NAME

DESCRIPTION

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

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

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

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

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

Security

Incompatible Changes

    U+0085 NEXT LINE
    U+200E LEFT-TO-RIGHT MARK
    U+200F RIGHT-TO-LEFT MARK
    U+2028 LINE SEPARATOR
    U+2029 PARAGRAPH SEPARATOR

    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

    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
    }

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

Performance Enhancements

• If a method or class name is known at compile time, a hash is 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 expression is optimised to the empty list, so long as the left-hand argument is a simple scalar or constant. (That is, "(foo())x0" is not subject to this optimisation.)
• "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.
• 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.
• There is a performance improvement of up to 20% when "length" is applied to a non-magical, non-tied string, and either "use bytes" is in scope or the string doesn't use UTF-8 internally.
• On most perl builds with 64-bit integers, memory usage for non-magical, non-tied scalars containing only a floating point value has been reduced by between 8 and 32 bytes, depending on OS.
• In "@array = split", the assignment can be optimized away, so that "split" writes 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. So for example "join "-", "a", "b"" is converted at compile-time to "a-b". Moreover, "join" with a scalar or constant for the separator and a single-item list to join is simplified to a stringification, and 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. [GH #13659] <https://github.com/Perl/perl5/issues/13659> [GH #13856] <https://github.com/Perl/perl5/issues/13856> [GH #13874] <https://github.com/Perl/perl5/issues/13874>
• "-T" and "-B" filetests will return sooner when an empty file is detected. [GH #13686] <https://github.com/Perl/perl5/issues/13686>
• Hash lookups where the key is a constant are faster.
• Subroutines with an empty prototype and a body containing just "undef" are now eligible for inlining. [GH #14077] <https://github.com/Perl/perl5/issues/14077>
• Subroutines in packages no longer need to be stored in typeglobs: declaring a subroutine will now put a simple sub reference directly in the stash if possible, saving memory. The typeglob still notionally exists, so accessing it will cause the stash entry to be upgraded to a typeglob (i.e. this is just an internal implementation detail). This optimization does not currently apply to XSUBs or exported subroutines, and method calls will undo it, since they cache things in typeglobs. [GH #13392] <https://github.com/Perl/perl5/issues/13392>
• 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, because some data is now memory mapped from disk and shared between processes from the same perl binary.

Modules and Pragmata

  corelist --diff 5.20.0 5.22.0

Documentation

Diagnostics

Utility Changes

Configuration and Compilation

• Configure now checks for "lrintl()", "lroundl()", "llrintl()", and "llroundl()".
• Configure with "-Dmksymlinks" should now be faster. [GH #13890] <https://github.com/Perl/perl5/issues/13890>.
• The "pthreads" and "cl" libraries 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.
• 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 on x86 and IA-64 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.

  [GH #13658] <https://github.com/Perl/perl5/issues/13658>

• 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".
• Pathtools no longer tries to load XS on miniperl. This speeds up building perl slightly.

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 infinity and NaN are working correctly. See "Infinity and NaN (not-a-number) handling improved".

Platform Support

Internal Changes

•

Experimental support has been added to allow ops in the optree to locate their parent, if any. This is enabled by the non-default build option "-DPERL_OP_PARENT". It is envisaged that this will eventually become enabled by default, so XS code which directly accesses the "op_sibling" field of ops should be updated to be future-proofed.

On "PERL_OP_PARENT" builds, the "op_sibling" field has been renamed "op_sibparent" and a new flag, "op_moresib", added. On the last op in a sibling chain, "op_moresib" is false and "op_sibparent" points to the parent (if any) rather than being "NULL".

To make existing code work transparently whether using "PERL_OP_PARENT" or not, a number of new macros and functions have been added that should be used, rather than directly manipulating "op_sibling".

For the case of just reading "op_sibling" to determine the next sibling, two new macros have been added. A simple scan through a sibling chain like this:

    for (; kid->op_sibling; kid = kid->op_sibling) { ... }
    

should now be written as:

    for (; OpHAS_SIBLING(kid); kid = OpSIBLING(kid)) { ... }
    

For altering optrees, a general-purpose function "op_sibling_splice()" has been added, which allows for manipulation of a chain of sibling ops. By analogy with the Perl function "splice()", it allows you to cut out zero or more ops from a sibling chain and replace them with zero or more new ops. It transparently handles all the updating of sibling, parent, op_last pointers etc.

If you need to manipulate ops at a lower level, then three new macros, "OpMORESIB_set", "OpLASTSIB_set" and "OpMAYBESIB_set" 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. Note that unlike "op_sibling_splice()" these macros won't maintain consistency in the parent at the same time (e.g. by updating "op_first" and "op_last" where appropriate).

A C-level "Perl_op_parent()" function and a Perl-level "B::OP::parent()" method have been added. The C function only exists under "PERL_OP_PARENT" builds (using it is build-time error on vanilla perls). "B::OP::parent()" exists always, but on a vanilla build it always returns "NULL". Under "PERL_OP_PARENT", they return the parent of the current op, if any. The variable $B::OP::does_parent allows you to determine whether "B" supports retrieving an op's parent.

"PERL_OP_PARENT" was introduced in 5.21.2, but the interface was changed considerably in 5.21.11. If you updated your code before the 5.21.11 changes, it may require further revision. The main changes after 5.21.2 were:

• 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.
• A new function, "Perl_sv_get_backrefs()", has been added which allows you retrieve the weak references, if any, which point at an SV.
• The "screaminstr()" function has been removed. Although marked as public API, it was undocumented and had no usage in CPAN modules. Calling it has been fatal since 5.17.0.
• The "newDEFSVOP()", "block_start()", "block_end()" and "intro_my()" functions have been added to the API.
• The internal "convert" function in op.c has been renamed "op_convert_list" and added to the API.
• The "sv_magic()" function 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. [GH #14202] <https://github.com/Perl/perl5/issues/14202>.
• Accessing "CvPADLIST" in perlapi on an XSUB is now forbidden.

  The "CvPADLIST" field has been reused for a different internal purpose for XSUBs. So in particular, you can no longer rely on it being NULL as a test of whether a CV is an XSUB. Use "CvISXSUB()" instead.

• SVs of type "SVt_NV" are now sometimes bodiless when the build configuration and platform allow it: specifically, when "sizeof(NV) <= sizeof(IV)". "Bodiless" means that the NV value is stored directly in the head of an SV, without requiring a separate body to be allocated. This trick has already been used for IVs since 5.9.2 (though in the case of IVs, it is always used, regardless of platform and build configuration).
• The $DB::single, $DB::signal and $DB::trace variables 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 infinitely if an overloaded object is assigned to any of those variables. [GH #14013] <https://github.com/Perl/perl5/issues/14013>.
• "Perl_tmps_grow()", which is marked as public API but is undocumented, has been removed from the public API. This change does not affect XS code that uses the "EXTEND_MORTAL" macro to pre-extend the mortal stack.
• Perl's internals no longer sets or uses the "SVs_PADMY" flag. "SvPADMY()" now returns a true value for anything not marked "PADTMP" and "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 five years ago, 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 UTF-8, respectively. (These flags have in fact been present since 5.16.0, but were formerly not regarded as part of the API.)
• A new opcode class, "METHOP", has been introduced. It holds information used at runtime to improve the performance of class/object method calls.

  "OP_METHOD" and "OP_METHOD_NAMED" have changed from being "UNOP/SVOP" to being "METHOP".

• "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.

  [GH #12767] <https://github.com/Perl/perl5/issues/12767> [GH #13392] <https://github.com/Perl/perl5/issues/13392>

• "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. [GH #12767] <https://github.com/Perl/perl5/issues/12767>
• "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 it 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. [GH #13392] <https://github.com/Perl/perl5/issues/13392>
• The "sync_locale" in perlapi function has been added to the public API. Changing the program's locale should be avoided by XS code. Nevertheless, certain non-Perl libraries called from XS need to do so, such as "Gtk". 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, debugging builds now have a new assertion in "op_free()" to ensure that the op doesn't have any unrecognized flags set in "op_private".
• 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 are new macros to manipulate the LC_NUMERIC locale, including "STORE_LC_NUMERIC_SET_TO_NEEDED" and "STORE_LC_NUMERIC_FORCE_TO_UNDERLYING". See "Locale-related functions and macros" in perlapi.
• 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".

  Note that the prefix-less versions of those functions that are part of the public API, such as "cast_i32()", remain unaffected.

• The "PADNAME" and "PADNAMELIST" types are now separate types, and no longer simply aliases for SV and AV. [GH #14250] <https://github.com/Perl/perl5/issues/14250>.
• Pad names are now always UTF-8. 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.
• 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 next item).
• 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

• "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.

• "no re" now can turn off everything that "use re" enables

  Previously, running "no re" would turn off only a few things. Now it can turn off all the enabled things. For example, the only way to stop debugging, once enabled, was to exit the enclosing block; that is now fixed.

• "pack("D", $x)" and "pack("F", $x)" now zero the padding on x86 long double builds. Under some build options on GCC 4.8 and later, they used to either overwrite the zero-initialized padding, or bypass the initialized buffer entirely. This caused op/pack.t to fail. [GH #14554] <https://github.com/Perl/perl5/issues/14554>
• 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. [GH #14605] <https://github.com/Perl/perl5/issues/14605>
• An assertion failure and subsequent crash with "*x=<y>" has been fixed. [GH #14493] <https://github.com/Perl/perl5/issues/14493>
• A possible crashing/looping bug related to compiling lexical subs has been fixed. [GH #14596] <https://github.com/Perl/perl5/issues/14596>
• UTF-8 now works correctly in function names, in unquoted HERE-document terminators, and in variable names used as array indexes. [GH #14601] <https://github.com/Perl/perl5/issues/14601>
• Repeated global pattern matches in scalar context on large tainted strings were exponentially slow depending on the current match position in the string. [GH #14238] <https://github.com/Perl/perl5/issues/14238>
• Various crashes due to the parser getting confused by syntax errors have been fixed. [GH #14496] <https://github.com/Perl/perl5/issues/14496> [GH #14497] <https://github.com/Perl/perl5/issues/14497> [GH #14548] <https://github.com/Perl/perl5/issues/14548> [GH #14564] <https://github.com/Perl/perl5/issues/14564>
• "split" in the scope of lexical $_ has been fixed not to fail assertions. [GH #14483] <https://github.com/Perl/perl5/issues/14483>
• "my $x : attr" syntax inside various list operators no longer fails assertions. [GH #14500] <https://github.com/Perl/perl5/issues/14500>
• 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. [GH #14553] <https://github.com/Perl/perl5/issues/14553>
• "*bar::=*foo::=*glob_with_hash" has been crashing since Perl 5.14, but no longer does. [GH #14512] <https://github.com/Perl/perl5/issues/14512>
• "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". [GH #14569] <https://github.com/Perl/perl5/issues/14569>
• Several cases of data used to store environment variable contents in core C code being potentially overwritten before being used have been fixed. [GH #14476] <https://github.com/Perl/perl5/issues/14476>
• Some patterns starting with "/.*..../" matched against long strings have been slow since v5.8, and some of the form "/.*..../i" have been slow since v5.18. They are now all fast again. [GH #14475] <https://github.com/Perl/perl5/issues/14475>.
• 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 $/ would see the array reference. [GH #14245] <https://github.com/Perl/perl5/issues/14245>.
• In a regular expression pattern, a POSIX class, like "[:ascii:]", must be inside a bracketed character class, like "qr/[[:ascii:]]/". A warning is issued when something looking like a POSIX class is not inside a bracketed class. That warning wasn't getting generated when the POSIX class was negated: "[:^ascii:]". This is now fixed.
• Perl 5.14.0 introduced a bug whereby "eval { LABEL: }" would crash. This has been fixed. [GH #14438] <https://github.com/Perl/perl5/issues/14438>.
• Various crashes due to the parser getting confused by syntax errors have been fixed. [GH #14421] <https://github.com/Perl/perl5/issues/14421>. [GH #14472] <https://github.com/Perl/perl5/issues/14472>. [GH #14480] <https://github.com/Perl/perl5/issues/14480>. [GH #14447] <https://github.com/Perl/perl5/issues/14447>.
• Code like "/$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. [GH #14462] <https://github.com/Perl/perl5/issues/14462>.
• Fix argument underflow for "pack". [GH #14525] <https://github.com/Perl/perl5/issues/14525>.
• 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". [GH #14499] <https://github.com/Perl/perl5/issues/14499>.
• 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 Infinity and NaN, including warnings when stringifying Infinity-like or NaN-like strings. For example, "NaNcy" doesn't numify to NaN anymore.
• 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 (which 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. [GH #14389] <https://github.com/Perl/perl5/issues/14389>.
• "s///g" now works on very long strings (where there are more than 2 billion iterations) instead of dying with 'Substitution loop'. [GH #11742] <https://github.com/Perl/perl5/issues/11742>. [GH #14190] <https://github.com/Perl/perl5/issues/14190>.
• "gmtime" no longer crashes with not-a-number values. [GH #14365] <https://github.com/Perl/perl5/issues/14365>.
• "\()" (a reference to an empty list), and "y///" with lexical $_ in scope, could both do a bad write past the end of the stack. They have both 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 $_ instead. [GH #14376] <https://github.com/Perl/perl5/issues/14376>.
• Some cases of lexical state subs declared inside predeclared subs could crash, for example when evalling a string including the name of an outer variable, but no longer do.
• Some cases of nested lexical state subs inside anonymous subs could cause 'Bizarre copy' errors or possibly even crashes.
• When trying to emit warnings, perl's default debugger (perl5db.pl) was sometimes giving 'Undefined subroutine &DB::db_warn called' instead. This bug, which started to occur in Perl 5.18, has been fixed. [GH #14400] <https://github.com/Perl/perl5/issues/14400>.
• Certain syntax errors in substitutions, such as "s/${<>{})//", 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. [GH #14391] <https://github.com/Perl/perl5/issues/14391>.
• Fix a couple of string grow size calculation overflows; in particular, 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. [GH #14401] <https://github.com/Perl/perl5/issues/14401>.
• "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. [GH #14388] <https://github.com/Perl/perl5/issues/14388>.
• A possible buffer overrun and crash when parsing a literal pattern during regular expression compilation has been fixed. [GH #14416] <https://github.com/Perl/perl5/issues/14416>.
• "fchmod()" and "futimes()" now set $! when they fail due to being passed a closed file handle. [GH #14073] <https://github.com/Perl/perl5/issues/14073>.
• "op_free()" and "scalarvoid()" no longer crash due to a stack overflow when freeing a deeply recursive op tree. [GH #11866] <https://github.com/Perl/perl5/issues/11866>.
• In Perl 5.20.0, $^N accidentally had the internal UTF-8 flag turned off if accessed from a code block within a regular expression, effectively UTF-8-encoding the value. This has been fixed. [GH #14211] <https://github.com/Perl/perl5/issues/14211>.
• A failed "semctl" call no longer overwrites existing items on the stack, which means that "(semctl(-1,0,0,0))[0]" no longer gives an "uninitialized" warning.
• "else{foo()}" with no space before "foo" is now better at assigning the right line number to that statement. [GH #14070] <https://github.com/Perl/perl5/issues/14070>.
• Sometimes the assignment in "@array = split" gets optimised so that "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.) It has now been fixed. [GH #14183] <https://github.com/Perl/perl5/issues/14183>.
• When an argument list fails the checks specified by a subroutine signature (which is still an experimental feature), the resulting error messages now give the file and line number of the caller, not of the called subroutine. [GH #13643] <https://github.com/Perl/perl5/issues/13643>.
• The 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. [GH #14110] <https://github.com/Perl/perl5/issues/14110>.
• 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. [GH #14110] <https://github.com/Perl/perl5/issues/14110>.
• "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. [GH #14088] <https://github.com/Perl/perl5/issues/14088>.
• 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 was an object with "x" overloading, causing erratic behavior. [GH #13811] <https://github.com/Perl/perl5/issues/13811>.
• Assignment to a lexical scalar is often optimised away; for example in "my $x; $x = $y + $z", the assign operator is optimised away and the add operator writes its result directly to $x. Various bugs related to this optimisation have been fixed. Certain operators on the right-hand side would sometimes fail to assign the value at all or assign the wrong value, or would call STORE twice or not at all on tied variables. The operators affected were "$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 use 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, of which all but the last were 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 UTF-8 string previously, then "crypt" would not turn off the UTF-8 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.
• "index()" and "rindex()" no longer crash when used on strings over 2GB in size. [GH #13700] <https://github.com/Perl/perl5/issues/13700>.
• 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 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:]]". This was 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 <http://www.coverity.com/> 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 UTF-8 strings corrupted "pos()". This bug, introduced in 5.20, is now fixed. [GH #13948] <https://github.com/Perl/perl5/issues/13948>.
• 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. [GH #13917] <https://github.com/Perl/perl5/issues/13917>.
• "" P" =~ /(?=.*P)P/" should match, but did not. This is now fixed. [GH #13954] <https://github.com/Perl/perl5/issues/13954>.
• 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. [GH #13926] <https://github.com/Perl/perl5/issues/13926>.
• "method { BLOCK } ARGS" syntax now correctly parses the arguments if they begin with an opening brace. [GH #9085] <https://github.com/Perl/perl5/issues/9085>.
• 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. [GH #13863] <https://github.com/Perl/perl5/issues/13863>.
• A bug has been fixed where zero-length assertions and code blocks inside of a regex could cause "pos" to see an incorrect value. [GH #14016] <https://github.com/Perl/perl5/issues/14016>.
• Dereferencing of constants now works correctly for typeglob constants. Previously the glob was stringified and its name looked up. Now the glob itself is used. [GH #9891] <https://github.com/Perl/perl5/issues/9891>
• When parsing a sigil ("$" "@" "%" "&)" 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. [GH #14032] <https://github.com/Perl/perl5/issues/14032>
• 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.
• "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. [GH #14059] <https://github.com/Perl/perl5/issues/14059>
• "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. [GH #10882] <https://github.com/Perl/perl5/issues/10882>
• 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.
• Perl 5.20.0 introduced a regression in which a UTF-8 encoded regular expression pattern that contains a single ASCII lowercase letter did not match its uppercase counterpart. That has been fixed in both 5.20.1 and 5.22.0. [GH #14051] <https://github.com/Perl/perl5/issues/14051>
• 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 Unicode tables during a regular expression match could cause assertion failures under debugging builds if the previous match used the very same regular expression. [GH #14081] <https://github.com/Perl/perl5/issues/14081>
• 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 using a lexical sub as 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. [GH #5788] <https://github.com/Perl/perl5/issues/5788>
• 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.) [GH #14086] <https://github.com/Perl/perl5/issues/14086>
• 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 were read-only to begin with. Previously, unlocking such values could result in crashes, hangs or other erratic behavior.
• 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. [GH #14087] <https://github.com/Perl/perl5/issues/14087>.
• "*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 into assuming a limit of 1.
• Fatal warnings no longer prevent the output of syntax errors. [GH #14155] <https://github.com/Perl/perl5/issues/14155>.
• 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 incorrectly leave some files behind. [GH #14108] <https://github.com/Perl/perl5/issues/14108>.
• AIX now sets the length in "getsockopt" correctly. [GH #13484] <https://github.com/Perl/perl5/issues/13484>. [cpan #91183] <https://rt.cpan.org/Ticket/Display.html?id=91183>. [cpan #85570] <https://rt.cpan.org/Ticket/Display.html?id=85570>.
• The optimization phase of a regexp compilation could run "forever" and exhaust all memory under certain circumstances; now fixed. [GH #13984] <https://github.com/Perl/perl5/issues/13984>.
• The test script t/op/crypt.t now uses the SHA-256 algorithm if the default one is disabled, rather than giving failures. [GH #13715] <https://github.com/Perl/perl5/issues/13715>.
• Fixed an off-by-one error when setting the size of a shared array. [GH #14151] <https://github.com/Perl/perl5/issues/14151>.
• Fixed a bug that could cause perl to enter an infinite loop during compilation. In particular, a while(1) within a sublist, e.g.

      sub foo { () = ($a, my $b, ($c, do { while(1) {} })) }
      

  The bug was introduced in 5.20.0 [GH #14165] <https://github.com/Perl/perl5/issues/14165>.

• On Win32, if a variable was "local"-ized in a pseudo-process that later forked, restoring the original value in the child pseudo-process caused memory corruption and a crash in the child pseudo-process (and therefore the OS process). [GH #8641] <https://github.com/Perl/perl5/issues/8641>.
• 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. [GH #14255] <https://github.com/Perl/perl5/issues/14255>.
• Non-ASCII lexical sub names now appear without trailing junk when they appear in error messages.
• The "\@" subroutine prototype no longer flattens parenthesized arrays (taking a reference to each element), but takes a reference to the array itself. [GH #9111] <https://github.com/Perl/perl5/issues/9111>.
• 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(...){...} }". [GH #14269] <https://github.com/Perl/perl5/issues/14269>.
• "scalar()" now propagates lvalue context, so that "for(scalar($#foo)) { ... }" can modify $#foo through $_.
• "qr/@array(?{block})/" no longer dies with "Bizarre copy of ARRAY". [GH #14292] <https://github.com/Perl/perl5/issues/14292>.
• "eval '$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 off the last 6 characters and treating the result as a sort sub name. The previous behavior of treating "CORE::fake" as a sort sub name has been restored. [GH #14323] <https://github.com/Perl/perl5/issues/14323>.
• 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. [GH #14174] <https://github.com/Perl/perl5/issues/14174>.

Known Problems

• "pack"-ing a NaN on a perl compiled with Visual C 6 does not behave properly, leading to a test failure in t/op/infnan.t. [GH #14705] <https://github.com/Perl/perl5/issues/14705>
• 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

•

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

Obituary

Acknowledgements

Excluding auto-generated files, documentation and release tools, there were approximately 370,000 lines of changes to 1,500 .pm, .t, .c and .h files.

Perl continues to flourish into its third decade thanks to a vibrant community of users and developers. The following people are known to have contributed the improvements that became Perl 5.22.0:

Aaron Crane, Abhijit Menon-Sen, Abigail, Alberto Simo~es, Alex Solovey, Alex Vandiver, Alexandr Ciornii, Alexandre (Midnite) Jousset, Andreas Koenig, Andreas Voegele, Andrew Fresh, Andy Dougherty, Anthony Heading, Aristotle Pagaltzis, brian d foy, Brian Fraser, Chad Granum, Chris 'BinGOs' Williams, Craig A. Berry, Dagfinn Ilmari Mannsaaker, Daniel Dragan, Darin McBride, Dave Rolsky, David Golden, David Mitchell, David Wheeler, Dmitri Tikhonov, Doug Bell, E. Choroba, Ed J, Eric Herman, Father Chrysostomos, George Greer, Glenn D. Golden, Graham Knop, H.Merijn Brand, Herbert Breunung, Hugo van der Sanden, James E Keenan, James McCoy, James Raspass, Jan Dubois, Jarkko Hietaniemi, Jasmine Ngan, Jerry D. Hedden, Jim Cromie, John Goodyear, kafka, Karen Etheridge, Karl Williamson, Kent Fredric, kmx, Lajos Veres, Leon Timmermans, Lukas Mai, Mathieu Arnold, Matthew Horsfall, Max Maischein, Michael Bunk, Nicholas Clark, Niels Thykier, Niko Tyni, Norman Koch, Olivier Mengue, Peter John Acklam, Peter Martini, Petr PisaX, Philippe Bruhat (BooK), Pierre Bogossian, Rafael Garcia-Suarez, Randy Stauner, Reini Urban, Ricardo Signes, Rob Hoelz, Rostislav Skudnov, Sawyer X, Shirakata Kentaro, Shlomi Fish, Sisyphus, Slaven Rezic, Smylers, Steffen Mueller, Steve Hay, Sullivan Beck, syber, Tadeusz SoXnierz, Thomas Sibley, Todd Rinaldo, Tony Cook, Vincent Pit, Vladimir Marek, Yaroslav Kuzmin, Yves Orton, AEvar Arnfjoerd` Bjarmason.

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 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 INSTALL file for how to build Perl.

The README file for general stuff.

The Artistic and Copying files for copyright information.