revise hash overhaul docs

[perl5.git] / pod / perldelta.pod

=encoding utf8

=head1 NAME

perldelta - what is new for perl v5.18.0

=head1 DESCRIPTION

This document describes differences between the 5.16.0 release and the 5.18.0
release.

If you are upgrading from an earlier release such as 5.14.0, first read
L<perl5160delta>, which describes differences between 5.14.0 and 5.16.0.

=head1 Core Enhancements

=head2 New mechanism for experimental features

Newly-added experimental features will now require this incantation:

    no warnings "experimental::feature_name";
    use feature "feature_name";  # would warn without the prev line

There is a new warnings category, called "experimental", containing
warnings that the L<feature> pragma emits when enabling experimental
features.

Newly-added experimental features will also be given special warning IDs,
which consist of "experimental::" followed by the name of the feature.  (The
plan is to extend this mechanism eventually to all warnings, to allow them
to be enabled or disabled individually, and not just by category.)

By saying

    no warnings "experimental::feature_name";

you are taking responsibility for any breakage that future changes to, or
removal of, the feature may cause.

Existing experimental features may begin emitting these warnings, too.  Please
consult L<perlexperiment> for information on which features are considered
experimental.

=head2 Hash overhaul

Changes to the implementation of hashes in perl 5.18.0 will be one of the most
visible changes to the behavior of existing code.

By default, two distinct hash variables with identical keys and values will now
provide their contents in a different order where it was previously identical.

When encountering these changes, the key to cleaning up from them is to accept
that B<hashes are unordered collections> and to act accordingly.

=head3 Hash randomization

The seed used by Perl's hash function is now random.  This means that the
order which keys/values will be returned from functions like C<keys()>,
C<values()>, and C<each()> will differ from run to run.

This change was introduced to make Perl's hashes more robust to algorithmic
complexity attacks, and also because we discovered that it exposes hash
ordering dependency bugs and makes them easier to track down.

Toolchain maintainers might want to invest in additional infrastructure to
test for things like this.  Running tests several times in a row and then
comparing results will make it easier to spot hash order dependencies in
code.  Authors are strongly encouraged not to expose the key order of
Perl's hashes to insecure audiences.

Further, every hash has its own iteration order, which should make it much
more difficult to determine what the current hash seed is.

=head3 New hash function: Murmurhash-32

We have switched Perl's hash function to use Murmurhash-32, and added build
support for several other hash functions.  This new function is expected to
perform equivalently to the old one for shorter strings and is faster for
hashing longer strings.

=head3 PERL_HASH_SEED environment variable now takes a hex value

C<PERL_HASH_SEED> no longer accepts an integer as a parameter;
instead the value is expected to be a binary value encoded in a hex
string, such as "0xf5867c55039dc724".  This is to make the
infrastructure support hash seeds of arbitrary lengths, which might
exceed that of an integer.  (SipHash uses a 16 byte seed).

=head3 PERL_PERTURB_KEYS environment variable added

The C<PERL_PERTURB_KEYS> environment variable allows one to control the level of
randomization applied to C<keys> and friends.

When C<PERL_PERTURB_KEYS> is 0, perl will not randomize the key order at all. The
chance that C<keys> changes due to an insert will be the same as in previous
perls, basically only when the bucket size is changed.

When C<PERL_PERTURB_KEYS> is 1, perl will randomize keys in a non-repeatable
way. The chance that C<keys> changes due to an insert will be very high.  This
is the most secure and default mode.

When C<PERL_PERTURB_KEYS> is 2, perl will randomize keys in a repeatable way.
Repeated runs of the same program should produce the same output every time.

C<PERL_HASH_SEED> implies a non-default C<PERL_PERTURB_KEYS> setting. Setting
C<PERL_HASH_SEED=0> (exactly one 0) implies C<PERL_PERTURB_KEYS=0> (hash key
randomization disabled); settng C<PERL_HASH_SEED> to any other value implies
C<PERL_PERTURB_KEYS=2> (deterministic and repeatable hash key randomization).
Specifying C<PERL_PERTURB_KEYS> explicitly to a different level overrides this
behavior.

=head3 Hash::Util::hash_seed() now returns a string

Hash::Util::hash_seed() now returns a string instead of an integer.  This
is to make the infrastructure support hash seeds of arbitrary lengths
which might exceed that of an integer.  (SipHash uses a 16 byte seed).

=head3 Output of PERL_HASH_SEED_DEBUG has been changed

The environment variable PERL_HASH_SEED_DEBUG now makes perl show both the
hash function perl was built with, I<and> the seed, in hex, in use for that
process. Code parsing this output, should it exist, must change to accommodate
the new format.  Example of the new format:

    $ PERL_HASH_SEED_DEBUG=1 ./perl -e1
    HASH_FUNCTION = MURMUR3 HASH_SEED = 0x1476bb9f

=head2 Upgrade to Unicode 6.2

Perl now supports Unicode 6.2.  A list of changes from Unicode
6.1 is at L<http://www.unicode.org/versions/Unicode6.2.0>.

=head2 Character name aliases may now include non-Latin1-range characters

It is possible to define your own names for characters for use in
C<\N{...}>, C<charnames::vianame()>, etc.  These names can now be
comprised of characters from the whole Unicode range.  This allows for
names to be in your native language, and not just English.  Certain
restrictions apply to the characters that may be used (you can't define
a name that has punctuation in it, for example).  See L<charnames/CUSTOM
ALIASES>.

=head2 New DTrace probes

The following new DTrace probes have been added:

=over 4

=item *

C<op-entry>

=item *

C<loading-file>

=item *

C<loaded-file>

=back

=head2 C<${^LAST_FH}>

This new variable provides access to the filehandle that was last read.
This is the handle used by C<$.> and by C<tell> and C<eof> without
arguments.

=head2 Regular Expression Set Operations

This is an B<experimental> feature to allow matching against the union,
intersection, etc., of sets of code points, similar to
L<Unicode::Regex::Set>.  It can also be used to extend C</x> processing
to [bracketed] character classes, and as a replacement of user-defined
properties, allowing more complex expressions than they do.  See
L<perlrecharclass/Extended Bracketed Character Classes>.

=head2 Lexical subroutines

This new feature is still considered B<experimental>.  To enable it:

    use 5.018;
    no warnings "experimental::lexical_subs";
    use feature "lexical_subs";

You can now declare subroutines with C<state sub foo>, C<my sub foo>, and
C<our sub foo>.  (C<state sub> requires that the "state" feature be
enabled, unless you write it as C<CORE::state sub foo>.)

C<state sub> creates a subroutine visible within the lexical scope in which
it is declared.  The subroutine is shared between calls to the outer sub.

C<my sub> declares a lexical subroutine that is created each time the
enclosing block is entered.  C<state sub> is generally slightly faster than
C<my sub>.

C<our sub> declares a lexical alias to the package subroutine of the same
name.

For more information, see L<perlsub/Lexical Subroutines>.

=head2 Computed Labels

The loop controls C<next>, C<last> and C<redo>, and the special C<dump>
operator, now allow arbitrary expressions to be used to compute labels at run
time.  Previously, any argument that was not a constant was treated as the
empty string.

=head2 More CORE:: subs

Several more built-in functions have been added as subroutines to the
CORE:: namespace - namely, those non-overridable keywords that can be
implemented without custom parsers: C<defined>, C<delete>, C<exists>,
C<glob>, C<pos>, C<protoytpe>, C<scalar>, C<split>, C<study>, and C<undef>.

As some of these have prototypes, C<prototype('CORE::...')> has been
changed to not make a distinction between overridable and non-overridable
keywords.  This is to make C<prototype('CORE::pos')> consistent with
C<prototype(&CORE::pos)>.

=head2 C<kill> with negative signal names

C<kill> has always allowed a negative signal number, which kills the
process group instead of a single process.  It has also allowed signal
names.  But it did not behave consistently, because negative signal names
were treated as 0.  Now negative signals names like C<-INT> are supported
and treated the same way as -2 [perl #112990].

=head1 Security

=head2 C<Storable> security warning in documentation

The documentation for C<Storable> now includes a section which warns readers
of the danger of accepting Storable documents from untrusted sources. The
short version is that deserializing certain types of data can lead to loading
modules and other code execution. This is documented behavior and wanted
behavior, but this opens an attack vector for malicious entities.

=head2 C<Locale::Maketext> allowed code injection via a malicious template

If users could provide a translation string to Locale::Maketext, this could be
used to invoke arbitrary Perl subroutines available in the current process.

This has been fixed, but it is still possible to invoke any method provided by
C<Locale::Maketext> itself or a subclass that you are using. One of these
methods in turn will invoke the Perl core's C<sprintf> subroutine.

In summary, allowing users to provide translation strings without auditing
them is a bad idea.

This vulnerability is documented in CVE-2012-6329.

=head2 Avoid calling memset with a negative count

Poorly written perl code that allows an attacker to specify the count to perl's
C<x> string repeat operator can already cause a memory exhaustion
denial-of-service attack. A flaw in versions of perl before 5.15.5 can escalate
that into a heap buffer overrun; coupled with versions of glibc before 2.16, it
possibly allows the execution of arbitrary code.

The flaw addressed to this commit has been assigned identifier CVE-2012-5195
and was researched by Tim Brown.

=head1 Incompatible Changes

=head2 See also: hash overhaul

Some of the changes in the L<hash overhaul|/"Hash overhaul"> are not fully
compatible with previous versions of perl.  Please read that section.

=head2 An unknown character name in C<\N{...}> is now a syntax error

Previously, it warned, and the Unicode REPLACEMENT CHARACTER was
substituted.  Unicode now recommends that this situation be a syntax
error.  Also, the previous behavior led to some confusing warnings and
behaviors, and since the REPLACEMENT CHARACTER has no use other than as
a stand-in for some unknown character, any code that has this problem is
buggy.

=head2 Formerly deprecated characters in C<\N{}> character name aliases are now errors.

Since v5.12.0, it has been deprecated to use certain characters in
user-defined C<\N{...}> character names.  These now cause a syntax
error.  For example, it is now an error to begin a name with a digit,
such as in

 my $undraftable = "\N{4F}";    # Syntax error!

or to have commas anywhere in the name.  See L<charnames/CUSTOM ALIASES>

=head2 C<\N{BELL}> now refers to U+1F514 instead of U+0007

Unicode 6.0 reused the name "BELL" for a different code point than it
traditionally had meant.  Since Perl v5.14, use of this name still
referred to U+0007, but would raise a deprecation warning.  Now, "BELL"
refers to U+1F514, and the name for U+0007 is "ALERT".  All the
functions in L<charnames> have been correspondingly updated.

=head2 New Restrictions in Multi-Character Case-Insensitive Matching in Regular Expression Bracketed Character Classes

Unicode has now withdrawn their previous recommendation for regular
expressions to automatically handle cases where a single character can
match multiple characters case-insensitively, for example the letter
LATIN SMALL LETTER SHARP S and the sequence C<ss>.  This is because
it turns out to be impracticable to do this correctly in all
circumstances.  Because Perl has tried to do this as best it can, it
will continue to do so.  (We are considering an option to turn it off.)
However, a new restriction is being added on such matches when they
occur in [bracketed] character classes.  People were specifying
things such as C</[\0-\xff]/i>, and being surprised that it matches the
two character sequence C<ss> (since LATIN SMALL LETTER SHARP S occurs in
this range).  This behavior is also inconsistent with using a
property instead of a range:  C<\p{Block=Latin1}> also includes LATIN
SMALL LETTER SHARP S, but C</[\p{Block=Latin1}]/i> does not match C<ss>.
The new rule is that for there to be a multi-character case-insensitive
match within a bracketed character class, the character must be
explicitly listed, and not as an end point of a range.  This more
closely obeys the Principle of Least Astonishment.  See
L<perlrecharclass/Bracketed Character Classes>.  Note that a bug [perl
#89774], now fixed as part of this change, prevented the previous
behavior from working fully.

=head2 Explicit rules for variable names and identifiers

Due to an oversight, length-one variable names in 5.16 were completely
unrestricted, and opened the door to several kinds of insanity.  As of
5.18, these now follow the rules of other identifiers, in addition
to accepting characters that match the C<\p{POSIX_Punct}> property.

There are no longer any differences in the parsing of identifiers
specified as C<$...> or C<${...}>; previously, they were dealt with in
different parts of the core, and so had slightly different behavior. For
instance, C<${foo:bar}> was a legal variable name.  Since they are now
both parsed by the same code, that is no longer the case.

=head2 C<\s> in regular expressions now matches a Vertical Tab

No one could recall why C<\s> didn't match C<\cK>, the vertical tab.
Now it does.  Given the extreme rarity of that character, very little
breakage is expected.

=head2 C</(?{})/> and C</(??{})/> have been heavily reworked

The implementation of this feature has been almost completely rewritten.
Although its main intent is to fix bugs, some behaviors, especially
related to the scope of lexical variables, will have changed.  This is
described more fully in the L</Selected Bug Fixes> section.

=head2 Stricter parsing of substitution replacement

It is no longer possible to abuse the way the parser parses C<s///e> like
this:

    %_=(_,"Just another ");
    $_="Perl hacker,\n";
    s//_}->{_/e;print

=head2 C<given> now aliases the global C<$_>

Instead of assigning to an implicit lexical C<$_>, C<given> now makes the
global C<$_> an alias for its argument, just like C<foreach>.  However, it
still uses lexical C<$_> if there is lexical C<$_> in scope (again, just like
C<foreach>) [perl #114020].

=head2 Lexical C<$_> is now experimental

Since it was introduced in Perl 5.10, it has caused much confusion with no
obvious solution:

=over

=item *

Various modules (e.g., List::Util) expect callback routines to use the
global C<$_>.  C<use List::Util 'first'; my $_; first { $_ == 1 } @list>
does not work as one would expect.

=item *

A C<my $_> declaration earlier in the same file can cause confusing closure
warnings.

=item *

The "_" subroutine prototype character allows called subroutines to access
your lexical C<$_>, so it is not really private after all.

=item *

Nevertheless, subroutines with a "(@)" prototype and methods cannot access
the caller's lexical C<$_>, unless they are written in XS.

=item *

But even XS routines cannot access a lexical C<$_> declared, not in the
calling subroutine, but in an outer scope, iff that subroutine happened not
to mention C<$_> or use any operators that default to C<$_>.

=back

It is our hope that lexical C<$_> can be rehabilitated, but this may
cause changes in its behavior.  Please use it with caution until it
becomes stable.

=head2 readline() with C<$/ = \N> now reads N characters, not N bytes

Previously, when reading from a stream with I/O layers such as
C<encoding>, the readline() function, otherwise known as the C<< <> >>
operator, would read I<N> bytes from the top-most layer. [perl #79960]

Now, I<N> characters are read instead.

There is no change in behaviour when reading from streams with no
extra layers, since bytes map exactly to characters.

=head2 Overridden C<glob> is now passed one argument

C<glob> overrides used to be passed a magical undocumented second argument
that identified the caller.  Nothing on CPAN was using this, and it got in
the way of a bug fix, so it was removed.  If you really need to identify
the caller, see L<Devel::Callsite> on CPAN.

=head2 Here-doc parsing

The body of a here-document inside a quote-like operator now always begins
on the line after the "<<foo" marker.  Previously, it was documented to
begin on the line following the containing quote-like operator, but that
was only sometimes the case [perl #114040].

=head2 Alphanumeric operators must now be separated from the closing
delimiter of regular expressions

You may no longer write something like:

 m/a/and 1

Instead you must write

 m/a/ and 1

with whitespace separating the operator from the closing delimiter of
the regular expression.  Not having whitespace has resulted in a
deprecation warning since Perl v5.14.0.

=head2 qw(...) can no longer be used as parentheses

C<qw> lists used to fool the parser into thinking they were always
surrounded by parentheses.  This permitted some surprising constructions
such as C<foreach $x qw(a b c) {...}>, which should really be written
C<foreach $x (qw(a b c)) {...}>.  These would sometimes get the lexer into
the wrong state, so they didn't fully work, and the similar C<foreach qw(a
b c) {...}> that one might expect to be permitted never worked at all.

This side effect of C<qw> has now been abolished.  It has been deprecated
since Perl 5.13.11.  It is now necessary to use real parentheses
everywhere that the grammar calls for them.

=head2 Interaction of lexical and default warnings

Turning on any lexical warnings used first to disable all default warnings
if lexical warnings were not already enabled:

    $*; # deprecation warning
    use warnings "void";
    $#; # void warning; no deprecation warning

Now, the C<debugging>, C<deprecated>, C<glob>, C<inplace> and C<malloc> warnings
categories are left on when turning on lexical warnings (unless they are
turned off by C<no warnings>, of course).

This may cause deprecation warnings to occur in code that used to be free
of warnings.

Those are the only categories consisting only of default warnings.  Default
warnings in other categories are still disabled by C<< use warnings "category" >>,
as we do not yet have the infrastructure for controlling
individual warnings.

=head2 C<state sub> and C<our sub>

Due to an accident of history, C<state sub> and C<our sub> were equivalent
to a plain C<sub>, so one could even create an anonymous sub with
C<our sub { ... }>.  These are now disallowed outside of the "lexical_subs"
feature.  Under the "lexical_subs" feature they have new meanings described
in L<perlsub/Lexical Subroutines>.

=head2 Defined values stored in environment are forced to byte strings

A value stored in an environment variable has always been stringified.  In this
release, it is converted to be only a byte string.  First, it is forced to be a
only a string.  Then if the string is utf8 and the equivalent of
C<utf8::downgrade()> works, that result is used; otherwise, the equivalent of
C<utf8::encode()> is used, and a warning is issued about wide characters
(L</Diagnostics>).

=head2 C<require> dies for unreadable files

When C<require> encounters an unreadable file, it now dies.  It used to
ignore the file and continue searching the directories in C<@INC>
[perl #113422].

=head2 C<gv_fetchmeth_*> and SUPER

The various C<gv_fetchmeth_*> XS functions used to treat a package whose
named ended with C<::SUPER> specially.  A method lookup on the C<Foo::SUPER>
package would be treated as a C<SUPER> method lookup on the C<Foo> package.  This
is no longer the case.  To do a C<SUPER> lookup, pass the C<Foo> stash and the
C<GV_SUPER> flag.

=head2 C<split>'s first argument is more consistently interpreted

After some changes earlier in 5.17, C<split>'s behavior has been
simplified: if the PATTERN argument evaluates to a literal string
containing one space, it is treated the way that a I<literal> string
containing one space once was.

=head1 Deprecations

=head2 Deprecated modules

The following modules will be removed from the core distribution in a
future release, and should be installed from CPAN instead. Distributions
on CPAN which require these should add them to their prerequisites.
The core versions of these modules will issue C<"deprecated">-category
warnings.

You can silence these deprecation warnings by installing the modules
in question from CPAN.

=over

=item L<Archive::Extract>

=item L<B::Lint>

=item L<B::Lint::Debug>

=item L<CPANPLUS> and all included C<CPANPLUS::*> modules

=item L<Devel::InnerPackage>

=item L<encoding>

=item L<Log::Message>

=item L<Log::Message::Config>

=item L<Log::Message::Handlers>

=item L<Log::Message::Item>

=item L<Log::Message::Simple>

=item L<Module::Pluggable>

=item L<Module::Pluggable::Object>

=item L<Object::Accessor>

=item L<Pod::LaTeX>

=item L<Term::UI>

=item L<Term::UI::History>

=back

=head2 Deprecated Utilities

The following utilities will be removed from the core distribution in a
future release as their associated modules have been deprecated. They
will remain available with the applicable CPAN distribution.

=over

=item L<cpanp>

=item C<cpanp-run-perl>

=item L<cpan2dist>

These items are part of the C<CPANPLUS> distribution.

=item L<pod2latex>

This item is part of the C<Pod::LaTeX> distribution.

=back

=head2 PL_sv_objcount

This interpreter-global variable used to track the total number of
Perl objects in the interpreter. It is no longer maintained and will
be removed altogether in Perl 5.20.

=head2 Five additional characters should be escaped in patterns with C</x>

When a regular expression pattern is compiled with C</x>, Perl treats 6
characters as white space to ignore, such as SPACE and TAB.  However,
Unicode recommends 11 characters be treated thusly.  We will conform
with this in a future Perl version.  In the meantime, use of any of the
missing characters will raise a deprecation warning, unless turned off.
The five characters 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.

=head2 User-defined charnames with surprising whitespace

A user-defined character name with trailing or multiple spaces in a row is
likely a typo.  This now generates a warning when defined, on the assumption
that uses of it will be unlikely to include the excess whitespace.

=head2 Various XS-callable functions are now deprecated

All the functions used to classify characters will be removed from a
future version of Perl, and should not be used.  With participating C
compilers (e.g., gcc), compiling any file that uses any of these will
generate a warning.  These were not intended for public use; there are
equivalent, faster, macros for most of them.
See L<perlapi/Character classes>.  The complete list is:
C<is_uni_alnum>, C<is_uni_alnumc>, C<is_uni_alnumc_lc>,
C<is_uni_alnum_lc>, C<is_uni_alpha>, C<is_uni_alpha_lc>,
C<is_uni_ascii>, C<is_uni_ascii_lc>, C<is_uni_blank>,
C<is_uni_blank_lc>, C<is_uni_cntrl>, C<is_uni_cntrl_lc>,
C<is_uni_digit>, C<is_uni_digit_lc>, C<is_uni_graph>,
C<is_uni_graph_lc>, C<is_uni_idfirst>, C<is_uni_idfirst_lc>,
C<is_uni_lower>, C<is_uni_lower_lc>, C<is_uni_print>,
C<is_uni_print_lc>, C<is_uni_punct>, C<is_uni_punct_lc>,
C<is_uni_space>, C<is_uni_space_lc>, C<is_uni_upper>,
C<is_uni_upper_lc>, C<is_uni_xdigit>, C<is_uni_xdigit_lc>,
C<is_utf8_alnum>, C<is_utf8_alnumc>, C<is_utf8_alpha>,
C<is_utf8_ascii>, C<is_utf8_blank>, C<is_utf8_char>,
C<is_utf8_cntrl>, C<is_utf8_digit>, C<is_utf8_graph>,
C<is_utf8_idcont>, C<is_utf8_idfirst>, C<is_utf8_lower>,
C<is_utf8_mark>, C<is_utf8_perl_space>, C<is_utf8_perl_word>,
C<is_utf8_posix_digit>, C<is_utf8_print>, C<is_utf8_punct>,
C<is_utf8_space>, C<is_utf8_upper>, C<is_utf8_xdigit>,
C<is_utf8_xidcont>, C<is_utf8_xidfirst>.

In addition these three functions that have never worked properly are
deprecated:
C<to_uni_lower_lc>, C<to_uni_title_lc>, and C<to_uni_upper_lc>.

=head2 Certain rare uses of backslashes within regexes are now deprecated

There are three pairs of characters that Perl recognizes as
metacharacters in regular expression patterns: C<{}>, C<[]>, and C<()>.
These can be used as well to delimit patterns, as in:

  m{foo}
  s(foo)(bar)

Since they are metacharacters, they have special meaning to regular
expression patterns, and it turns out that you can't turn off that
special meaning by the normal means of preceding them with a backslash,
if you use them, paired, within a pattern delimited by them.  For
example, in

  m{foo\{1,3\}}

the backslashes do not change the behavior, and this matches
S<C<"f o">> followed by one to three more occurrences of C<"o">.

Usages like this, where they are interpreted as metacharacters, are
exceedingly rare; we think there are none, for example, in all of CPAN.
Hence, this deprecation should affect very little code.  It does give
notice, however, that any such code needs to change, which will in turn
allow us to change the behavior in future Perl versions so that the
backslashes do have an effect, and without fear that we are silently
breaking any existing code.

=head2 Splitting the tokens C<(?> and C<(*> in regular expressions

A deprecation warning is now raised if the C<(> and C<?> are separated
by white space or comments in C<(?...)> regular expression constructs.
Similarly, if the C<(> and C<*> are separated in C<(*VERB...)>
constructs.

=head2 Pre-PerlIO IO implementations

Perl supports being built without PerlIO proper, using a stdio or sfio
wrapper instead.  A perl build like this will not support IO layers and
thus Unicode IO, making it rather handicapped.

PerlIO supports a C<stdio> layer if stdio use is desired, and similarly a
sfio layer could be produced.

=head1 Future Deprecations

=over

=item *

Platforms without support infrastructure

Both Windows CE and z/OS have been historically under-maintained, and are
currently neither successfully building nor regularly being smoke tested.
Efforts are underway to change this situation, but it should not be taken for
granted that the platforms are safe and supported.  If they do not become
buildable and regularly smoked, support for them may be actively removed in
future releases.  If you have an interest in these platforms and you can lend
your time, expertise, or hardware to help support these platforms, please let
the perl development effort know by emailing C<perl5-porters@perl.org>.

Some platforms that appear otherwise entirely dead are also on the short list
for removal between now and 5.20.0:

=over

=item DG/UX

=item NeXT

=back

=item *

Swapping of $< and $>

For more information about this future deprecation, see L<the relevant RT
ticket|https://rt.perl.org/rt3/Ticket/Display.html?id=96212>.

=item *

C<microperl>, long broken and of unclear present purpose, will be removed.

=item *

Revamping C<< "\Q" >> semantics in double-quotish strings when combined with
other escapes.

There are several bugs and inconsistencies involving combinations
of C<\Q> and escapes like C<\x>, C<\L>, etc., within a C<\Q...\E> pair.
These need to be fixed, and doing so will necessarily change current
behavior.  The changes have not yet been settled.

=item *

Use of C<$^>, where C<^> stands for any actual (non-printing) C0 control
character will be disallowed in a future Perl version.  Use C<${^}>
instead (where again C<^> stands for a control character),
or better, C<$^A> , where C<^> this time is a caret (CIRCUMFLEX ACCENT),
and C<A> stands for any of the characters listed at the end of
L<perlebcdic/OPERATOR DIFFERENCES>.

=back

=head1 Performance Enhancements

=over 4

=item *

Lists of lexical variable declarations (C<my($x, $y)>) are now optimised
down to a single op and are hence faster than before.

=item *

A new C preprocessor define C<NO_TAINT_SUPPORT> was added that, if set,
disables Perl's taint support altogether.  Using the -T or -t command
line flags will cause a fatal error.  Beware that both core tests as
well as many a CPAN distribution's tests will fail with this change.  On
the upside, it provides a small performance benefit due to reduced
branching.

B<Do not enable this unless you know exactly what you are getting yourself
into.>

=item *

C<pack> with constant arguments is now constant folded in most cases
[perl #113470].

=item *

Speed up in regular expression matching against Unicode properties.  The
largest gain is for C<\X>, the Unicode "extended grapheme cluster."  The
gain for it is about 35% - 40%.  Bracketed character classes, e.g.,
C<[0-9\x{100}]> containing code points above 255 are also now faster.

=item *

On platforms supporting it, several former macros are now implemented as static
inline functions. This should speed things up slightly on non-GCC platforms.

=item *

The optimisation of hashes in boolean context has been extended to
affect C<scalar(%hash)>, C<%hash ? ... : ...>, and C<sub { %hash || ... }>.

=item *

Filetest operators manage the stack in a fractionally more efficient manner.

=item *

Globs used in a numeric context are now numified directly in most cases,
rather than being numified via stringification.

=item *

The C<x> repetition operator is now folded to a single constant at compile
time if called in scalar context with constant operands and no parentheses
around the left operand.

=back

=head1 Modules and Pragmata

=head2 New Modules and Pragmata

=over 4

=item *

L<Config::Perl::V> version 0.16 has been added as a dual-lifed module.
It provides structured data retrieval of C<perl -V> output including
information only known to the C<perl> binary and not available via L<Config>.

=back

=head2 Updated Modules and Pragmata

This is only an overview of selected module updates.  For a complete
list of updates, run:

  $ corelist --diff 5.16.0 5.18.0

You can substitute your favorite version in place of 5.16.0, too.

=over 4

=item *

L<XXX> has been upgraded from version A.xx to B.yy.

=back

=head2 Removed Modules and Pragmata

=over 

=item *

L<Version::Requirements> has been removed from the core distribution.  It is
available under a different name: L<CPAN::Meta::Requirements>.

=back

=head1 Documentation

=head2 Changes to Existing Documentation

=head3 L<perlcheat>

=over 4

=item *

L<perlcheat> has been reorganized, and a few new sections were added.

=back

=head3 L<perldata>

=over 4

=item *

Now explicitly documents the behaviour of hash initializer lists that
contain duplicate keys.

=back

=head3 L<perldiag>

=over 4

=item *

The explanation of symbolic references being prevented by "strict refs"
now doesn't assume that the reader knows what symbolic references are.

=back

=head3 L<perlfaq>

=over 4

=item *

L<perlfaq> has been synchronized with version 5.0150040 from CPAN.

=back

=head3 L<perlfunc>

=over 4

=item *

The return value of C<pipe> is now documented.

=item *

Clarified documentation of C<our>.

=back

=head3 L<perlop>

=over 4

=item *

Loop control verbs (C<dump>, C<goto>, C<next>, C<last> and C<redo>) have always
had the same precedence as assignment operators, but this was not documented
until now.

=back

=head3 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 L<perldiag>.

XXX New or changed warnings emitted by the core's C<C> code go here.  Also
include any changes in L<perldiag> that reconcile it to the C<C> code.

=head2 New Diagnostics

XXX Newly added diagnostic messages go under here, separated into New Errors
and New Warnings

=head3 New Errors

=over 4

=item *

L<Unterminated delimiter for here document|perldiag/"Unterminated delimiter for here document">

This message now occurs when a here document label has an initial quotation
mark but the final quotation mark is missing.

This replaces a bogus and misleading error message about not finding the label
itself [perl #114104].

=item *

L<panic: child pseudo-process was never scheduled|perldiag/"panic: child pseudo-process was never scheduled">

This error is thrown when a child pseudo-process in the ithreads implementation
on Windows was not scheduled within the time period allowed and therefore was
not able to initialize properly [perl #88840].

=item *

L<Group name must start with a non-digit word character in regex; marked by <-- HERE in mE<sol>%sE<sol>|perldiag/"Group name must start with a non-digit word character in regex; marked by <-- HERE in m/%s/">

This error has been added for C<(?&0)>, which is invalid.  It used to
produce an incomprehensible error message [perl #101666].

=item *

L<Can't use an undefined value as a subroutine reference|perldiag/"Can't use an undefined value as %s reference">

Calling an undefined value as a subroutine now produces this error message.
It used to, but was accidentally disabled, first in Perl 5.004 for
non-magical variables, and then in Perl 5.14 for magical (e.g., tied)
variables.  It has now been restored.  In the mean time, undef was treated
as an empty string [perl #113576].

=item *

L<Experimental "%s" subs not enabled|perldiag/"Experimental "%s" subs not enabled">

To use lexical subs, you must first enable them:

    no warnings 'experimental::lexical_subs';
    use feature 'lexical_subs';
    my sub foo { ... }

=back

=head3 New Warnings

=over 4

=item *

XXX: This needs more detail.

Strings with code points over 0xFF may not be mapped into in-memory file
handles

=item *

L<'%s' resolved to '\o{%s}%d'|perldiag/"'%s' resolved to '\o{%s}%d'">

=item *

L<'Trailing white-space in a charnames alias definition is deprecated'|perldiag/"Trailing white-space in a charnames alias definition is deprecated">

=item *

L<'A sequence of multiple spaces in a charnames alias definition is deprecated'|perldiag/"A sequence of multiple spaces in a charnames alias definition is deprecated">

=item *

L<'Passing malformed UTF-8 to "%s" is deprecated'|perldiag/"Passing malformed UTF-8 to "%s" is deprecated">

=item *

L<Subroutine "&%s" is not available|perldiag/"Subroutine "&%s" is not available">

(W closure) During compilation, an inner named subroutine or eval is
attempting to capture an outer lexical subroutine that is not currently
available.  This can happen for one of two reasons.  First, the lexical
subroutine may be declared in an outer anonymous subroutine that has not
yet been created.  (Remember that named subs are created at compile time,
while anonymous subs are created at run-time.)  For example,

    sub { my sub a {...} sub f { \&a } }

At the time that f is created, it can't capture the current the "a" sub,
since the anonymous subroutine hasn't been created yet.  Conversely, the
following won't give a warning since the anonymous subroutine has by now
been created and is live:

    sub { my sub a {...} eval 'sub f { \&a }' }->();

The second situation is caused by an eval accessing a variable that has
gone out of scope, for example,

    sub f {
        my sub a {...}
        sub { eval '\&a' }
    }
    f()->();

Here, when the '\&a' in the eval is being compiled, f() is not currently
being executed, so its &a is not available for capture.

=item *

L<"%s" subroutine &%s masks earlier declaration in same %s|perldiag/"%s" subroutine &%s masks earlier declaration in same %s>

(W misc) A "my" or "state" subroutine has been redeclared in the
current scope or statement, effectively eliminating all access to
the previous instance.  This is almost always a typographical error.
Note that the earlier subroutine will still exist until the end of
the scope or until all closure references to it are destroyed.

=item *

L<The %s feature is experimental|perldiag/"The %s feature is experimental">

(S experimental) This warning is emitted if you enable an experimental
feature via C<use feature>.  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::lexical_subs";
    use feature "lexical_subs";

=item *

L<sleep(%u) too large|perldiag/"sleep(%u) too large">

(W overflow) You called C<sleep> with a number that was larger than it can
reliably handle and C<sleep> probably slept for less time than requested.

=item *

L<Wide character in setenv|perldiag/"Wide character in %s">

Attempts to put wide characters into environment variables via C<%ENV> now
provoke this warning.

=item *

"L<Invalid negative number (%s) in chr|perldiag/"Invalid negative number (%s) in chr">"

C<chr()> now warns when passed a negative value [perl #83048].

=item *

"L<Integer overflow in srand|perldiag/"Integer overflow in srand">"

C<srand()> now warns when passed a value that doesn't fit in a C<UV> (since the
value will be truncated rather than overflowing) [perl #40605].

=item *

"L<-i used with no filenames on the command line, reading from STDIN|perldiag/"-i used with no filenames on the command line, reading from STDIN">"

Running perl with the C<-i> flag now warns if no input files are provided on
the command line [perl #113410].

=back

=head2 Changes to Existing Diagnostics

=over 4

=item *

L<$* is no longer supported|perldiag/"$* is no longer supported">

The warning that use of C<$*> and C<$#> is no longer supported is now
generated for every location that references them.  Previously it would fail
to be generated if another variable using the same typeglob was seen first
(e.g. C<@*> before C<$*>), and would not be generated for the second and
subsequent uses.  (It's hard to fix the failure to generate warnings at all
without also generating them every time, and warning every time is
consistent with the warnings that C<$[> used to generate.)

=item *

The warnings for C<\b{> and C<\B{> were added.  They are a deprecation
warning which should be turned off by that category.  One should not
have to turn off regular regexp warnings as well to get rid of these.

=item *

L<Constant(%s): Call to &{$^H{%s}} did not return a defined value|perldiag/Constant(%s): Call to &{$^H{%s}} did not return a defined value>

Constant overloading that returns C<undef> results in this error message.
For numeric constants, it used to say "Constant(undef)".  "undef" has been
replaced with the number itself.

=item *

The error produced when a module cannot be loaded now includes a hint that
the module may need to be installed: "Can't locate hopping.pm in @INC (you
may need to install the hopping module) (@INC contains: ...)"

=item *

L<vector argument not supported with alpha versions|perldiag/vector argument not supported with alpha versions>

This warning was not suppressable, even with C<no warnings>.  Now it is
suppressible, and has been moved from the "internal" category to the
"printf" category.

=item *

C<< Can't do {n,m} with n > m in regex; marked by <-- HERE in m/%s/ >>

This fatal error has been turned into a warning that reads:

L<< Quantifier {n,m} with n > m can't match in regex | perldiag/Quantifier {n,m} with n > m can't match in regex >>

(W regexp) Minima should be less than or equal to maxima.  If you really want
your regexp to match something 0 times, just put {0}.

=item *

The "Runaway prototype" warning that occurs in bizarre cases has been
removed as being unhelpful and inconsistent.

=item *

The "Not a format reference" error has been removed, as the only case in
which it could be triggered was a bug.

=item *

The "Unable to create sub named %s" error has been removed for the same
reason.

=item *

The 'Can't use "my %s" in sort comparison' error has been downgraded to a
warning, '"my %s" used in sort comparison' (with 'state' instead of 'my'
for state variables).  In addition, the heuristics for guessing whether
lexical $a or $b has been misused have been improved to generate fewer
false positives.  Lexical $a and $b are no longer disallowed if they are
outside the sort block.  Also, a named unary or list operator inside the
sort block no longer causes the $a or $b to be ignored [perl #86136].

=back

=head1 Utility Changes

=head3 L<h2xs>

=over 4

=item *

F<h2xs> no longer produces invalid code for empty defines.  [perl #20636]

=back

=head1 Configuration and Compilation

=over 4

=item *

Added C<useversionedarchname> option to Configure

When set, it includes 'api_versionstring' in 'archname'. E.g.
x86_64-linux-5.13.6-thread-multi.  It is unset by default.

This feature was requested by Tim Bunce, who observed that
C<INSTALL_BASE> creates a library structure that does not
differentiate by perl version.  Instead, it places architecture
specific files in "$install_base/lib/perl5/$archname".  This makes
it difficult to use a common C<INSTALL_BASE> library path with
multiple versions of perl.

By setting C<-Duseversionedarchname>, the $archname will be
distinct for architecture I<and> API version, allowing mixed use of
C<INSTALL_BASE>.

=item *

Add a C<PERL_NO_INLINE_FUNCTIONS> option

If C<PERL_NO_INLINE_FUNCTIONS> is defined, don't include "inline.h"

This permits test code to include the perl headers for definitions without
creating a link dependency on the perl library (which may not exist yet).

=item *

Configure will honour the external C<MAILDOMAIN> environment variable, if set.

=item *

C<installman> no longer ignores the silent option

=item *

Both C<META.yml> and C<META.json> files are now included in the distribution.

=item *

F<Configure> will now correctly detect C<isblank()> when compiling with a C++
compiler.

=item *

The pager detection in F<Configure> has been improved to allow responses which
specify options after the program name, e.g. B</usr/bin/less -R>, if the user
accepts the default value.  This helps B<perldoc> when handling ANSI escapes
[perl #72156].

=back

=head1 Testing

=over 4

=item *

The test suite now has a section for tests that require very large amounts
of memory.  These tests won't run by default; they can be enabled by
setting the C<PERL_TEST_MEMORY> environment variable to the number of
gibibytes of memory that may be safely used.

=back

=head1 Platform Support

=head2 Discontinued Platforms

=over 4

=item BeOS

BeOS was an operating system for personal computers developed by Be Inc,
initially for their BeBox hardware. The OS Haiku was written as an open
source replacement for/continuation of BeOS, and its perl port is current and
actively maintained.

=item UTS Global

Support code relating to UTS global has been removed.  UTS was a mainframe
version of System V created by Amdahl, subsequently sold to UTS Global.  The
port has not been touched since before Perl 5.8.0, and UTS Global is now
defunct.

=item VM/ESA

Support for VM/ESA has been removed. The port was tested on 2.3.0, which
IBM ended service on in March 2002. 2.4.0 ended service in June 2003, and
was superseded by Z/VM. The current version of Z/VM is V6.2.0, and scheduled
for end of service on 2015/04/30.

=item MPE/IX

Support for MPE/IX has been removed.

=item EPOC

Support code relating to EPOC has been removed.  EPOC was a family of
operating systems developed by Psion for mobile devices.  It was the
predecessor of Symbian.  The port was last updated in April 2002.

=item Rhapsody

Support for Rhapsody has been removed.

=back

=head2 Platform-Specific Notes

=head3 AIX

Configure now always adds C<-qlanglvl=extc99> to the CC flags on AIX when
using xlC.  This will make it easier to compile a number of XS-based modules
that assume C99 [perl #113778].

=head3 clang++

There is now a workaround for a compiler bug that prevented compiling
with clang++ since Perl 5.15.7 [perl #112786].

=head3 C++

When compiling the Perl core as C++ (which is only semi-supported), the
mathom functions are now compiled as C<extern "C">, to ensure proper
binary compatibility.  (However, binary compatibility isn't generally
guaranteed anyway in the situations where this would matter.)

=head3 Darwin

Stop hardcoding an alignment on 8 byte boundaries to fix builds using
-Dusemorebits.

=head3 Haiku

Perl should now work out of the box on Haiku R1 Alpha 4.

=head3 MidnightBSD

C<libc_r> was removed from recent versions of MidnightBSD and older versions
work better with C<pthread>. Threading is now enabled using C<pthread> which
corrects build errors with threading enabled on 0.4-CURRENT.

=head3 Solaris

In Configure, avoid running sed commands with flags not supported on Solaris.

=head3 VMS

=over

=item *

Where possible, the case of filenames and command-line arguments is now
preserved by enabling the CRTL features C<DECC$EFS_CASE_PRESERVE> and
C<DECC$ARGV_PARSE_STYLE> at start-up time.  The latter only takes effect
when extended parse is enabled in the process from which Perl is run.

=item *

The character set for Extended Filename Syntax (EFS) is now enabled by default
on VMS.  Among other things, this provides better handling of dots in directory
names, multiple dots in filenames, and spaces in filenames.  To obtain the old
behavior, set the logical name C<DECC$EFS_CHARSET> to C<DISABLE>.

=item *

Fixed linking on builds configured with C<-Dusemymalloc=y>.

=item *

Experimental support for building Perl with the HP C++ compiler is available
by configuring with C<-Dusecxx>.

=item *

All C header files from the top-level directory of the distribution are now
installed on VMS, providing consistency with a long-standing practice on other
platforms. Previously only a subset were installed, which broke non-core
extension builds for extensions that depended on the missing include files.

=item *

Quotes are now removed from the command verb (but not the parameters) for
commands spawned via C<system>, backticks, or a piped C<open>.  Previously,
quotes on the verb were passed through to DCL, which would fail to recognize
the command.  Also, if the verb is actually a path to an image or command
procedure on an ODS-5 volume, quoting it now allows the path to contain spaces.

=item *

The B<a2p> build has been fixed for the HP C++ compiler on OpenVMS.

=back

=head3 Win32

=over

=item *

Perl can now be built using Microsoft's Visual C++ 2012 compiler by specifying
CCTYPE=MSVC110 (or MSVC110FREE if you are using the free Express edition for
Windows Desktop) in F<win32/Makefile>.

=item *

The option to build without C<USE_SOCKETS_AS_HANDLES> has been removed.

=item *

Fixed a problem where perl could crash while cleaning up threads (including the
main thread) in threaded debugging builds on Win32 and possibly other platforms
[perl #114496].

=item *

A rare race condition that would lead to L<sleep|perlfunc/sleep> taking more
time than requested, and possibly even hanging, has been fixed [perl #33096].

=item *

C<link> on Win32 now attempts to set C<$!> to more appropriate values
based on the Win32 API error code. [perl #112272]

Perl no longer mangles the environment block, e.g. when launching a new
sub-process, when the environment contains non-ASCII characters. Known
problems still remain, however, when the environment contains characters
outside of the current ANSI codepage (e.g. see the item about Unicode in
C<%ENV> in L<http://perl5.git.perl.org/perl.git/blob/HEAD:/Porting/todo.pod>).
[perl #113536]

=item *

Building perl with some Windows compilers used to fail due to a problem
with miniperl's C<glob> operator (which uses the C<perlglob> program)
deleting the PATH environment variable [perl #113798].

=item *

A new makefile option, C<USE_64_BIT_INT>, has been added to the Windows
makefiles.  Set this to "define" when building a 32-bit perl if you want
it to use 64-bit integers.

Machine code size reductions, already made to the DLLs of XS modules in
Perl 5.17.2, have now been extended to the perl DLL itself.

Building with VC++ 6.0 was inadvertently broken in Perl 5.17.2 but has
now been fixed again.

=back

=head3 WinCE

Building on WinCE is now possible once again, although more work is required
to fully restore a clean build.

=head1 Internal Changes

=over

=item *

Synonyms for the misleadingly named C<av_len()> have been created:
C<av_top_index()> and C<av_tindex>.  All three of these return the
number of the highest index in the array, not the number of elements it
contains.

=item *

SvUPGRADE() is no longer an expression. Originally this macro (and its
underlying function, sv_upgrade()) were documented as boolean, although
in reality they always croaked on error and never returned false. In 2005
the documentation was updated to specify a void return value, but
SvUPGRADE() was left always returning 1 for backwards compatibility. This
has now been removed, and SvUPGRADE() is now a statement with no return
value.

So this is now a syntax error:

    if (!SvUPGRADE(sv)) { croak(...); }

If you have code like that, simply replace it with

    SvUPGRADE(sv);

or to to avoid compiler warnings with older perls, possibly

    (void)SvUPGRADE(sv);

=item *

Perl has a new copy-on-write mechanism that allows any SvPOK scalar to be
upgraded to a copy-on-write scalar.  A reference count on the string buffer
is stored in the string buffer itself.

This breaks a few XS modules by allowing copy-on-write scalars to go
through code paths that never encountered them before.

This behaviour can still be disabled by running F<Configure> with
B<-Accflags=-DPERL_NO_COW>.  This option will probably be removed in Perl
5.20.

=item *

Copy-on-write no longer uses the SvFAKE and SvREADONLY flags.  Hence,
SvREADONLY indicates a true read-only SV.

Use the SvIsCOW macro (as before) to identify a copy-on-write scalar.

=item *

C<PL_glob_index> is gone.

=item *

The private Perl_croak_no_modify has had its context parameter removed.  It is
now has a void prototype.  Users of the public API croak_no_modify remain
unaffected.

=item *

Copy-on-write (shared hash key) scalars are no longer marked read-only.
C<SvREADONLY> returns false on such an SV, but C<SvIsCOW> still returns
true.

=item *

A new op type, C<OP_PADRANGE> has been introduced.  The perl peephole
optimiser will, where possible, substitute a single padrange op for a
pushmark followed by one or more pad ops, and possibly also skipping list
and nextstate ops.  In addition, the op can carry out the tasks associated
with the RHS of a C<< my(...) = @_ >> assignment, so those ops may be optimised
away too.

=item *

Case-insensitive matching inside a [bracketed] character class with a
multi-character fold no longer excludes one of the possibilities in the
circumstances that it used to. [perl #89774].

=item *

C<PL_formfeed> has been removed.

=item *

The regular expression engine no longer reads one byte past the end of the
target string.  While for all internally well-formed scalars this should
never have been a problem, this change facilitates clever tricks with
string buffers in CPAN modules.  [perl #73542]

=item *

Inside a BEGIN block, C<PL_compcv> now points to the currently-compiling
subroutine, rather than the BEGIN block itself.

=item *

C<mg_length> has been deprecated.

=item *

C<sv_len> now always returns a byte count and C<sv_len_utf8> a character
count.  Previously, C<sv_len> and C<sv_len_utf8> were both buggy and would
sometimes returns bytes and sometimes characters.  C<sv_len_utf8> no longer
assumes that its argument is in UTF8.  Neither of these creates UTF8 caches
for tied or overloaded values or for non-PVs any more.

=item *

C<sv_mortalcopy> now copies string buffers of shared hash key scalars when
called from XS modules [perl #79824].

=item *

C<RXf_SPLIT> and C<RXf_SKIPWHITE> are no longer used.  They are now
#defined as 0.

=item *

The new C<RXf_MODIFIES_VARS> flag can be set by custom regular expression
engines to indicate that the execution of the regular expression may cause
variables to be modified.  This lets C<s///> know to skip certain
optimisations.  Perl's own regular expression engine sets this flag for the
special backtracking verbs that set $REGMARK and $REGERROR.

=item *

The APIs for accessing lexical pads have changed considerably.

C<PADLIST>s are now longer C<AV>s, but their own type instead.
C<PADLIST>s now contain a C<PAD> and a C<PADNAMELIST> of C<PADNAME>s,
rather than C<AV>s for the pad and the list of pad names.  C<PAD>s,
C<PADNAMELIST>s, and C<PADNAME>s are to be accessed as such through the
newly added pad API instead of the plain C<AV> and C<SV> APIs.  See
L<perlapi> for details.

=item *

In the regex API, the numbered capture callbacks are passed an index
indicating what match variable is being accessed. There are special
index values for the C<$`, $&, $&> variables. Previously the same three
values were used to retrieve C<${^PREMATCH}, ${^MATCH}, ${^POSTMATCH}>
too, but these have now been assigned three separate values. See
L<perlreapi/Numbered capture callbacks>.

=item *

C<PL_sawampersand> was previously a boolean indicating that any of
C<$`, $&, $&> had been seen; it now contains three one-bit flags
indicating the presence of each of the variables individually.

=item *

The C<CV *> typemap entry now supports C<&{}> overloading and typeglobs,
just like C<&{...}> [perl #96872].

=item *

The C<SVf_AMAGIC> flag to indicate overloading is now on the stash, not the
object.  It is now set automatically whenever a method or @ISA changes, so
its meaning has changed, too.  It now means "potentially overloaded".  When
the overload table is calculated, the flag is automatically turned off if
there is no overloading, so there should be no noticeable slowdown.

The staleness of the overload tables is now checked when overload methods
are invoked, rather than during C<bless>.

"A" magic is gone.  The changes to the handling of the C<SVf_AMAGIC> flag
eliminate the need for it.

C<PL_amagic_generation> has been removed as no longer necessary.  For XS
modules, it is now a macro alias to C<PL_na>.

The fallback overload setting is now stored in a stash entry separate from
overloadedness itself.

=item *

The character-processing code has been cleaned up in places.  The changes
should be operationally invisible.

=item *

The C<study> function was made a no-op in 5.16.  It was simply disabled via
a C<return> statement; the code was left in place.  Now the code supporting
what C<study> used to do has been removed.

=item *

Under threaded perls, there is no longer a separate PV allocated for every
COP to store its package name (C<< cop->stashpv >>).  Instead, there is an
offset (C<< cop->stashoff >>) into the new C<PL_stashpad> array, which
holds stash pointers.

=item *

In the pluggable regex API, the C<regexp_engine> struct has acquired a new
field C<op_comp>, which is currently just for perl's internal use, and
should be initialized to NULL by other regex plugin modules.

=item *

A new function C<alloccoptash> has been added to the API, but is considered
experimental.  See L<perlapi>.

=item *

Perl used to implement get magic in a way that would sometimes hide bugs in
code that could call mg_get() too many times on magical values.  This hiding of
errors no longer occurs, so long-standing bugs may become visible now.  If
you see magic-related errors in XS code, check to make sure it, together
with the Perl API functions it uses, calls mg_get() only once on SvGMAGICAL()
values.

=item *

OP allocation for CVs now uses a slab allocator.  This simplifies
memory management for OPs allocated to a CV, so cleaning up after a
compilation error is simpler and safer [perl #111462][perl #112312].

=item *

C<PERL_DEBUG_READONLY_OPS> has been rewritten to work with the new slab
allocator, allowing it to catch more violations than before.

=item *

The old slab allocator for ops, which was only enabled for C<PERL_IMPLICIT_SYS>
and C<PERL_DEBUG_READONLY_OPS>, has been retired.

=back

=head1 Selected Bug Fixes

=over 4

=item *

Here-doc terminators no longer require a terminating newline character when
they occur at the end of a file.  This was already the case at the end of a
string eval [perl #65838].

=item *

C<-DPERL_GLOBAL_STRUCT> builds now free the global struct B<after>
they've finished using it.

=item *

A trailing '/' on a path in @INC will no longer have an additional '/'
appended.

=item *

The C<:crlf> layer now works when unread data doesn't fit into its own
buffer. [perl #112244].

=item *

C<ungetc()> now handles UTF-8 encoded data. [perl #116322].

=item *

A bug in the core typemap caused any C types that map to the T_BOOL core
typemap entry to not be set, updated, or modified when the T_BOOL variable was
used in an OUTPUT: section with an exception for RETVAL. T_BOOL in an INPUT:
section was not affected. Using a T_BOOL return type for an XSUB (RETVAL)
was not affected. A side effect of fixing this bug is, if a T_BOOL is specified
in the OUTPUT: section (which previous did nothing to the SV), and a read only
SV (literal) is passed to the XSUB, croaks like "Modification of a read-only
value attempted" will happen. [perl #115796]

=item *

On many platforms, providing a directory name as the script name caused perl
to do nothing and report success.  It should now universally report an error
and exit nonzero. [perl #61362]

=item *

C<sort {undef} ...> under fatal warnings no longer crashes.  It had
begun crashing in Perl 5.16.

=item *

Stashes blessed into each other
(C<bless \%Foo::, 'Bar'; bless \%Bar::, 'Foo'>) no longer result in double
frees.  This bug started happening in Perl 5.16.

=item *

Numerous memory leaks have been fixed, mostly involving fatal warnings and
syntax errors.

=item *

Some failed regular expression matches such as C<'f' =~ /../g> were not
resetting C<pos>.  Also, "match-once" patterns (C<m?...?g>) failed to reset
it, too, when invoked a second time [perl #23180].

=item *

Accessing C<$&> after a pattern match now works if it had not been seen
before the match.  I.e., this applies to C<${'&'}> (under C<no strict>) and
C<eval '$&'>.  The same applies to C<$'> and C<$`> [perl #4289].

=item *

Several bugs involving C<local *ISA> and C<local *Foo::> causing stale
MRO caches have been fixed.

=item *

Defining a subroutine when its typeglob has been aliased no longer results
in stale method caches.  This bug was introduced in Perl 5.10.

=item *

Localising a typeglob containing a subroutine when the typeglob's package
has been deleted from its parent stash no longer produces an error.  This
bug was introduced in Perl 5.14.

=item *

Under some circumstances, C<local *method=...> would fail to reset method
caches upon scope exit.

=item *

C</[.foo.]/> is no longer an error, but produces a warning (as before) and
is treated as C</[.fo]/> [perl #115818].

=item *

C<goto $tied_var> now calls FETCH before deciding what type of goto
(subroutine or label) this is.

=item *

Renaming packages through glob assignment
(C<*Foo:: = *Bar::; *Bar:: = *Baz::>) in combination with C<m?...?> and
C<reset> no longer makes threaded builds crash.

=item *

A number of bugs related to assigning a list to hash have been fixed. Many of
these involve lists with repeated keys like C<(1, 1, 1, 1)>.

=over 4

=item *

The expression C<scalar(%h = (1, 1, 1, 1))> now returns C<4>, not C<2>.

=item *

The return value of C<%h = (1, 1, 1)> in list context was wrong. Previously
this would return C<(1, undef, 1)>, now it returns C<(1, undef)>.

=item *

Perl now issues the same warning on C<($s, %h) = (1, {})> as it does for
C<(%h) = ({})>, "Reference found where even-sized list expected".

=item *

A number of additional edge cases in list assignment to hashes were
corrected. For more details see commit 23b7025ebc.

=back

=item *

Attributes applied to lexical variables no longer leak memory.
[perl #114764]

=item *

C<dump>, C<goto>, C<last>, C<next>, C<redo> or C<require> followed by a
bareword (or version) and then an infix operator is no longer a syntax
error.  It used to be for those infix operators (like C<+>) that have a
different meaning where a term is expected.  [perl #105924]

=item *

C<require a::b . 1> and C<require a::b + 1> no longer produce erroneous
ambiguity warnings.  [perl #107002]

=item *

Class method calls are now allowed on any string, and not just strings
beginning with an alphanumeric character.  [perl #105922]

=item *

An empty pattern created with C<qr//> used in C<m///> no longer triggers
the "empty pattern reuses last pattern" behaviour.  [perl #96230]

=item *

Tying a hash during iteration no longer results in a memory leak.

=item *

Freeing a tied hash during iteration no longer results in a memory leak.

=item *

List assignment to a tied array or hash that dies on STORE no longer
results in a memory leak.

=item *

If the hint hash (C<%^H>) is tied, compile-time scope entry (which copies
the hint hash) no longer leaks memory if FETCH dies.  [perl #107000]

=item *

Constant folding no longer inappropriately triggers the special
C<split " "> behaviour.  [perl #94490]

=item *

C<defined scalar(@array)>, C<defined do { &foo }>, and similar constructs
now treat the argument to C<defined> as a simple scalar.  [perl #97466]

=item *

Running a custom debugging that defines no C<*DB::DB> glob or provides a
subroutine stub for C<&DB::DB> no longer results in a crash, but an error
instead.  [perl #114990]

=item *

C<reset ""> now matches its documentation.  C<reset> only resets C<m?...?>
patterns when called with no argument.  An empty string for an argument now
does nothing.  (It used to be treated as no argument.)  [perl #97958]

=item *

C<printf> with an argument returning an empty list no longer reads past the
end of the stack, resulting in erratic behaviour.  [perl #77094]

=item *

C<--subname> no longer produces erroneous ambiguity warnings.
[perl #77240]

=item *

C<v10> is now allowed as a label or package name.  This was inadvertently
broken when v-strings were added in Perl 5.6.  [perl #56880]

=item *

C<length>, C<pos>, C<substr> and C<sprintf> could be confused by ties,
overloading, references and typeglobs if the stringification of such
changed the internal representation to or from UTF8.  [perl #114410]

=item *

utf8::encode now calls FETCH and STORE on tied variables.  utf8::decode now
calls STORE (it was already calling FETCH).

=item *

C<$tied =~ s/$non_utf8/$utf8/> no longer loops infinitely if the tied
variable returns a Latin-1 string, shared hash key scalar, or reference or
typeglob that stringifies as ASCII or Latin-1.  This was a regression from
5.12.x.

=item *

C<s///> without /e is now better at detecting when it needs to forego
certain optimisations, fixing some buggy cases:

=over

=item *

Match variables in certain constructs (C<&&>, C<||>, C<..> and others) in
the replacement part; e.g., C<s/(.)/$l{$a||$1}/g>.  [perl #26986]

=item *

Aliases to match variables in the replacement.

=item *

C<$REGERROR> or C<$REGMARK> in the replacement.  [perl #49190]

=item *

An empty pattern (C<s//$foo/>) that causes the last-successful pattern to
be used, when that pattern contains code blocks that modify the variables
in the replacement.

=back

=item *

The taintedness of the replacement string no longer affects the taintedness
of the return value of C<s///e>.

=item *

The C<$|> autoflush variable is created on-the-fly when needed.  If this
happened (e.g., if it was mentioned in a module or eval) when the
currently-selected filehandle was a typeglob with an empty IO slot, it used
to crash.  [perl #115206]

=item *

Line numbers at the end of a string eval are no longer off by one.
[perl #114658]

=item *

@INC filters (subroutines returned by subroutines in @INC) that set $_ to a
copy-on-write scalar no longer cause the parser to modify that string
buffer in place.

=item *

C<length($object)> no longer returns the undefined value if the object has
string overloading that returns undef.  [perl #115260]

=item *

The use of C<PL_stashcache>, the stash name lookup cache for method calls, has
been restored,

Commit da6b625f78f5f133 in August 2011 inadvertently broke the code that looks
up values in C<PL_stashcache>. As it's a only cache, quite correctly everything
carried on working without it.

=item *

The error "Can't localize through a reference" had disappeared in 5.16.0
when C<local %$ref> appeared on the last line of an lvalue subroutine.
This error disappeared for C<\local %$ref> in perl 5.8.1.  It has now
been restored.

=item *

The parsing of here-docs has been improved significantly, fixing several
parsing bugs and crashes and one memory leak, and correcting wrong
subsequent line numbers under certain conditions.

=item *

Inside an eval, the error message for an unterminated here-doc no longer
has a newline in the middle of it [perl #70836].

=item *

A substitution inside a substitution pattern (C<s/${s|||}//>) no longer
confuses the parser.

=item *

It may be an odd place to allow comments, but C<s//"" # hello/e> has
always worked, I<unless> there happens to be a null character before the
first #.  Now it works even in the presence of nulls.

=item *

An invalid range in C<tr///> or C<y///> no longer results in a memory leak.

=item *

String eval no longer treats a semicolon-delimited quote-like operator at
the very end (C<eval 'q;;'>) as a syntax error.

=item *

C<< warn {$_ => 1} + 1 >> is no longer a syntax error.  The parser used to
get confused with certain list operators followed by an anonymous hash and
then an infix operator that shares its form with a unary operator.

=item *

C<(caller $n)[6]> (which gives the text of the eval) used to return the
actual parser buffer.  Modifying it could result in crashes.  Now it always
returns a copy.  The string returned no longer has "\n;" tacked on to the
end.  The returned text also includes here-doc bodies, which used to be
omitted.

=item *

Reset the utf8 position cache when accessing magical variables to avoid the
string buffer and the utf8 position cache getting out of sync
[perl #114410].

=item *

Various cases of get magic being called twice for magical utf8 strings have been
fixed.

=item *

This code (when not in the presence of C<$&> etc)

    $_ = 'x' x 1_000_000;
    1 while /(.)/;

used to skip the buffer copy for performance reasons, but suffered from C<$1>
etc changing if the original string changed.  That's now been fixed.

=item *

Perl doesn't use PerlIO anymore to report out of memory messages, as PerlIO
might attempt to allocate more memory.

=item *

In a regular expression, if something is quantified with C<{n,m}> where
C<S<n E<gt> m>>, it can't possibly match.  Previously this was a fatal
error, but now is merely a warning (and that something won't match).
[perl #82954].

=item *

It used to be possible for formats defined in subroutines that have
subsequently been undefined and redefined to close over variables in the
wrong pad (the newly-defined enclosing sub), resulting in crashes or
"Bizarre copy" errors.

=item *

Redefinition of XSUBs at run time could produce warnings with the wrong
line number.

=item *

The %vd sprintf format does not support version objects for alpha versions.
It used to output the format itself (%vd) when passed an alpha version, and
also emit an "Invalid conversion in printf" warning.  It no longer does,
but produces the empty string in the output.  It also no longer leaks
memory in this case.

=item *

C<< $obj->SUPER::method >> calls in the main package could fail if the
SUPER package had already been accessed by other means.

=item *

Stash aliasing (C<< *foo:: = *bar:: >>) no longer causes SUPER calls to ignore
changes to methods or @ISA or use the wrong package.

=item *

Method calls on packages whose names end in ::SUPER are no longer treated
as SUPER method calls, resulting in failure to find the method.
Furthermore, defining subroutines in such packages no longer causes them to
be found by SUPER method calls on the containing package [perl #114924].

=item *

C<\w> now matches the code points U+200C (ZERO WIDTH NON-JOINER) and U+200D
(ZERO WIDTH JOINER).  C<\W> no longer matches these.  This change is because
Unicode corrected their definition of what C<\w> should match.

=item *

C<dump LABEL> no longer leaks its label.

=item *

Constant folding no longer changes the behaviour of functions like C<stat()>
and C<truncate()> that can take either filenames or handles.
C<stat 1 ? foo : bar> nows treats its argument as a file name (since it is an
arbitrary expression), rather than the handle "foo".

=item *

C<truncate FOO, $len> no longer falls back to treating "FOO" as a file name if
the filehandle has been deleted.  This was broken in Perl 5.16.0.

=item *

Subroutine redefinitions after sub-to-glob and glob-to-glob assignments no
longer cause double frees or panic messages.

=item *

C<s///> now turns vstrings into plain strings when performing a substitution,
even if the resulting string is the same (C<s/a/a/>).

=item *

Prototype mismatch warnings no longer erroneously treat constant subs as having
no prototype when they actually have "".

=item *

Constant subroutines and forward declarations no longer prevent prototype
mismatch warnings from omitting the sub name.

=item *

C<undef> on a subroutine now clears call checkers.

=item *

The C<ref> operator started leaking memory on blessed objects in Perl 5.16.0.
This has been fixed [perl #114340].

=item *

C<use> no longer tries to parse its arguments as a statement, making
C<use constant { () };> a syntax error [perl #114222].

=item *

On debugging builds, "uninitialized" warnings inside formats no longer cause
assertion failures.

=item *

On debugging builds, subroutines nested inside formats no longer cause
assertion failures [perl #78550].

=item *

Formats and C<use> statements are now permitted inside formats.

=item *

C<print $x> and C<sub { print $x }-E<gt>()> now always produce the same output.
It was possible for the latter to refuse to close over $x if the variable was
not active; e.g., if it was defined outside a currently-running named
subroutine.

=item *

Similarly, C<print $x> and C<print eval '$x'> now produce the same output.
This also allows "my $x if 0" variables to be seen in the debugger [perl
#114018].

=item *

Formats called recursively no longer stomp on their own lexical variables, but
each recursive call has its own set of lexicals.

=item *

Attempting to free an active format or the handle associated with it no longer
results in a crash.

=item *

Format parsing no longer gets confused by braces, semicolons and low-precedence
operators.  It used to be possible to use braces as format delimiters (instead
of C<=> and C<.>), but only sometimes.  Semicolons and low-precedence operators
in format argument lines no longer confuse the parser into ignoring the line's
return value.  In format argument lines, braces can now be used for anonymous
hashes, instead of being treated always as C<do> blocks.

=item *

Formats can now be nested inside code blocks in regular expressions and other
quoted constructs (C</(?{...})/> and C<qq/${...}/>) [perl #114040].

=item *

Formats are no longer created after compilation errors.

=item *

Under debugging builds, the B<-DA> command line option started crashing in Perl
5.16.0.  It has been fixed [perl #114368].

=item *

A potential deadlock scenario involving the premature termination of a pseudo-
forked child in a Windows build with ithreads enabled has been fixed.  This
resolves the common problem of the F<t/op/fork.t> test hanging on Windows [perl
#88840].

=item *

The code which generates errors from C<require()> could potentially read one or
two bytes before the start of the filename for filenames less than three bytes
long and ending C</\.p?\z/>.  This has now been fixed.  Note that it could
never have happened with module names given to C<use()> or C<require()> anyway.

=item *

The handling of pathnames of modules given to C<require()> has been made
thread-safe on VMS.

=item *

Non-blocking sockets have been fixed on VMS.

=item *

A bug in the compilation of a C</(?{})/> expression which affected the TryCatch
test suite has been fixed [perl #114242].

=item *

Pod can now be nested in code inside a quoted construct outside of a string
eval.  This used to work only within string evals [perl #114040].

=item *

C<goto ''> now looks for an empty label, producing the "goto must have
label" error message, instead of exiting the program [perl #111794].

=item *

C<goto "\0"> now dies with "Can't find label" instead of "goto must have
label".

=item *

The C function C<hv_store> used to result in crashes when used on C<%^H>
[perl #111000].

=item *

A call checker attached to a closure prototype via C<cv_set_call_checker>
is now copied to closures cloned from it.  So C<cv_set_call_checker> now
works inside an attribute handler for a closure.

=item *

Writing to C<$^N> used to have no effect.  Now it croaks with "Modification
of a read-only value" by default, but that can be overridden by a custom
regular expression engine, as with C<$1> [perl #112184].

=item *

C<undef> on a control character glob (C<undef *^H>) no longer emits an
erroneous warning about ambiguity [perl #112456].

=item *

For efficiency's sake, many operators and built-in functions return the
same scalar each time.  Lvalue subroutines and subroutines in the CORE::
namespace were allowing this implementation detail to leak through.
C<print &CORE::uc("a"), &CORE::uc("b")> used to print "BB".  The same thing
would happen with an lvalue subroutine returning the return value of C<uc>.
Now the value is copied in such cases.

=item *

C<method {}> syntax with an empty block or a block returning an empty list
used to crash or use some random value left on the stack as its invocant.
Now it produces an error.

=item *

C<vec> now works with extremely large offsets (E<gt>2 GB) [perl #111730].

=item *

Changes to overload settings now take effect immediately, as do changes to
inheritance that affect overloading.  They used to take effect only after
C<bless>.

Objects that were created before a class had any overloading used to remain
non-overloaded even if the class gained overloading through C<use overload>
or @ISA changes, and even after C<bless>.  This has been fixed
[perl #112708].

=item *

Classes with overloading can now inherit fallback values.

=item *

Overloading was not respecting a fallback value of 0 if there were
overloaded objects on both sides of an assignment operator like C<+=>
[perl #111856].

=item *

C<pos> now croaks with hash and array arguments, instead of producing
erroneous warnings.

=item *

C<while(each %h)> now implies C<while(defined($_ = each %h))>, like
C<readline> and C<readdir>.

=item *

Subs in the CORE:: namespace no longer crash after C<undef *_> when called
with no argument list (C<&CORE::time> with no parentheses).

=item *

C<unpack> no longer produces the "'/' must follow a numeric type in unpack"
error when it is the data that are at fault [perl #60204].

=item *

C<join> and C<"@array"> now call FETCH only once on a tied C<$">
[perl #8931].

=item *

Some subroutine calls generated by compiling core ops affected by a
C<CORE::GLOBAL> override had op checking performed twice.  The checking
is always idempotent for pure Perl code, but the double checking can
matter when custom call checkers are involved.

=item *

A race condition used to exist around fork that could cause a signal sent to
the parent to be handled by both parent and child. Signals are now blocked
briefly around fork to prevent this from happening [perl #82580].

=item *

The implementation of code blocks in regular expressions, such as C<(?{})>
and C<(??{})>, has been heavily reworked to eliminate a whole slew of bugs.
The main user-visible changes are:

=over 4

=item *

Code blocks within patterns are now parsed in the same pass as the
surrounding code; in particular it is no longer necessary to have balanced
braces: this now works:

    /(?{  $x='{'  })/

This means that this error message is no longer generated:

    Sequence (?{...}) not terminated or not {}-balanced in regex

but a new error may be seen:

    Sequence (?{...}) not terminated with ')'

In addition, literal code blocks within run-time patterns are only
compiled once, at perl compile-time:

    for my $p (...) {
        # this 'FOO' block of code is compiled once,
	# at the same time as the surrounding 'for' loop
        /$p{(?{FOO;})/;
    }

=item *

Lexical variables are now sane as regards scope, recursion and closure
behavior. In particular, C</A(?{B})C/> behaves (from a closure viewpoint)
exactly like C</A/ && do { B } && /C/>, while  C<qr/A(?{B})C/> is like
C<sub {/A/ && do { B } && /C/}>. So this code now works how you might
expect, creating three regexes that match 0, 1, and 2:

    for my $i (0..2) {
        push @r, qr/^(??{$i})$/;
    }
    "1" =~ $r[1]; # matches

=item *

The C<use re 'eval'> pragma is now only required for code blocks defined
at runtime; in particular in the following, the text of the C<$r> pattern is
still interpolated into the new pattern and recompiled, but the individual
compiled code-blocks within C<$r> are reused rather than being recompiled,
and C<use re 'eval'> isn't needed any more:

    my $r = qr/abc(?{....})def/;
    /xyz$r/;

=item *

Flow control operators no longer crash. Each code block runs in a new
dynamic scope, so C<next> etc. will not see
any enclosing loops. C<return> returns a value
from the code block, not from any enclosing subroutine.

=item *

Perl normally caches the compilation of run-time patterns, and doesn't
recompile if the pattern hasn't changed, but this is now disabled if
required for the correct behavior of closures. For example:

    my $code = '(??{$x})';
    for my $x (1..3) {
	# recompile to see fresh value of $x each time
        $x =~ /$code/;
    }

=item *

The C</msix> and C<(?msix)> etc. flags are now propagated into the return
value from C<(??{})>; this now works:

    "AB" =~ /a(??{'b'})/i;

=item *

Warnings and errors will appear to come from the surrounding code (or for
run-time code blocks, from an eval) rather than from an C<re_eval>:

    use re 'eval'; $c = '(?{ warn "foo" })'; /$c/;
    /(?{ warn "foo" })/;

formerly gave:

    foo at (re_eval 1) line 1.
    foo at (re_eval 2) line 1.

and now gives:

    foo at (eval 1) line 1.
    foo at /some/prog line 2.

=back

=item *

Perl now can be recompiled to use any Unicode version.  In v5.16, it
worked on Unicodes 6.0 and 6.1, but there were various bugs if earlier
releases were used; the older the release the more problems.

=item *

C<vec> no longer produces "uninitialized" warnings in lvalue context
[perl #9423].

=item *

An optimization involving fixed strings in regular expressions could cause
a severe performance penalty in edge cases.  This has been fixed
[perl #76546].

=item *

In certain cases, including empty subpatterns within a regular expression (such
as C<(?:)> or C<(?:|)>) could disable some optimizations. This has been fixed.

=item *

The "Can't find an opnumber" message that C<prototype> produces when passed
a string like "CORE::nonexistent_keyword" now passes UTF-8 and embedded
NULs through unchanged [perl #97478].

=item *

C<prototype> now treats magical variables like C<$1> the same way as
non-magical variables when checking for the CORE:: prefix, instead of
treating them as subroutine names.

=item *

Under threaded perls, a runtime code block in a regular expression could
corrupt the package name stored in the op tree, resulting in bad reads
in C<caller>, and possibly crashes [perl #113060].

=item *

Referencing a closure prototype (C<\&{$_[1]}> in an attribute handler for a
closure) no longer results in a copy of the subroutine (or assertion
failures on debugging builds).

=item *

C<eval '__PACKAGE__'> now returns the right answer on threaded builds if
the current package has been assigned over (as in
C<*ThisPackage:: = *ThatPackage::>) [perl #78742].

=item *

If a package is deleted by code that it calls, it is possible for C<caller>
to see a stack frame belonging to that deleted package.  C<caller> could
crash if the stash's memory address was reused for a scalar and a
substitution was performed on the same scalar [perl #113486].

=item *

C<UNIVERSAL::can> no longer treats its first argument differently
depending on whether it is a string or number internally.

=item *

C<open> with C<< <& >> for the mode checks to see whether the third argument is
a number, in determining whether to treat it as a file descriptor or a handle
name.  Magical variables like C<$1> were always failing the numeric check and
being treated as handle names.

=item *

C<warn>'s handling of magical variables (C<$1>, ties) has undergone several
fixes.  C<FETCH> is only called once now on a tied argument or a tied C<$@>
[perl #97480].  Tied variables returning objects that stringify as "" are
no longer ignored.  A tied C<$@> that happened to return a reference the
I<previous> time it was used is no longer ignored.

=item *

C<warn ""> now treats C<$@> with a number in it the same way, regardless of
whether it happened via C<$@=3> or C<$@="3">.  It used to ignore the
former.  Now it appends "\t...caught", as it has always done with
C<$@="3">.

=item *

Numeric operators on magical variables (e.g., S<C<$1 + 1>>) used to use
floating point operations even where integer operations were more appropriate,
resulting in loss of accuracy on 64-bit platforms [perl #109542].

=item *

Unary negation no longer treats a string as a number if the string happened
to be used as a number at some point.  So, if C<$x> contains the string "dogs",
C<-$x> returns "-dogs" even if C<$y=0+$x> has happened at some point.

=item *

In Perl 5.14, C<-'-10'> was fixed to return "10", not "+10".  But magical
variables (C<$1>, ties) were not fixed till now [perl #57706].

=item *

Unary negation now treats strings consistently, regardless of the internal
C<UTF8> flag.

=item *

A regression introduced in Perl v5.16.0 involving
C<tr/I<SEARCHLIST>/I<REPLACEMENTLIST>/> has been fixed.  Only the first
instance is supposed to be meaningful if a character appears more than
once in C<I<SEARCHLIST>>.  Under some circumstances, the final instance
was overriding all earlier ones.  [perl #113584]

=item *

Regular expressions like C<qr/\87/> previously silently inserted a NUL
character, thus matching as if it had been written C<qr/\00087/>.  Now it
matches as if it had been written as C<qr/87/>, with a message that the
sequence C<"\8"> is unrecognized.

=item *

C<__SUB__> now works in special blocks (C<BEGIN>, C<END>, etc.).

=item *

Thread creation on Windows could theoretically result in a crash if done
inside a C<BEGIN> block.  It still does not work properly, but it no longer
crashes [perl #111610].

=item *

C<\&{''}> (with the empty string) now autovivifies a stub like any other
sub name, and no longer produces the "Unable to create sub" error
[perl #94476].

=item *

A regression introduced in v5.14.0 has been fixed, in which some calls
to the C<re> module would clobber C<$_> [perl #113750].

=item *

C<do FILE> now always either sets or clears C<$@>, even when the file can't be
read. This ensures that testing C<$@> first (as recommended by the
documentation) always returns the correct result.

=item *

The array iterator used for the C<each @array> construct is now correctly
reset when C<@array> is cleared (RT #75596). This happens for example when the
array is globally assigned to, as in C<@array = (...)>, but not when its
B<values> are assigned to. In terms of the XS API, it means that C<av_clear()>
will now reset the iterator.

This mirrors the behaviour of the hash iterator when the hash is cleared.

=item *

C<< $class->can >>, C<< $class->isa >>, and C<< $class->DOES >> now return
correct results, regardless of whether that package referred to by C<$class>
exists [perl #47113].

=item *

Arriving signals no longer clear C<$@> [perl #45173].

=item *

Allow C<my ()> declarations with an empty variable list [perl #113554].

=item *

During parsing, subs declared after errors no longer leave stubs
[perl #113712].

=item *

Closures containing no string evals no longer hang on to their containing
subroutines, allowing variables closed over by outer subroutines to be
freed when the outer sub is freed, even if the inner sub still exists
[perl #89544].

=item *

Duplication of in-memory filehandles by opening with a "<&=" or ">&=" mode
stopped working properly in 5.16.0.  It was causing the new handle to
reference a different scalar variable.  This has been fixed [perl #113764].

=item *

C<qr//> expressions no longer crash with custom regular expression engines
that do not set C<offs> at regular expression compilation time
[perl #112962].

=item *

C<delete local> no longer crashes with certain magical arrays and hashes
[perl #112966].

=item *

C<local> on elements of certain magical arrays and hashes used not to
arrange to have the element deleted on scope exit, even if the element did
not exist before C<local>.

=item *

C<scalar(write)> no longer returns multiple items [perl #73690].

=item *

String to floating point conversions no longer misparse certain strings under
C<use locale> [perl #109318].

=item *

C<@INC> filters that die no longer leak memory [perl #92252].

=item *

The implementations of overloaded operations are now called in the correct
context. This allows, among other things, being able to properly override
C<< <> >> [perl #47119].

=item *

Specifying only the C<fallback> key when calling C<use overload> now behaves
properly [perl #113010].

=item *

C<< sub foo { my $a = 0; while ($a) { ... } } >> and
C<< sub foo { while (0) { ... } } >> now return the same thing [perl #73618].

=item *

String negation now behaves the same under C<use integer;> as it does
without [perl #113012].

=item *

C<chr> now returns the Unicode replacement character (U+FFFD) for -1,
regardless of the internal representation.  -1 used to wrap if the argument
was tied or a string internally.

=item *

Using a C<format> after its enclosing sub was freed could crash as of
perl 5.12.0, if the format referenced lexical variables from the outer sub.

=item *

Using a C<format> after its enclosing sub was undefined could crash as of
perl 5.10.0, if the format referenced lexical variables from the outer sub.

=item *

Using a C<format> defined inside a closure, which format references
lexical variables from outside, never really worked unless the C<write>
call was directly inside the closure.  In 5.10.0 it even started crashing.
Now the copy of that closure nearest the top of the call stack is used to
find those variables.

=item *

Formats that close over variables in special blocks no longer crash if a
stub exists with the same name as the special block before the special
block is compiled.

=item *

The parser no longer gets confused, treating C<eval foo ()> as a syntax
error if preceded by C<print;> [perl #16249].

=item *

The return value of C<syscall> is no longer truncated on 64-bit platforms
[perl #113980].

=item *

Constant folding no longer causes C<print 1 ? FOO : BAR> to print to the
FOO handle [perl #78064].

=item *

C<do subname> now calls the named subroutine and uses the file name it
returns, instead of opening a file named "subname".

=item *

Subroutines looked up by rv2cv check hooks (registered by XS modules) are
now taken into consideration when determining whether C<foo bar> should be
the sub call C<foo(bar)> or the method call C<< "bar"->foo >>.

=item *

C<CORE::foo::bar> is no longer treated specially, allowing global overrides
to be called directly via C<CORE::GLOBAL::uc(...)> [perl #113016].

=item *

Calling an undefined sub whose typeglob has been undefined now produces the
customary "Undefined subroutine called" error, instead of "Not a CODE
reference".

=item *

Two bugs involving @ISA have been fixed.  C<*ISA = *glob_without_array> and
C<undef *ISA; @{*ISA}> would prevent future modifications to @ISA from
updating the internal caches used to look up methods.  The
*glob_without_array case was a regression from Perl 5.12.

=item *

Regular expression optimisations sometimes caused C<$> with C</m> to
produce failed or incorrect matches [perl #114068].

=item *

C<__SUB__> now works in a C<sort> block when the enclosing subroutine is
predeclared with C<sub foo;> syntax [perl #113710].

=item *

Unicode properties only apply to Unicode code points, which leads to
some subtleties when regular expressions are matched against
above-Unicode code points.  There is a warning generated to draw your
attention to this.  However, this warning was being generated
inappropriately in some cases, such as when a program was being parsed.
Non-Unicode matches such as C<\w> and C<[:word;]> should not generate the
warning, as their definitions don't limit them to apply to only Unicode
code points.  Now the message is only generated when matching against
C<\p{}> and C<\P{}>.  There remains a bug, [perl #114148], for the very
few properties in Unicode that match just a single code point.  The
warning is not generated if they are matched against an above-Unicode
code point.

=item *

Uninitialized warnings mentioning hash elements would only mention the
element name if it was not in the first bucket of the hash, due to an
off-by-one error.

=item *

A regular expression optimizer bug could cause multiline "^" to behave
incorrectly in the presence of line breaks, such that
C<"/\n\n" =~ m#\A(?:^/$)#im> would not match [perl #115242].

=item *

Failed C<fork> in list context no longer corrupts the stack.
C<@a = (1, 2, fork, 3)> used to gobble up the 2 and assign C<(1, undef, 3)>
if the C<fork> call failed.

=item *

Numerous memory leaks have been fixed, mostly involving tied variables that
die, regular expression character classes and code blocks, and syntax
errors.

=item *

Assigning a regular expression (C<${qr//}>) to a variable that happens to
hold a floating point number no longer causes assertion failures on
debugging builds.

=item *

Assigning a regular expression to a scalar containing a number no longer
causes subsequent numification to produce random numbers.

=item *

Assigning a regular expression to a magic variable no longer wipes away the
magic.  This was a regression from 5.10.

=item *

Assigning a regular expression to a blessed scalar no longer results in
crashes.  This was also a regression from 5.10.

=item *

Regular expression can now be assigned to tied hash and array elements with
flattening into strings.

=item *

Numifying a regular expression no longer results in an uninitialized
warning.

=item *

Negative array indices no longer cause EXISTS methods of tied variables to
be ignored.  This was a regression from 5.12.

=item *

Negative array indices no longer result in crashes on arrays tied to
non-objects.

=item *

C<$byte_overload .= $utf8> no longer results in doubly-encoded UTF8 if the
left-hand scalar happened to have produced a UTF8 string the last time
overloading was invoked.

=item *

C<goto &sub> now uses the current value of @_, instead of using the array
the subroutine was originally called with.  This means
C<local @_ = (...); goto &sub> now works [perl #43077].

=item *

If a debugger is invoked recursively, it no longer stomps on its own
lexical variables.  Formerly under recursion all calls would share the same
set of lexical variables [perl #115742].

=item *

C<*_{ARRAY}> returned from a subroutine no longer spontaneously
becomes empty.

=back

=head1 Known Problems

=over 4

=item *

XXX: the imperfect behavior of the ** deprecation

=back

=head1 Acknowledgements

XXX Generate this with:

  perl Porting/acknowledgements.pl v5.18.0..HEAD

=head1 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
http://rt.perl.org/perlbug/ .  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 L<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 C<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.

=head1 SEE ALSO

The F<Changes> file for an explanation of how to view exhaustive details on
what changed.

The F<INSTALL> file for how to build Perl.

The F<README> file for general stuff.

The F<Artistic> and F<Copying> files for copyright information.

=cut