure, or a dirfd(3) function as specified by POSIX.1-2008.

List form of pipe open implemented for Win32

The list form of pipe:

open my $fh, "-|", "program", @arguments;

is now implemented on Win32. It has the same limitations as system LIST on Win32, since the Win32 API doesn't accept program arguments as a list.

close now sets $!

When an I/O error occurs, the fact that there has been an error is recorded in the handle. close returns false for such a handle. Previously, the value of $! would be untouched by close, so the common convention of writing close $fh or die $! did not work reliably. Now the handle records the value of $!, too, and close restores it.

Assignment to list repetition

(...) x ... can now be used within a list that is assigned to, as long as the left-hand side is a valid lvalue. This allows (undef,undef,$foo) = that_function() to be written as ((undef)x2, $foo) = that_function().

Infinity and NaN (not-a-number) handling improved

Floating point values are able to hold the special values infinity (also -infinity), and NaN (not-a-number). Now we more robustly recognize and propagate the value in computations, and on output normalize them to Inf and NaN.

See also the POSIX enhancements.

Floating point parsing has been improved

Parsing and printing of floating point values has been improved.

As a completely new feature, hexadecimal floating point literals (like 0x1.23p-4) are now supported, and they can be output with printf %a.

Packing infinity or not-a-number into a character is now fatal

Before, when trying to pack infinity or not-a-number into a (signed) character, Perl would warn, and assumed you tried to pack 0xFF; if you gave it as an argument to chr, U+FFFD was returned.

But now, all such actions (pack, chr, and print '%c') result in a fatal error.

Experimental C Backtrace API

Starting from Perl 5.21.1, on some platforms Perl supports retrieving the C level backtrace (similar to what symbolic debuggers like gdb do).

The backtrace returns the stack trace of the C call frames, with the symbol names (function names), the object names (like "perl"), and if it can, also the source code locations (file:line).

The supported platforms are Linux and OS X (some *BSD might work at least partly, but they have not yet been tested).

The feature needs to be enabled with Configure -Dusecbacktrace.

Also included is a C API to retrieve backtraces.

See "C backtrace" in perlhacktips for more information.

Security

Perl is now compiled with -fstack-protector-strong if available

Perl has been compiled with the anti-stack-smashing option -fstack-protector since 5.10.1. Now Perl uses the newer variant called -fstack-protector-strong, if available.

The Safe module could allow outside packages to be replaced

Critical bugfix: outside packages could be replaced. Safe has been patched to 2.38 to address this.

Perl is now always compiled with -D_FORTIFY_SOURCE=2 if available

The 'code hardening' option called _FORTIFY_SOURCE, available in gcc 4.*, is now always used for compiling Perl, if available.

Note that this isn't necessarily a huge step since in many platforms the step had already been taken several years ago: many Linux distributions (like Fedora) have been using this option for Perl, and OS X has enforced the same for many years.

Incompatible Changes

Subroutine signatures moved before attributes

The experimental sub signatures feature, as introduced in 5.20, parsed signatures after attributes. In this release, the positioning has been moved such that signatures occur after the subroutine name (if any) and before the attribute list (if any).

& and \& prototypes accepts only subs

The & prototype character now accepts only anonymous subs (sub {...}) and things beginning with \&. Formerly it erroneously also allowed undef and references to array, hashes, and lists. [perl #4539] [perl #123062]

The \& prototype was allowing subroutine calls, whereas now it only allows subroutines. &foo is permitted. &foo() and foo() are not. [perl #77860]

use encoding is now lexical

The encoding pragma's effect is now limited to lexical scope. This pragma is deprecated, but in the meantime, it could adversely affect unrelated modules that are included in the same program.

List slices returning empty lists

List slices return an empty list now only if the original list was empty (or if there are no indices). Formerly, a list slice would return an empty list if all indices fell outside the original list. [perl #114498]

\N{} with a sequence of multiple spaces is now a fatal error.

This has been deprecated since v5.18.

use UNIVERSAL '...' is now a fatal error

Importing functions from UNIVERSAL has been deprecated since v5.12, and is now a fatal error. "use UNIVERSAL" without any arguments is still allowed.

In double-quotish \cX, X must now be a printable ASCII character

In prior releases, failure to do this raised a deprecation warning.

Splitting the tokens (? and (* in regular expressions is now a fatal compilation error.

These had been deprecated since v5.18.

5 additional characters are treated as white space under /x in regex patterns (unless escaped)

The use of these characters with /x outside bracketed character classes and when not preceded by a backslash has raised a deprecation warning since v5.18. Now they will be ignored. See "qr/foo/x" for the list of the five characters.

Comment lines within (?[ ]) now are ended only by a \n

(?[ ]) is an experimental feature, introduced in v5.18. It operates as if /x is always enabled. But there was a difference, comment lines (following a # character) were terminated by anything matching \R which includes all vertical whitespace, such as form feeds. For consistency, this is now changed to match what terminates comment lines outside (?[ ]), namely a \n (even if escaped), which is the same as what terminates a heredoc string and formats.

(?[...]) operators now follow standard Perl precedence

This experimental feature allows set operations in regular expression patterns. Prior to this, the intersection operator had the same precedence as the other binary operators. Now it has higher precedence. This could lead to different outcomes than existing code expects (though the documentation has always noted that this change might happen, recommending fully parenthesizing the expressions). See "Extended Bracketed Character Classes" in perlrecharclass.

Omitting % and @ on hash and array names is no longer permitted

Really old Perl let you omit the @ on array names and the % on hash names in some spots. This has issued a deprecation warning since Perl 5.0, and is no longer permitted.

"$!" text is now in English outside "use locale" scope

Previously, the text, unlike almost everything else, always came out based on the current underlying locale of the program. (Also affected on some systems is "$^E".) For programs that are unprepared to handle locale, this can cause garbage text to be displayed. It's better to display text that is translatable via some tool than garbage text which is much harder to figure out.

"$!" text will be returned in UTF-8 when appropriate

The stringification of $! and $^E will have the UTF-8 flag set when the text is actually non-ASCII UTF-8. This will enable programs that are set up to be locale-aware to properly output messages in the user's native language. Code that needs to continue the 5.20 and earlier behavior can do the stringification within the scopes of both 'use bytes' and 'use locale ":messages". No other Perl operations will be affected by locale; only $! and $^E stringification. The 'bytes' pragma causes the UTF-8 flag to not be set, just as in previous Perl releases. This resolves [perl #112208].

Support for ?PATTERN? without explicit operator has been removed

Starting regular expressions matching only once directly with the question mark delimiter is now a syntax error, so that the question mark can be available for use in new operators. Write m?PATTERN? instead, explicitly using the m operator: the question mark delimiter still invokes match-once behaviour.

defined(@array) and defined(%hash) are now fatal errors

These have been deprecated since v5.6.1 and have raised deprecation warnings since v5.16.

Using a hash or an array as a reference are now fatal errors.

For example, %foo->{"bar"} now causes a fatal compilation error. These have been deprecated since before v5.8, and have raised deprecation warnings since then.

Changes to the * prototype

The * character in a subroutine's prototype used to allow barewords to take precedence over most, but not all subroutines. It was never consistent and exhibited buggy behaviour.

Now it has been changed, so subroutines always take precedence over barewords, which brings it into conformity with similarly prototyped built-in functions:

sub splat(*) { ... }
sub foo { ... }
splat(foo); # now always splat(foo())
splat(bar); # still splat('bar') as before
close(foo); # close(foo())
close(bar); # close('bar')

Deprecations

Setting ${^ENCODING} to anything but undef

This variable allows Perl scripts to be written in a non-ASCII, non-UTF-8 encoding. However, it affects all modules globally, leading to wrong answers and segmentation faults. New scripts should be written in UTF-8; old scripts should be converted to UTF-8, which is easily done with the encoding pragma.

/\C/ character class

This character class, which matches a single byte, even if it appears in a multi-byte character has been deprecated. Matching single bytes in a multi-byte character breaks encapsulation, and can corrupt utf8 strings.

Use of non-graphic characters in single-character variable names

The syntax for single-character variable names is more lenient than for longer variable names, allowing the one-character name to be a punctuation character or even invisible (a non-graphic). Perl v5.20 deprecated the ASCII-range controls as such a name. Now, all non-graphic characters that formerly were allowed are deprecated. The practical effect of this occurs only when not under "use utf8", and affects just the C1 controls (code points 0x80 through 0xFF), NO-BREAK SPACE, and SOFT HYPHEN.

Inlining of sub () { $var } with observable side-effects

In many cases Perl makes sub () { $var } into an inlinable constant subroutine, capturing the value of $var at the time the sub expression is evaluated. This can break the closure behaviour in those cases where $var is subsequently modified. The subroutine won't return the new value.

This usage is now deprecated in those cases where the variable could be modified elsewhere. Perl detects those cases and emits a deprecation warning. Such code will likely change in the future and stop producing a constant.

If your variable is only modified in the place where it is declared, then Perl will continue to make the sub inlinable with no warnings.

sub make_constant {
    my $var = shift;
    return sub () { $var }; # fine
}

sub make_constant_deprecated {
    my $var;
    $var = shift;
    return sub () { $var }; # deprecated
}

sub make_constant_deprecated2 {
    my $var = shift;
    log_that_value($var); # could modify $var
    return sub () { $var }; # deprecated
}

In the second example above, detecting that $var is assigned to only once is too hard to detect. That it happens in a spot other than the my declaration is enough for Perl to find it suspicious.

This deprecation warning happens only for a simple variable for the body of the sub. (A BEGIN block or use statement inside the sub is ignored, because it does not become part of the sub's body.) For more complex cases, such as sub () { do_something() if 0; $var } the behaviour has changed such that inlining does not happen if the variable is modifiable elsewhere. Such cases should be rare.

Use of multiple /x regexp modifiers

It is now deprecated to say something like any of the following:

qr/foo/xx;
/(?xax:foo)/;
use re qw(/amxx);

That is, now x should only occur once in any string of contiguous regular expression pattern modifiers. We do not believe there are any occurrences of this in all of CPAN. This is in preparation for a future Perl release having /xx mean to allow white-space for readability in bracketed character classes (those enclosed in square brackets: [...]).

Using a NO-BREAK space in a character alias for \N{...} is now deprecated

This non-graphic character is essentially indistinguishable from a regular space, and so should not be allowed. See "CUSTOM ALIASES" in charnames.

A literal "{" should now be escaped in a pattern

If you want a literal left curly bracket (also called a left brace) in a regular expression pattern, you should now escape it by either preceding it with a backslash ("\{") or enclosing it within square brackets "[{]", or by using \Q; otherwise a deprecation warning will be raised. This was first announced as forthcoming in the v5.16 release; it will allow future extensions to the language to happen.

Making all warnings fatal is discouraged

The documentation for fatal warnings notes that use warnings FATAL => 'all' is discouraged and provides stronger language about the risks of fatal warnings in general.

Performance Enhancements

  • If method and class names are known at compile time, hashes are precomputed to speed up run-time method lookup. Also, compound method names like SUPER::new are parsed at compile time, to save having to parse them at run time.

  • Array and hash lookups (especially nested ones) that use only constants or simple variables as keys, are now considerably faster. See "Internal Changes" for more details.

  • (...)x1, ("constant")x0 and ($scalar)x0 are now optimised in list context. If the right-hand argument is a constant 1, the repetition operator disappears. If the right-hand argument is a constant 0, the whole expressions is optimised to the empty list, so long as the left-hand argument is a simple scalar or constant. (foo())x0 is not optimised.

  • substr assignment is now optimised into 4-argument substr at the end of a subroutine (or as the argument to return). Previously, this optimisation only happened in void context.

  • Assignment to lexical variables is often optimised away. For instance, in $lexical = chr $foo, the chr operator writes directly to the lexical variable instead of returning a value that gets copied. This optimisation has been extended to split, x and vec on the right-hand side. It has also been made to work with state variable initialization.

  • In "\L...", "\Q...", etc., the extra "stringify" op is now optimised away, making these just as fast as lcfirst, quotemeta, etc.

  • Assignment to an empty list is now sometimes faster. In particular, it never calls FETCH on tied arguments on the right-hand side, whereas it used to sometimes.

  • length is up to 20% faster for non-magical/non-tied scalars containing a string if it is a non-utf8 string or if use bytes; is in scope.

  • Non-magical/non-tied scalars that contain only a floating point value and are on most Perl builds with 64 bit integers now use 8-32 less bytes of memory depending on OS.

  • In @array = split, the assignment can be optimized away with split writing directly to the array. This optimisation was happening only for package arrays other than @_ and only sometimes. Now this optimisation happens almost all the time.

  • join is now subject to constant folding. Moreover, join with a scalar or constant for the separator and a single-item list to join is simplified to a stringification. The separator doesn't even get evaluated.

  • qq(@array) is implemented using two ops: a stringify op and a join op. If the qq contains nothing but a single array, the stringification is optimized away.

  • our $var and our($s,@a,%h) in void context are no longer evaluated at run time. Even a whole sequence of our $foo; statements will simply be skipped over. The same applies to state variables.

  • Many internal functions have been refactored to improve performance and reduce their memory footprints.

    [perl #121436] [perl #121906] [perl #121969]

  • -T and -B filetests will return sooner when an empty file is detected.

    perl #121489

  • Refactoring of pp_tied and Cpp_ref for small improvements.

  • Pathtools don't try to load XS on miniperl.

  • A typo fix reduces the size of the OP structure.

  • Hash lookups where the key is a constant is faster.

  • Subroutines with an empty prototype and bodies containing just undef are now eligible for inlining. [perl #122728]

  • Subroutines in packages no longer need to carry typeglobs around with them. Declaring a subroutine will now put a simple sub reference in the stash if possible, saving memory. The typeglobs still notionally exist, so accessing them will cause the subroutine reference to be upgraded to a typeglob. This optimization does not currently apply to XSUBs or exported subroutines, and method calls will undo it, since they cache things in typeglobs. [perl #120441]

  • The functions utf8::native_to_unicode() and utf8::unicode_to_native() (see utf8) are now optimized out on ASCII platforms. There is now not even a minimal performance hit in writing code portable between ASCII and EBCDIC platforms.

  • Win32 Perl uses 8 KB less of per-process memory than before for every perl process of this version. This data is now memory mapped from disk and shared between perl processes from the same perl binary.

Modules and Pragmata

XXX All changes to installed files in cpan/, dist/, ext/ and lib/ go here. If Module::CoreList is updated, generate an initial draft of the following sections using Porting/corelist-perldelta.pl. A paragraph summary for important changes should then be added by hand. In an ideal world, dual-life modules would have a Changes file that could be cribbed.

[ Within each section, list entries as a =item entry ]

New Modules and Pragmata

  • XXX

Updated Modules and Pragmata

  • XXX has been upgraded from version A.xx to B.yy.

Removed Modules and Pragmata

  • XXX

Documentation

New Documentation

perlunicook

This document, by Tom Christiansen, provides examples of handling Unicode in Perl.

Changes to Existing Documentation

perlapi

  • Note that SvSetSV doesn't do set magic.

  • sv_usepvn_flags - Fix documentation to mention the use of NewX instead of malloc.

    [perl #121869]

  • Clarify where NUL may be embedded or is required to terminate a string.

  • Previously missing documentation due to formatting errors are now included.

  • Entries are now organized into groups rather than by file where they are found.

  • Alphabetical sorting of entries is now handled by the POD generator to make entries easier to find when scanning.

perldata

  • The syntax of single-character variable names has been brought up-to-date and more fully explained.

perlebcdic

  • This document has been significantly updated in the light of recent improvements to EBCDIC support.

perlfunc

  • Mention that study() is currently a no-op.

  • Calling delete or exists on array values is now described as "strongly discouraged" rather than "deprecated".

  • Improve documentation of our.

  • -l now notes that it will return false if symlinks aren't supported by the file system.

    [perl #121523]

  • Note that exec LIST and system LIST may fall back to the shell on Win32. Only exec PROGRAM LIST and system PROGRAM LIST indirect object syntax will reliably avoid using the shell.

    This has also been noted in perlport.

    [perl #122046]

perlguts

  • The OOK example has been updated to account for COW changes and a change in the storage of the offset.

  • Details on C level symbols and libperl.t added.

perlhacktips

  • Documentation has been added illustrating the perils of assuming the contents of static memory pointed to by the return values of Perl wrappers for C library functions doesn't change.

  • Recommended replacements for tmpfile, atoi, strtol, and strtoul added.

  • Updated documentation for the test.valgrind make target.

    [perl #121431]

perlmodstyle

  • Instead of pointing to the module list, we are now pointing to PrePAN.

perlpolicy

  • We now have a code of conduct for the p5p mailing list, as documented in "STANDARDS OF CONDUCT" in perlpolicy.

  • The conditions for marking an experimental feature as non-experimental are now set out.

perlport

  • Out-of-date VMS-specific information has been fixed/simplified.

perlre

  • The /x modifier has been clarified to note that comments cannot be continued onto the next line by escaping them.

perlrebackslash

  • Added documentation of \b{sb}, \b{wb}, \b{gcb}, and \b{g}.

perlrecharclass

  • Clarifications have been added to "Character Ranges" in perlrecharclass to the effect that Perl guarantees that [A-Z], [a-z], [0-9] and any subranges thereof in regular expression bracketed character classes are guaranteed to match exactly what a naive English speaker would expect them to match, even on platforms (such as EBCDIC) where special handling is required to accomplish this.

  • The documentation of Bracketed Character Classes has been expanded to cover the improvements in qr/[\N{named sequence}]/ (see under "Selected Bug Fixes").

perlsec

  • Comments added on algorithmic complexity and tied hashes.

perlsyn

  • An ambiguity in the documentation of the ... statement has been corrected. [perl #122661]

  • The empty conditional in for and while is now documented in perlsyn.

perlunicode

perluniintro

  • Advice for how to make sure your strings and regular expression patterns are interpreted as Unicode has been revised to account for the new Perl 5.22 EBCDIC handling.

perlvar

  • Further clarify version number representations and usage.

perlvms

  • Out-of-date and/or incorrect material has been removed.

  • Updated documentation on environment and shell interaction in VMS.

perlxs

  • Added a discussion of locale issues in XS code.

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

  • '%s' is an unknown bound type in regex

    You used \b{...} or \B{...} and the ... is not known to Perl. The current valid ones are given in "\b{}, \b, \B{}, \B" in perlrebackslash.

  • "%s" is more clearly written simply as "%s" in regex; marked by <-- HERE in m/%s/

    (W regexp) (only under use re 'strict' or within (?[...]))

    You specified a character that has the given plainer way of writing it, and which is also portable to platforms running with different character sets.

  • Argument "%s" treated as 0 in increment (++)

    (W numeric) The indicated string was fed as an argument to the ++ operator which expects either a number or a string matching /^[a-zA-Z]*[0-9]*\z/. See "Auto-increment and Auto-decrement" in perlop for details.

  • Both or neither range ends should be Unicode in regex; marked by <-- HERE in m/%s/

    (W regexp) (only under use re 'strict' or within (?[...]))

    In a bracketed character class in a regular expression pattern, you had a range which has exactly one end of it specified using \N{}, and the other end is specified using a non-portable mechanism. Perl treats the range as a Unicode range, that is, all the characters in it are considered to be the Unicode characters, and which may be different code points on some platforms Perl runs on. For example, [\N{U+06}-\x08] is treated as if you had instead said [\N{U+06}-\N{U+08}], that is it matches the characters whose code points in Unicode are 6, 7, and 8. But that \x08 might indicate that you meant something different, so the warning gets raised.

  • Character in 'C' format overflow in pack

    (W pack) You tried converting an infinity or not-a-number to an unsigned character, which makes no sense. Perl behaved as if you tried to pack 0xFF.

  • Character in 'c' format overflow in pack

    (W pack) You tried converting an infinity or not-a-number to a signed character, which makes no sense. Perl behaved as if you tried to pack 0xFF.

  • :const is experimental

    (S experimental::const_attr) The "const" attribute is experimental. If you want to use the feature, disable the warning with no warnings 'experimental::const_attr', but know that in doing so you are taking the risk that your code may break in a future Perl version.

  • gmtime(%f) failed

    (W overflow) You called gmtime with a number that it could not handle: too large, too small, or NaN. The returned value is undef.

  • Hexadecimal float: exponent overflow

    (W overflow) The hexadecimal floating point has larger exponent than the floating point supports.

  • Hexadecimal float: exponent underflow

    (W overflow) The hexadecimal floating point has smaller exponent than the floating point supports.

  • Hexadecimal float: mantissa overflow

    (W overflow) The hexadecimal floating point literal had more bits in the mantissa (the part between the 0x and the exponent, also known as the fraction or the significand) than the floating point supports.

  • Hexadecimal float: precision loss

    (W overflow) The hexadecimal floating point had internally more digits than could be output. This can be caused by unsupported long double formats, or by 64-bit integers not being available (needed to retrieve the digits under some configurations).

    Invalid number (%f) in chr

    (W utf8) You passed an invalid number (like an infinity or not-a-number) to chr. Those are not valid character numbers, so it returned the Unicode replacement character (U+FFFD).

  • localtime(%f) failed

    (W overflow) You called localtime with a number that it could not handle: too large, too small, or NaN. The returned value is undef.

  • Negative repeat count does nothing

    (W numeric) You tried to execute the x repetition operator fewer than 0 times, which doesn't make sense.

  • NO-BREAK SPACE in a charnames alias definition is deprecated

    (D deprecated) You defined a character name which contained a no-break space character. Change it to a regular space. Usually these names are defined in the :alias import argument to use charnames, but they could be defined by a translator installed into $^H{charnames}. See "CUSTOM ALIASES" in charnames.

  • Non-finite repeat count does nothing

    (W numeric) You tried to execute the x repetition operator Inf (or -Inf) or NaN times, which doesn't make sense.

  • PerlIO layer ':win32' is experimental

    (S experimental::win32_perlio) The :win32 PerlIO layer is experimental. If you want to take the risk of using this layer, simply disable this warning:

    no warnings "experimental::win32_perlio";

  • Ranges of ASCII printables should be some subset of "0-9", "A-Z", or "a-z" in regex; marked by <-- HERE in m/%s/

    (W regexp) (only under use re 'strict' or within (?[...]))

    Stricter rules help to find typos and other errors. Perhaps you didn't even intend a range here, if the "-" was meant to be some other character, or should have been escaped (like "\-"). If you did intend a range, the one that was used is not portable between ASCII and EBCDIC platforms, and doesn't have an obvious meaning to a casual reader.

    [3-7]    # OK; Obvious and portable
[d-g]    # OK; Obvious and portable
[A-Y]    # OK; Obvious and portable
[A-z]    # WRONG; Not portable; not clear what is meant
[a-Z]    # WRONG; Not portable; not clear what is meant
[%-.]    # WRONG; Not portable; not clear what is meant
[\x41-Z] # WRONG; Not portable; not obvious to non-geek

    (You can force portability by specifying a Unicode range, which means that the endpoints are specified by \N{...}, but the meaning may still not be obvious.) The stricter rules require that ranges that start or stop with an ASCII character that is not a control have all their endpoints be the literal character, and not some escape sequence (like "\x41"), and the ranges must be all digits, or all uppercase letters, or all lowercase letters.

  • Ranges of digits should be from the same group in regex; marked by <-- HERE in m/%s/

    (W regexp) (only under use re 'strict' or within (?[...]))

    Stricter rules help to find typos and other errors. You included a range, and at least one of the end points is a decimal digit. Under the stricter rules, when this happens, both end points should be digits in the same group of 10 consecutive digits.

  • Redundant argument in %s

    (W redundant) You called a function with more arguments than other arguments you supplied indicated would be needed. Currently only emitted when a printf-type format required fewer arguments than were supplied, but might be used in the future for e.g. "pack" in perlfunc.

    The warnings category redundant is new. See also [RT #121025]

  • Use of \b{} for non-UTF-8 locale is wrong. Assuming a UTF-8 locale

    You are matching a regular expression using locale rules, and a Unicode boundary is being matched, but the locale is not a Unicode one. This doesn't make sense. Perl will continue, assuming a Unicode (UTF-8) locale, but the results could well be wrong except if the locale happens to be ISO-8859-1 (Latin1) where this message is spurious and can be ignored.

  • Using /u for '%s' instead of /%s in regex; marked by <-- HERE in m/%s/

    You used a Unicode boundary (\b{...} or \B{...}) in a portion of a regular expression where the character set modifiers /a or /aa are in effect. These two modifiers indicate an ASCII interpretation, and this doesn't make sense for a Unicode definition. The generated regular expression will compile so that the boundary uses all of Unicode. No other portion of the regular expression is affected.

  • The bitwise feature is experimental

    This warning is emitted if you use bitwise operators (& | ^ ~ &. |. ^. ~.) with the "bitwise" feature enabled. Simply suppress the warning if you want to use the feature, but know that in doing so you are taking the risk of using an experimental feature which may change or be removed in a future Perl version:

    no warnings "experimental::bitwise";
use feature "bitwise";
$x |.= $y;

  • Unescaped left brace in regex is deprecated, passed through in regex; marked by <-- HERE in m/%s/

    (D deprecated, regexp) You used a literal "{" character in a regular expression pattern. You should change to use "\{" instead, because a future version of Perl (tentatively v5.26) will consider this to be a syntax error. If the pattern delimiters are also braces, any matching right brace ("}") should also be escaped to avoid confusing the parser, for example,

    qr{abc\{def\}ghi}

  • Use of literal non-graphic characters in variable names is deprecated

  • Useless use of attribute "const"

    (W misc) The "const" attribute has no effect except on anonymous closure prototypes. You applied it to a subroutine via attributes.pm. This is only useful inside an attribute handler for an anonymous subroutine.

  • "use re 'strict'" is experimental

    (S experimental::re_strict) The things that are different when a regular expression pattern is compiled under 'strict' are subject to change in future Perl releases in incompatible ways. This means that a pattern that compiles today may not in a future Perl release. This warning is to alert you to that risk.

  • Warning: unable to close filehandle %s properly: %s

  • Wide character (U+%X) in %s

    (W locale) While in a single-byte locale (i.e., a non-UTF-8 one), a multi-byte character was encountered. Perl considers this character to be the specified Unicode code point. Combining non-UTF8 locales and Unicode is dangerous. Almost certainly some characters will have two different representations. For example, in the ISO 8859-7 (Greek) locale, the code point 0xC3 represents a Capital Gamma. But so also does 0x393. This will make string comparisons unreliable.

    You likely need to figure out how this multi-byte character got mixed up with your single-byte locale (or perhaps you thought you had a UTF-8 locale, but Perl disagrees).

  • The following two warnings for tr/// used to be skipped if the transliteration contained wide characters, but now they occur regardless of whether there are wide characters or not:

    Useless use of /d modifier in transliteration operator

    Replacement list is longer than search list

  • A new locale warning category has been created, with the following warning messages currently in it:

Changes to Existing Diagnostics

Diagnostic Removals

  • "Ambiguous use of -foo resolved as -&foo()"

    There is actually no ambiguity here, and this impedes the use of negated constants; e.g., -Inf.

  • "Constant is not a FOO reference"

    Compile-time checking of constant dereferencing (e.g., my_constant->()) has been removed, since it was not taking overloading into account. [perl #69456] [perl #122607]

Utility Changes

x2p/

  • The x2p/ directory has been removed from the Perl core.

    This removes find2perl, s2p and a2p. They have all been released to CPAN as separate distributions (App::find2perl, App::s2p, App::a2p).

h2ph

  • h2ph now handles hexadecimal constants in the compiler's predefined macro definitions, as visible in $Config{cppsymbols}. [rt.perl.org #123784]

encguess

  • No longer depends on non-core module anymore.

Configuration and Compilation

  • Configure now checks for lrintl, lroundl, llrintl, and llroundl.

  • Configure with -Dmksymlinks should now be faster. [perl #122002]

  • pthreads and lcl will be linked by default if present. This allows XS modules that require threading to work on non-threaded perls. Note that you must still pass -Dusethreads if you want a threaded perl.

  • For long doubles (to get more precision and range for floating point numbers) one can now use the GCC quadmath library which implements the quadruple precision floating point numbers in x86 and ia64 platforms. See INSTALL for details.

  • MurmurHash64A and MurmurHash64B can now be configured as the internal hash function.

  • make test.valgrind now supports parallel testing.

    For example:

    TEST_JOBS=9 make test.valgrind

    See "valgrind" in perlhacktips for more information.

    [perl #121431]

  • The MAD (Misc Attribute Decoration) build option has been removed

    This was an unmaintained attempt at preserving the Perl parse tree more faithfully so that automatic conversion of Perl 5 to Perl 6 would have been easier.

    This build-time configuration option had been unmaintained for years, and had probably seriously diverged on both Perl 5 and Perl 6 sides.

  • A new compilation flag, -DPERL_OP_PARENT is available. For details, see the discussion below at "Internal Changes".

Testing

  • t/porting/re_context.t has been added to test that utf8 and its dependencies only use the subset of the $1..$n capture vars that Perl_save_re_context() is hard-coded to localize, because that function has no efficient way of determining at runtime what vars to localize.

  • Tests for performance issues have been added in the file t/perf/taint.t.

  • Some regular expression tests are written in such a way that they will run very slowly if certain optimizations break. These tests have been moved into new files, t/re/speed.t and t/re/speed_thr.t, and are run with a watchdog().

  • test.pl now allows plan skip_all => $reason, to make it more compatible with Test::More.

  • A new test script, op/infnan.t, has been added to test if Inf and NaN are working correctly. See "Infinity and NaN (not-a-number) handling improved".

Platform Support

Regained Platforms

IRIX and Tru64 platforms are working again.

(Some make test failures remain.)

z/OS running EBCDIC Code Page 1047

Core perl now works on this EBCDIC platform. Earlier perls also worked, but, even though support wasn't officially withdrawn, recent perls would not compile and run well. Perl 5.20 would work, but had many bugs which have now been fixed. Many CPAN modules that ship with Perl still fail tests, including Pod::Simple. However the version of Pod::Simple currently on CPAN should work; it was fixed too late to include in Perl 5.22. Work is under way to fix many of the still-broken CPAN modules, which likely will be installed on CPAN when completed, so that you may not have to wait until Perl 5.24 to get a working version.

Discontinued Platforms

NeXTSTEP/OPENSTEP

NeXTSTEP was proprietary OS bundled with NeXT's workstations in the early to mid 90's; OPENSTEP was an API specification that provided a NeXTSTEP-like environment on a non-NeXTSTEP system. Both are now long dead, so support for building Perl on them has been removed.

Platform-Specific Notes

EBCDIC

Special handling is required on EBCDIC platforms to get qr/[i-j]/ to match only "i" and "j", since there are 7 characters between the code points for "i" and "j". This special handling had only been invoked when both ends of the range are literals. Now it is also invoked if any of the \N{...} forms for specifying a character by name or Unicode code point is used instead of a literal. See "Character Ranges" in perlrecharclass.

HP-UX

The archname now distinguishes use64bitint from use64bitall.

Android

Build support has been improved for cross-compiling in general and for Android in particular.

VMS

  • When spawning a subprocess without waiting, the return value is now the correct PID.

  • Fix a prototype so linking doesn't fail under the VMS C++ compiler.

  • finite, finitel, and isfinite detection has been added to configure.com, environment handling has had some minor changes, and a fix for legacy feature checking status.

Win32

  • miniperl.exe is now built with -fno-strict-aliasing, allowing 64-bit builds to complete on GCC 4.8. [perl #123976]

  • test-prep again depends on test-prep-gcc for GCC builds. [perl #124221]

  • Perl can now be built in C++ mode on Windows by setting the makefile macro USE_CPLUSPLUS to the value "define".

  • List form pipe open no longer falls back to the shell.

  • In release 5.21.8 compiling on VC with dmake was broken. Fixed.

  • New DebugSymbols and DebugFull configuration options added to Windows makefiles.

  • B now compiles again on Windows.

  • Previously, on Visual C++ for Win64 built Perls only, when compiling every Perl XS module (including CPAN ones) and Perl aware .c file with a 64 bit Visual C++, would unconditionally have around a dozen warnings from hv_func.h. These warnings have been silenced. GCC all bitness and Visual C++ for Win32 were not affected.

  • Support for building without PerlIO has been removed from the Windows makefiles. Non-PerlIO builds were all but deprecated in Perl 5.18.0 and are already not supported by Configure on POSIX systems.

  • Between 2 and 6 ms and 7 I/O calls have been saved per attempt to open a perl module for each path in @INC.

  • Intel C builds are now always built with C99 mode on.

  • %I64d is now being used instead of %lld for MinGW.

  • In the experimental :win32 layer, a crash in open was fixed. Also opening /dev/null, which works the Win32 Perl's normal :unix layer, was implemented for :win32. [perl #122224]

  • A new makefile option, USE_LONG_DOUBLE, has been added to the Windows dmake makefile for gcc builds only. Set this to "define" if you want perl to use long doubles to give more accuracy and range for floating point numbers.

OpenBSD

On OpenBSD, Perl will now default to using the system malloc due to the security features it provides. Perl's own malloc wrapper has been in use since v5.14 due to performance reasons, but the OpenBSD project believes the tradeoff is worth it and would prefer that users who need the speed specifically ask for it.

[perl #122000].

Solaris

  • We now look for the Sun Studio compiler in both /opt/solstudio* and /opt/solarisstudio*.

  • Builds on Solaris 10 with -Dusedtrace would fail early since make didn't follow implied dependencies to build perldtrace.h. Added an explicit dependency to depend. [perl #120120]

  • c99 options have been cleaned up, hints look for solstudio as well as SUNWspro, and support for native setenv has been added.

Internal Changes

  • Perl 5.21.2 introduced a new build option, -DPERL_OP_PARENT, which causes the last op_sibling pointer to refer back to the parent rather than being NULL, and where instead a new flag indicates the end of the chain. In this release, the new implementation has been revised; in particular:

    • On PERL_OP_PARENT builds, the op_sibling field has been renamed op_sibparent to reflect its new dual purpose. Since the intention is that this field should primarily be accessed via macros, this change should be transparent for code written to work under PERL_OP_PARENT.

    • The newly-introduced op_lastsib flag bit has been renamed op_moresib and its logic inverted; i.e. it is initialised to zero in a new op, and is changed to 1 when an op gains a sibling.

    • The function Perl_op_parent is now only available on PERL_OP_PARENT builds. Using it on a plain build will be a compile-timer error.

    • Three new macros, OpMORESIB_set, OpLASTSIB_set, OpMAYBESIB_set have been added, which are intended to be a low-level portable way to set op_sibling / op_sibparent while also updating op_moresib. The first sets the sibling pointer to a new sibling, the second makes the op the last sibling, and the third conditionally does the first or second action. The op_sibling_splice() function is retained as a higher-level interface that can also maintain consistency in the parent at the same time (e.g. by updating op_first and op_last where appropriate).

    • The macro OpSIBLING_set, added in Perl 5.21.2, has been removed. It didn't manipulate op_moresib and has been superseded by OpMORESIB_set et al.

    • The op_sibling_splice function now accepts a null parent argument where the splicing doesn't affect the first or last ops in the sibling chain, and thus where the parent doesn't need to be updated accordingly.

  • Macros have been created to allow XS code to better manipulate the POSIX locale category LC_NUMERIC. See "Locale-related functions and macros" in perlapi.

  • The previous atoi et al replacement function, grok_atou, has now been superseded by grok_atoUV. See perlclib for details.

  • Added Perl_sv_get_backrefs() to determine if an SV is a weak-referent.

    Function either returns an SV * of type AV, which contains the set of weakreferences which reference the passed in SV, or a simple RV * which is the only weakref to this item.

  • screaminstr has been removed. Although marked as public API, it is undocumented and has no usage in modern perl versions on CPAN Grep. Calling it has been fatal since 5.17.0.

  • newDEFSVOP, block_start, block_end and intro_my have been added to the API.

  • The internal convert function in op.c has been renamed op_convert_list and added to the API.

  • sv_magic no longer forbids "ext" magic on read-only values. After all, perl can't know whether the custom magic will modify the SV or not. [perl #123103]

  • Starting in 5.21.6, accessing "CvPADLIST" in perlapi in an XSUB is forbidden. CvPADLIST has be reused for a different internal purpose for XSUBs. Guard all CvPADLIST expressions with CvISXSUB() if your code doesn't already block XSUB CV*s from going through optree CV* expecting code.

  • SVs of type SVt_NV are now bodyless when a build configure and platform allow it, specifically sizeof(NV) <= sizeof(IV). The bodyless trick is the same one as for IVs since 5.9.2, but for NVs, unlike IVs, is not guaranteed on all platforms and build configurations.

  • The $DB::single, $DB::signal and $DB::trace now have set and get magic that stores their values as IVs and those IVs are used when testing their values in pp_dbstate. This prevents perl from recursing infinity if an overloaded object is assigned to any of those variables. [perl #122445]

  • Perl_tmps_grow which is marked as public API but undocumented has been removed from public API. If you use EXTEND_MORTAL macro in your XS code to preextend the mortal stack, you are unaffected by this change.

  • cv_name, which was introduced in 5.21.4, has been changed incompatibly. It now has a flags field that allows the caller to specify whether the name should be fully qualified. See "cv_name" in perlapi.

  • Internally Perl no longer uses the SVs_PADMY flag. SvPADMY() now returns a true value for anything not marked PADTMP. SVs_PADMY is now defined as 0.

  • The macros SETsv and SETsvUN have been removed. They were no longer used in the core since commit 6f1401dc2a, and have not been found present on CPAN.

  • The SvFAKE bit (unused on HVs) got informally reserved by David Mitchell for future work on vtables.

  • The sv_catpvn_flags function accepts SV_CATBYTES and SV_CATUTF8 flags, which specify whether the appended string is bytes or utf8, respectively.

  • A new opcode class, METHOP has been introduced, which holds class/method related info needed at runtime to improve performance of class/object method calls.

    OP_METHOD and OP_METHOD_NAMED are moved from being UNOP/SVOP to being METHOP.

  • save_re_context no longer does anything and has been moved to mathoms.c.

  • cv_name is a new API function that can be passed a CV or GV. It returns an SV containing the name of the subroutine for use in diagnostics. [perl #116735] [perl #120441]

  • cv_set_call_checker_flags is a new API function that works like cv_set_call_checker, except that it allows the caller to specify whether the call checker requires a full GV for reporting the subroutine's name, or whether it could be passed a CV instead. Whatever value is passed will be acceptable to cv_name. cv_set_call_checker guarantees there will be a GV, but it may have to create one on the fly, which is inefficient. [perl #116735]

  • CvGV (which is not part of the API) is now a more complex macro, which may call a function and reify a GV. For those cases where is has been used as a boolean, CvHASGV has been added, which will return true for CVs that notionally have GVs, but without reifying the GV. CvGV also returns a GV now for lexical subs. [perl #120441]

  • Added "sync_locale" in perlapi. Changing the program's locale should be avoided by XS code. Nevertheless, certain non-Perl libraries called from XS, such as Gtk do so. When this happens, Perl needs to be told that the locale has changed. Use this function to do so, before returning to Perl.

  • The defines and labels for the flags in the op_private field of OPs are now auto-generated from data in regen/op_private. The noticeable effect of this is that some of the flag output of Concise might differ slightly, and the flag output of perl -Dx may differ considerably (they both use the same set of labels now). Also in debugging builds, there is a new assert in op_free() that checks that the op doesn't have any unrecognized flags set in op_private.

  • Added "sync_locale" in perlapi. Changing the program's locale should be avoided by XS code. Nevertheless, certain non-Perl libraries called from XS, such as Gtk do so. When this happens, Perl needs to be told that the locale has changed. Use this function to do so, before returning to Perl.

  • The deprecated variable PL_sv_objcount has been removed.

  • Perl now tries to keep the locale category LC_NUMERIC set to "C" except around operations that need it to be set to the program's underlying locale. This protects the many XS modules that cannot cope with the decimal radix character not being a dot. Prior to this release, Perl initialized this category to "C", but a call to POSIX::setlocale() would change it. Now such a call will change the underlying locale of the LC_NUMERIC category for the program, but the locale exposed to XS code will remain "C". There is an API under development for those relatively few modules that need to use the underlying locale. This API will be nailed down during the course of developing v5.21. Send email to mailto:perl5-porters@perl.org for guidance.

  • A new macro isUTF8_CHAR has been written which efficiently determines if the string given by its parameters begins with a well-formed UTF-8 encoded character.

  • The following private API functions had their context parameter removed, Perl_cast_ulong, Perl_cast_i32, Perl_cast_iv, Perl_cast_uv, Perl_cv_const_sv, Perl_mg_find, Perl_mg_findext, Perl_mg_magical, Perl_mini_mktime, Perl_my_dirfd, Perl_sv_backoff, Perl_utf8_hop.

    Users of the public API prefix-less calls remain unaffected.

  • Experimental support for ops in the optree to be able to locate their parent, if any. A general-purpose function, op_sibling_splice() allows for general manipulating an op_sibling chain. The last op in such a chain is now marked with the field op_lastsib.

    A new build define, -DPERL_OP_PARENT has been added; if given, it forces the core to use op_lastsib to detect the last sibling in a chain, freeing the last op_sibling pointer, which then points back to the parent (instead of being NULL).

    A C-level op_parent() function, and a B parent() method have been added; under a default build, they return NULL, but when -DPERL_OP_PARENT has been set, they return the parent of the current op.

  • The PADNAME and PADNAMELIST types are now separate types, and no longer simply aliases for SV and AV. [perl #123223]

  • Pad names are now always UTF8. The PadnameUTF8 macro always returns true. Previously, this was effectively the case already, but any support for two different internal representations of pad names has now been removed.

  • The OP_SIBLING and OP_HAS_SIBLING macros added in an earlier 5.21.x release have been renamed OpSIBLING and OpHAS_SIBLING, following the existing convention.

  • A new op class, UNOP_AUX, has been added. This is a subclass of UNOP with an op_aux field added, which points to an array of unions of UV, SV* etc. It is intended for where an op needs to store more data than a simple op_sv or whatever. Currently the only op of this type is OP_MULTIDEREF (see below).

  • A new op has been added, OP_MULTIDEREF, which performs one or more nested array and hash lookups where the key is a constant or simple variable. For example the expression $a[0]{$k}[$i], which previously involved ten rv2Xv, Xelem, gvsv and const ops is now performed by a single multideref op. It can also handle local, exists and delete. A non-simple index expression, such as [$i+1] is still done using aelem/helem, and single-level array lookup with a small constant index is still done using aelemfast.

Selected Bug Fixes

  • pack("D", $x) and pack("F", $x) now zero the padding on x86 long double builds. GCC 4.8 and later, under some build options, would either overwrite the zero-initialized padding, or bypass the initialized buffer entirely. This caused op/pack.t to fail. [perl #123971]

  • Extending an array cloned from a parent thread could result in "Modification of a read-only value attempted" errors when attempting to modify the new elements. [perl #124127]

  • An assertion failure and subsequent crash with *x=<y> has been fixed. [perl #123790]

  • An optimization for state variable initialization introduced in Perl 5.21.6 has been reverted because it was found to exacerbate some other existing buggy behaviour. [perl #124160]

  • The extension of another optimization to cover more ops in Perl 5.21 has also been reverted to its Perl 5.20 state as a temporary fix for regression issues that it caused. [perl #123790]

  • New bitwise ops added in Perl 5.21.9 accidentally caused $^H |= 0x1c020000 to enable all features. This has now been fixed.

  • A possible crashing/looping bug has been fixed. [perl #124099]

  • UTF-8 variable names used in array indexes, unquoted UTF-8 HERE-document terminators and UTF-8 function names all now work correctly. [perl #124113]

  • Repeated global pattern matches in scalar context on large tainted strings were exponentially slow depending on the current match position in the string. [perl #123202]

  • Various crashes due to the parser getting confused by syntax errors have been fixed. [perl #123801] [perl #123802] [perl #123955] [perl #123995]

  • split in the scope of lexical $_ has been fixed not to fail assertions. [perl #123763]

  • my $x : attr syntax inside various list operators no longer fails assertions. [perl #123817]

  • An @ sign in quotes followed by a non-ASCII digit (which is not a valid identifier) would cause the parser to crash, instead of simply trying the @ as literal. This has been fixed. [perl #123963]

  • *bar::=*foo::=*glob_with_hash has been crashing since Perl 5.14, but no longer does. [perl #123847]

  • foreach in scalar context was not pushing an item on to the stack, resulting in bugs. (print 4, scalar do { foreach(@x){} } + 1 would print 5.) It has been fixed to return undef. [perl #124004]

  • A memory leak introduced in Perl 5.21.6 has been fixed. [perl #123922]

  • A regression in the behaviour of the readline built-in function, caused by the introduction of the <<>> operator, has been fixed. [perl #123990]

  • Several cases of data used to store environment variable contents in core C code being potentially overwritten before being used have been fixed. [perl #123748]

  • Patterns starting with /.*/ are now fast again. [rt.perl.org #123743]

  • The original visible value of $/ is now preserved when it is set to an invalid value. Previously if you set $/ to a reference to an array, for example, perl would produce a runtime error and not set PL_rs, but perl code that checked $/ would see the array reference. [rt.perl.org #123218]

  • In a regular expression pattern, a POSIX class, like [:ascii:], must be inside a bracketed character class, like /qr[[:ascii:]]. A warning is issued when something looking like a POSIX class is not inside a bracketed class. That warning wasn't getting generated when the POSIX class was negated: [:^ascii:]. This is now fixed.

  • Fix a couple of other size calculation overflows. [rt.perl.org #123554]

  • A bug introduced in 5.21.6, dump LABEL acted the same as goto LABEL. This has been fixed. [rt.perl.org #123836]

  • Perl 5.14.0 introduced a bug whereby eval { LABEL: } would crash. This has been fixed. [rt.perl.org #123652]

  • Various crashes due to the parser getting confused by syntax errors have been fixed. [rt.perl.org #123617] [rt.perl.org #123737] [rt.perl.org #123753] [rt.perl.org #123677]

  • Code like /$a[/ used to read the next line of input and treat it as though it came immediately after the opening bracket. Some invalid code consequently would parse and run, but some code caused crashes, so this is now disallowed. [rt.perl.org #123712]

  • Fix argument underflow for pack. [rt.perl.org #123874]

  • Fix handling of non-strict \x{}. Now \x{} is equivalent to \x{0} instead of faulting.

  • stat -t is now no longer treated as stackable, just like -t stat. [rt.perl.org #123816]

  • The following no longer causes a SEGV: qr{x+(y(?0))*}.

  • Fixed infinite loop in parsing backrefs in regexp patterns.

  • Several minor bug fixes in behavior of Inf and NaN, including warnings when stringifying Inf-like or NaN-like strings. For example, "NaNcy" doesn't numify to NaN anymore.

  • Only stringy classnames are now shared. This fixes some failures in autobox. [rt.cpan.org #100819]

  • A bug in regular expression patterns that could lead to segfaults and other crashes has been fixed. This occurred only in patterns compiled with "/i", while taking into account the current POSIX locale (this usually means they have to be compiled within the scope of "use locale"), and there must be a string of at least 128 consecutive bytes to match. [perl #123539]

  • s/// now works on very long strings instead of dying with 'Substitution loop'. [perl #103260] [perl #123071]

  • gmtime no longer crashes with not-a-number values. [perl #123495]

  • \() (reference to an empty list) and y/// with lexical $_ in scope could do a bad write past the end of the stack. They have been fixed to extend the stack first.

  • prototype() with no arguments used to read the previous item on the stack, so print "foo", prototype() would print foo's prototype. It has been fixed to infer $_ instead. [perl #123514]

  • Some cases of lexical state subs inside predeclared subs could crash but no longer do.

  • Some cases of nested lexical state subs inside anonymous subs could cause 'Bizarre copy' errors or possibly even crash.

  • When trying to emit warnings, perl's default debugger (perl5db.pl) was sometimes giving 'Undefined subroutine &DB::db_warn called' instead. This bug, which started to occur in Perl 5.18, has been fixed. [perl #123553]

  • Certain syntax errors in substitutions, such as s/${<>{})//, would crash, and had done so since Perl 5.10. (In some cases the crash did not start happening till 5.16.) The crash has, of course, been fixed. [perl #123542]

  • A repeat expression like 33 x ~3 could cause a large buffer overflow since the new output buffer size was not correctly handled by SvGROW(). An expression like this now properly produces a memory wrap panic. [perl 123554]

  • formline("@...", "a"); would crash. The FF_CHECKNL case in pp_formline() didn't set the pointer used to mark the chop position, which led to the FF_MORE case crashing with a segmentation fault. This has been fixed. [perl #123538]

  • A possible buffer overrun and crash when parsing a literal pattern during regular expression compilation has been fixed. [perl #123604]

  • fchmod() and futimes() now set $! when they fail due to being passed a closed file handle. [perl #122703]

  • Perl now comes with a corrected Unicode 7.0 for the erratum issued on October 21, 2014 (see http://www.unicode.org/errata/#current_errata), dealing with glyph shaping in Arabic.

  • op_free() no longer crashes due to a stack overflow when freeing a deeply recursive op tree. [perl #108276]

  • scalarvoid() would crash due to a stack overflow when processing a deeply recursive op tree. [perl #108276]

  • In Perl 5.20.0, $^N accidentally had the internal UTF8 flag turned off if accessed from a code block within a regular expression, effectively UTF8-encoding the value. This has been fixed. [perl #123135]

  • A failed semctl call no longer overwrites existing items on the stack, causing (semctl(-1,0,0,0))[0] to give an "uninitialized" warning.

  • else{foo()} with no space before foo is now better at assigning the right line number to that statement. [perl #122695]

  • Sometimes the assignment in @array = split gets optimised and split itself writes directly to the array. This caused a bug, preventing this assignment from being used in lvalue context. So (@a=split//,"foo")=bar() was an error. (This bug probably goes back to Perl 3, when the optimisation was added.) This optimisation, and the bug, started to happen in more cases in 5.21.5. It has now been fixed. [perl #123057]

  • When argument lists that fail the checks installed by subroutine signatures, the resulting error messages now give the file and line number of the caller, not of the called subroutine. [perl #121374]

  • Flip-flop operators (.. and ... in scalar context) used to maintain a separate state for each recursion level (the number of times the enclosing sub was called recursively), contrary to the documentation. Now each closure has one internal state for each flip-flop. [perl #122829]

  • use, no, statement labels, special blocks (BEGIN) and pod are now permitted as the first thing in a map or grep block, the block after print or say (or other functions) returning a handle, and within ${...}, @{...}, etc. [perl #122782]

  • The repetition operator x now propagates lvalue context to its left-hand argument when used in contexts like foreach. That allows for(($#that_array)x2) { ... } to work as expected if the loop modifies $_.

  • (...) x ... in scalar context used to corrupt the stack if one operand were an object with "x" overloading, causing erratic behaviour. [perl #121827]

  • Assignment to a lexical scalar is often optimised away (as mentioned under "Performance Enhancements"). Various bugs related to this optimisation have been fixed. Certain operators on the right-hand side would sometimes fail to assign the value at all or assign the wrong value, or would call STORE twice or not at all on tied variables. The operators affected were $foo++, $foo--, and -$foo under use integer, chomp, chr and setpgrp.

  • List assignments were sometimes buggy if the same scalar ended up on both sides of the assignment due to used of tied, values or each. The result would be the wrong value getting assigned.

  • setpgrp($nonzero) (with one argument) was accidentally changed in 5.16 to mean setpgrp(0). This has been fixed.

  • __SUB__ could return the wrong value or even corrupt memory under the debugger (the -d switch) and in subs containing eval $string.

  • When sub () { $var } becomes inlinable, it now returns a different scalar each time, just as a non-inlinable sub would, though Perl still optimises the copy away in cases where it would make no observable difference.

  • my sub f () { $var } and sub () : attr { $var } are no longer eligible for inlining. The former would crash; the latter would just throw the attributes away. An exception is made for the little-known ":method" attribute, which does nothing much.

  • Inlining of subs with an empty prototype is now more consistent than before. Previously, a sub with multiple statements, all but the last optimised away, would be inlinable only if it were an anonymous sub containing a string eval or state declaration or closing over an outer lexical variable (or any anonymous sub under the debugger). Now any sub that gets folded to a single constant after statements have been optimised away is eligible for inlining. This applies to things like sub () { jabber() if DEBUG; 42 }.

    Some subroutines with an explicit return were being made inlinable, contrary to the documentation, Now return always prevents inlining.

  • On some systems, such as VMS, crypt can return a non-ASCII string. If a scalar assigned to had contained a UTF8 string previously, then crypt would not turn off the UTF8 flag, thus corrupting the return value. This would happen with $lexical = crypt ....

  • crypt no longer calls FETCH twice on a tied first argument.

  • An unterminated here-doc on the last line of a quote-like operator (qq[${ <<END }], /(?{ <<END })/) no longer causes a double free. It started doing so in 5.18.

  • Fixed two assertion failures introduced into -DPERL_OP_PARENT builds. [perl #108276]

  • index() and rindex() no longer crash when used on strings over 2GB in size. [perl #121562].

  • A small previously intentional memory leak in PERL_SYS_INIT/PERL_SYS_INIT3 on Win32 builds was fixed. This might affect embedders who repeatedly create and destroy perl engines within the same process.

  • POSIX::localeconv() now returns the data for the program's underlying locale even when called from outside the scope of use locale.

  • POSIX::localeconv() now works properly on platforms which don't have LC_NUMERIC and/or LC_MONETARY, or for which Perl has been compiled to disregard either or both of these locale categories. In such circumstances, there are now no entries for the corresponding values in the hash returned by localeconv().

  • POSIX::localeconv() now marks appropriately the values it returns as UTF-8 or not. Previously they were always returned as a bytes, even if they were supposed to be encoded as UTF-8.

  • On Microsoft Windows, within the scope of use locale, the following POSIX character classes gave results for many locales that did not conform to the POSIX standard: [[:alnum:]], [[:alpha:]], [[:blank:]], [[:digit:]], [[:graph:]], [[:lower:]], [[:print:]], [[:punct:]], [[:upper:]], [[:word:]], and [[:xdigit:]]. These are because the underlying Microsoft implementation does not follow the standard. Perl now takes special precautions to correct for this.

  • Many issues have been detected by Coverity and fixed.

  • system() and friends should now work properly on more Android builds.

    Due to an oversight, the value specified through -Dtargetsh to Configure would end up being ignored by some of the build process. This caused perls cross-compiled for Android to end up with defective versions of system(), exec() and backticks: the commands would end up looking for /bin/sh instead of /system/bin/sh, and so would fail for the vast majority of devices, leaving $! as ENOENT.

  • qr(...\(...\)...), qr[...\[...\]...], and qr{...\{...\}...} now work. Previously it was impossible to escape these three left-characters with a backslash within a regular expression pattern where otherwise they would be considered metacharacters, and the pattern opening delimiter was the character, and the closing delimiter was its mirror character.

  • s///e on tainted utf8 strings got pos() messed up. This bug, introduced in 5.20, is now fixed. [RT #122148]

  • A non-word boundary in a regular expression (\B) did not always match the end of the string; in particular q{} =~ /\B/ did not match. This bug, introduced in perl 5.14, is now fixed. [RT #122090]

  • " P" =~ /(?=.*P)P/ should match, but did not. This is now fixed. [RT #122171].

  • Failing to compile use Foo in an eval could leave a spurious BEGIN subroutine definition, which would produce a "Subroutine BEGIN redefined" warning on the next use of use, or other BEGIN block. [perl #122107]

  • method { BLOCK } ARGS syntax now correctly parses the arguments if they begin with an opening brace. [perl #46947]

  • External libraries and Perl may have different ideas of what the locale is. This is problematic when parsing version strings if the locale's numeric separator has been changed. Version parsing has been patched to ensure it handles the locales correctly. [perl #121930]

  • A bug has been fixed where zero-length assertions and code blocks inside of a regex could cause pos to see an incorrect value. [perl #122460]

  • Constant dereferencing now works correctly for typeglob constants. Previously the glob was stringified and its name looked up. Now the glob itself is used. [perl #69456]

  • When parsing a funny character ($ @ % &) followed by braces, the parser no longer tries to guess whether it is a block or a hash constructor (causing a syntax error when it guesses the latter), since it can only be a block.

  • undef $reference now frees the referent immediately, instead of hanging on to it until the next statement. [perl #122556]

  • Various cases where the name of a sub is used (autoload, overloading, error messages) used to crash for lexical subs, but have been fixed.

  • Bareword lookup now tries to avoid vivifying packages if it turns out the bareword is not going to be a subroutine name.

  • Compilation of anonymous constants (e.g., sub () { 3 }) no longer deletes any subroutine named __ANON__ in the current package. Not only was *__ANON__{CODE} cleared, but there was a memory leak, too. This bug goes back to Perl 5.8.0.

  • Stub declarations like sub f; and sub f (); no longer wipe out constants of the same name declared by use constant. This bug was introduced in Perl 5.10.0.

  • Under some conditions a warning raised in compilation of regular expression patterns could be displayed multiple times. This is now fixed.

  • qr/[\N{named sequence}]/ now works properly in many instances. Some names known to \N{...} refer to a sequence of multiple characters, instead of the usual single character. Bracketed character classes generally only match single characters, but now special handling has been added so that they can match named sequences, but not if the class is inverted or the sequence is specified as the beginning or end of a range. In these cases, the only behavior change from before is a slight rewording of the fatal error message given when this class is part of a ?[...]) construct. When the [...] stands alone, the same non-fatal warning as before is raised, and only the first character in the sequence is used, again just as before.

  • Tainted constants evaluated at compile time no longer cause unrelated statements to become tainted. [perl #122669]

  • open $$fh, ..., which vivifies a handle with a name like "main::_GEN_0", was not giving the handle the right reference count, so a double free could happen.

  • When deciding that a bareword was a method name, the parser would get confused if an "our" sub with the same name existed, and look up the method in the package of the "our" sub, instead of the package of the invocant.

  • The parser no longer gets confused by \U= within a double-quoted string. It used to produce a syntax error, but now compiles it correctly. [perl #80368]

  • It has always been the intention for the -B and -T file test operators to treat UTF-8 encoded files as text. (perlfunc has been updated to say this.) Previously, it was possible for some files to be considered UTF-8 that actually weren't valid UTF-8. This is now fixed. The operators now work on EBCDIC platforms as well.

  • Under some conditions warning messages raised during regular expression pattern compilation were being output more than once. This has now been fixed.

  • A regression has been fixed that was introduced in Perl 5.20.0 (fixed in Perl 5.20.1 as well as here) in which a UTF-8 encoded regular expression pattern that contains a single ASCII lowercase letter does not match its uppercase counterpart. [perl #122655]

  • Constant folding could incorrectly suppress warnings if lexical warnings (use warnings or no warnings) were not in effect and $^W were false at compile time and true at run time.

  • Loading UTF8 tables during a regular expression match could cause assertion failures under debugging builds if the previous match used the very same regular expression. [perl #122747]

  • Thread cloning used to work incorrectly for lexical subs, possibly causing crashes or double frees on exit.

  • Since Perl 5.14.0, deleting $SomePackage::{__ANON__} and then undefining an anonymous subroutine could corrupt things internally, resulting in Devel::Peek crashing or B.pm giving nonsensical data. This has been fixed.

  • (caller $n)[3] now reports names of lexical subs, instead of treating them as "(unknown)".

  • sort subname LIST now supports lexical subs for the comparison routine.

  • Aliasing (e.g., via *x = *y) could confuse list assignments that mention the two names for the same variable on either side, causing wrong values to be assigned. [perl #15667]

  • Long here-doc terminators could cause a bad read on short lines of input. This has been fixed. It is doubtful that any crash could have occurred. This bug goes back to when here-docs were introduced in Perl 3.000 twenty-five years ago.

  • An optimization in split to treat split/^/ like split/^/m had the unfortunate side-effect of also treating split/\A/ like split/^/m, which it should not. This has been fixed. (Note, however, that split/^x/ does not behave like split/^x/m, which is also considered to be a bug and will be fixed in a future version.) [perl #122761]

  • The little-known my Class $var syntax (see fields and attributes) could get confused in the scope of use utf8 if Class were a constant whose value contained Latin-1 characters.

  • Locking and unlocking values via Hash::Util or Internals::SvREADONLY no longer has any effect on values that are read-only to begin. Previously, unlocking such values could result in crashes, hangs or other erratic behaviour.

  • The internal looks_like_number function (which Scalar::Util provides access to) began erroneously to return true for "-e1" in 5.21.4, affecting also -'-e1'. This has been fixed.

  • The flip-flop operator (.. in scalar context) would return the same scalar each time, unless the containing subroutine was called recursively. Now it always returns a new scalar. [perl #122829]

  • Some unterminated (?(...)...) constructs in regular expressions would either crash or give erroneous error messages. /(?(1)/ is one such example.

  • pack "w", $tied no longer calls FETCH twice.

  • List assignments like ($x, $z) = (1, $y) now work correctly if $x and $y have been aliased by foreach.

  • Some patterns including code blocks with syntax errors, such as / (?{(^{})/, would hang or fail assertions on debugging builds. Now they produce errors.

  • An assertion failure when parsing sort with debugging enabled has been fixed. [perl #122771]

  • *a = *b; @a = split //, $b[1] could do a bad read and produce junk results.

  • In () = @array = split, the () = at the beginning no longer confuses the optimizer, making it assume a limit of 1.

  • Fatal warnings no longer prevent the output of syntax errors. [perl #122966]

  • Fixed a NaN double to long double conversion error on VMS. For quiet NaNs (and only on Itanium, not Alpha) negative infinity instead of NaN was produced.

  • Fixed the issue that caused make distclean to leave files behind that shouldn't. [perl #122820]

  • AIX now sets the length in getsockopt correctly. [perl #120835], [rt #91183], [rt #85570].

  • During the pattern optimization phase, we no longer recurse into GOSUB/GOSTART when not SCF_DO_SUBSTR. This prevents the optimizer to run "forever" and exhaust all memory. [perl #122283]

  • t/op/crypt.t now performs SHA-256 algorithm if the default one is disabled. [perl #121591]

  • Fixed an off-by-one error when setting the size of shared array. [perl #122950]

  • Fixed a bug that could cause perl to execute an infinite loop during compilation. [perl #122995]

  • On Win32, restoring in a child pseudo-process a variable that was local()ed in a parent pseudo-process before the fork happened caused memory corruption and a crash in the child pseudo-process (and therefore OS process). [perl #40565]

  • Calling write on a format with a ^** field could produce a panic in sv_chop() if there were insufficient arguments or if the variable used to fill the field was empty. [perl #123245]

  • Non-ASCII lexical sub names (use in error messages) on longer have extra junk on the end.

  • The \@ subroutine prototype no longer flattens parenthesized arrays (taking a reference to each element), but takes a reference to the array itself. [perl #47363]

  • A block containing nothing except a C-style for loop could corrupt the stack, causing lists outside the block to lose elements or have elements overwritten. This could happen with map { for(...){...} } ... and with lists containing do { for(...){...} }. [perl #123286]

  • scalar() now propagates lvalue context, so that for(scalar($#foo)) { ... } can modify $#foo through $_.

  • qr/@array(?{block})/ no longer dies with "Bizarre copy of ARRAY". [#123344]

  • eval '$variable' in nested named subroutines would sometimes look up a global variable even with a lexical variable in scope.

  • In perl 5.20.0, sort CORE::fake where 'fake' is anything other than a keyword started chopping of the last 6 characters and treating the result as a sort sub name. The previous behaviour of treating "CORE::fake" as a sort sub name has been restored. [perl #123410]

  • Outside of use utf8, a single-character Latin-1 lexical variable is disallowed. The error message for it, "Can't use global $foo...", was giving garbage instead of the variable name.

  • readline on a nonexistent handle was causing ${^LAST_FH} to produce a reference to an undefined scalar (or fail an assertion). Now ${^LAST_FH} ends up undefined.

  • (...)x... in void context now applies scalar context to the left-hand argument, instead of the context the current sub was called in. [perl #123020]

Known Problems

  • A goal is for Perl to be able to be recompiled to work reasonably well on any Unicode version. In Perl 5.22, though, the earliest such version is Unicode 5.1 (current is 7.0).

  • EBCDIC platforms

    • Encode and encoding are mostly broken.

    • Many cpan modules that are shipped with core show failing tests.

    • pack/unpack with "U0" format may not work properly.

  • The following modules are known to have test failures with this version of Perl. Patches have been submitted, so there will hopefully be new releases soon:

Acknowledgements

XXX Generate this with:

perl Porting/acknowledgements.pl v5.20.0..HEAD

Reporting Bugs

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 https://rt.perl.org/ . There may also be information at http://www.perl.org/ , 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.

If the bug you are reporting has security implications, which make it inappropriate to send to a publicly archived mailing list, then please send it to perl5-security-report@perl.org. This points to a closed subscription unarchived mailing list, which includes all the core committers, who will be able to help assess the impact of issues, figure out a resolution, and help co-ordinate the release of patches to mitigate or fix the problem across all platforms on which Perl is supported. Please only use this address for security issues in the Perl core, not for modules independently distributed on CPAN.

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.