This document describes differences between the 5.12.0 release and
the 5.14.0 release.
+If you are upgrading from an earlier release such as 5.10.0, first read
+L<perl5120delta>, which describes differences between 5.10.0 and
+5.12.0.
+
Some of the bug fixes in this release have been backported to subsequent
releases of 5.12.x. Those are indicated with the 5.12.x version in
parentheses.
-XXX Go through the perl512*delta files and do that.
-
=head1 Notice
XXX Any important notices here
with one exception noted below.
See L<http://unicode.org/versions/Unicode6.0.0> for details on the new
release. Perl does not support any Unicode provisional properties,
-including the new ones for this release, but their database files are
-packaged with Perl.
+including the new ones for this release.
Unicode 6.0 has chosen to use the name C<BELL> for the character at U+1F514,
which is a symbol that looks like a bell, and is used in Japanese cell
See L<charnames> for details on all these changes.
+=head3 New warnings categories for problematic (non-)Unicode code points.
+
+Three new warnings subcategories of "utf8" have been added. These
+allow you to turn off some "utf8" warnings, while allowing
+others warnings to remain on. The three categories are:
+C<surrogate> when UTF-16 surrogates are encountered;
+C<nonchar> when Unicode non-character code points are encountered;
+and C<non_unicode> when code points that are above the legal Unicode
+maximum of 0x10FFFF are encountered.
+
=head3 Any unsigned value can be encoded as a character
With this release, Perl is adopting a model that any unsigned value can
be treated as a code point and encoded internally (as utf8) without
warnings - not just the code points that are legal in Unicode.
-However, unless utf8 warnings have been
+However, unless utf8 or the corresponding sub-category (see previous
+item) warnings have been
explicitly lexically turned off, outputting or performing a
Unicode-defined operation (such as upper-casing) on such a code point
will generate a warning. Attempting to input these using strict rules
them will work the same as for the non-legal Unicode code points, as the
Unicode standard says they are illegal for "open interchange".
-=head3 New warnings categories for problematic (non-)Unicode code points.
+=head3 Unicode database files not installed
-Three new warnings subcategories of "utf8" have been added. These
-allow you to turn off some "utf8" warnings, while allowing
-others warnings to remain on. The three categories are:
-C<surrogate> when UTF-16 surrogates are encountered;
-C<nonchar> when Unicode non-character code points are encountered;
-and C<non_unicode> when code points that are above the legal Unicode
-maximum of 0x10FFFF are encountered.
+The Unicode database files are no longer installed with Perl. This
+doesn't affect any functionality in Perl and saves significant disk
+space. If you previously were explicitly opening and reading those
+files, you can download them from
+L<http://www.unicode.org/Public/zipped/6.0.0/>.
=head2 Regular Expressions
of compiling the regular expression.
The C</a> regular expression modifier restricts C<\s>, C<\d> and C<\w> and
-the Posix (C<[[:posix:]]>) character classes to the ASCII range. The
+the Posix (C<[[:posix:]]>) character classes to the ASCII range. Their
complements and C<\b> and C<\B> are correspondingly
affected. Otherwise, C</a> behaves like the C</u> modifier, in that
case-insensitive matching uses Unicode semantics.
Regular expression debugging output (turned on by C<use re 'debug';>) now
uses hexadecimal when escaping non-ASCII characters, instead of octal.
+=head3 Return value of C<delete $+{...}>
+
+Custom regular expression engines can now determine the return value of
+C<delete> on an entry of C<%+> or C<%->.
+
=head2 Syntactical Enhancements
=head3 Array and hash container functions accept references
Perl more internally consistent. A round-trip with C<eval sprintf
"%#X", 0x10> now returns C<16>, the way C<eval sprintf "%#x", 0x10> does.
+=head3 Overridable tie functions
+
+C<tie>, C<tied> and C<untie> can now be overridden [perl #75902].
+
=head2 Exception Handling
Several changes have been made to the way C<die>, C<warn>, and C<$@>
=head2 New C APIs
-=head3 CLONE_PARAMS structure added to ease correct thread creation
-
-Modules that create threads should now create C<CLONE_PARAMS> structures
-by calling the new function C<Perl_clone_params_new()>, and free them with
-C<Perl_clone_params_del()>. This will ensure compatibility with any future
-changes to the internals of the C<CLONE_PARAMS> structure layout, and that
-it is correctly allocated and initialised.
-
-=head3 API function to parse statements
-
-The C<parse_fullstmt> function has been added to allow parsing of a single
-complete Perl statement. See L<perlapi> for details.
-
-=head3 API functions for accessing the runtime hinthash
-
-A new C API for introspecting the hinthash C<%^H> at runtime has been added.
-See C<cop_hints_2hv>, C<cop_hints_fetchpvn>, C<cop_hints_fetchpvs>,
-C<cop_hints_fetchsv>, and C<hv_copy_hints_hv> in L<perlapi> for details.
-
-=head3 C interface to C<caller()>
-
-The C<caller_cx> function has been added as an XSUB-writer's equivalent of
-C<caller()>. See L<perlapi> for details.
-
-=head3 Custom per-subroutine check hooks
-
-XS code in an extension module can now annotate a subroutine (whether
-implemented in XS or in Perl) so that nominated XS code will be called
-at compile time (specifically as part of op checking) to change the op
-tree of that subroutine. The compile-time check function (supplied by
-the extension module) can implement argument processing that can't be
-expressed as a prototype, generate customised compile-time warnings,
-perform constant folding for a pure function, inline a subroutine
-consisting of sufficiently simple ops, replace the whole call with a
-custom op, and so on. This was previously all possible by hooking the
-C<entersub> op checker, but the new mechanism makes it easy to tie the
-hook to a specific subroutine. See L<perlapi/cv_set_call_checker>.
-
-To help in writing custom check hooks, several subtasks within standard
-C<entersub> op checking have been separated out and exposed in the API.
-
-=head3 Improved support for custom OPs
-
-Custom ops can now be registered with the new C<custom_op_register> C
-function and the C<XOP> structure. This will make it easier to add new
-properties of custom ops in the future. Two new properties have been added
-already, C<xop_class> and C<xop_peep>.
-
-C<xop_class> is one of the OA_*OP constants, and allows L<B> and other
-introspection mechanisms to work with custom ops that aren't BASEOPs.
-C<xop_peep> is a pointer to a function that will be called for ops of this
-type from C<Perl_rpeep>.
-
-See L<perlguts/Custom Operators> and L<perlapi/Custom Operators> for more
-detail.
-
-The old C<PL_custom_op_names>/C<PL_custom_op_descs> interface is still
-supported but discouraged.
-
-=head3 Return value of C<delete $+{...}>
-
-Custom regular expression engines can now determine the return value of
-C<delete> on an entry of C<%+> or C<%->.
-
-XXX Mention the actual API.
+See L</Internal Changes>.
=head1 Security
Perl 5.14.0 is not binary-compatible with any previous stable release.
-=head2 Regular Expressions and String Escapes
-
-=head3 C<\cI<X>>
+In addition to the sections that follow, see L</C API Changes>.
-The backslash-c construct was designed as a way of specifying
-non-printable characters, but there were no restrictions (on ASCII
-platforms) on what the character following the C<c> could be. Now, that
-character must be one of the ASCII characters.
+=head2 Regular Expressions and String Escapes
=head3 \400-\777
m/\p{ASCII_Hex_Digit}+/i
could previously match non-ASCII characters because of the Unicode
-matching rules. There were a number of bugs in this feature until an
-earlier release in the 5.13 series. Now this release reverts, and
-removes the feature completely except for the few properties where
-people have come to expect it, namely the ones where casing is an
-integral part of their functionality, such as C<m/\p{Uppercase}/i> and
-C<m/\p{Lowercase}/i>, both of which match the exact same code points,
-namely those matched by C<m/\p{Cased}/i>. Details are in
-L<perlrecharclass/Unicode Properties>.
-
-XXX The mention of ‘until an earlier release in the 5.13 series’ needs to
-change, but I do not fully understand what happened here.
+matching rules (although there were a number of bugs with this). Now
+matching under C</i> gives the same results as non-C</i> matching except
+for those few properties where people have come to expect differences,
+namely the ones where casing is an integral part of their meaning, such
+as C<m/\p{Uppercase}/i> and C<m/\p{Lowercase}/i>, both of which match
+the exact same code points, namely those matched by C<m/\p{Cased}/i>.
+Details are in L<perlrecharclass/Unicode Properties>.
User-defined property handlers that need to match differently under
-C</i> must change to read the new boolean parameter passed to it which is
+C</i> must change to read the new boolean parameter passed to them which is
non-zero if case-insensitive matching is in effect or 0 otherwise. See
L<perluniprops/User-Defined Character Properties>.
Now, a Unicode property match specified in the pattern will indicate
that the pattern is meant for matching according to Unicode rules, the way
-C<\x{}> does.
+C<\N{}> does.
=head3 Regular expressions retain their localeness when interpolated
To fix this, the C<*{}> operator (including the C<*foo> and C<*$foo> forms)
has been modified to make a new immutable glob if its operand is a glob
-copy. Various operators that make a distinction between globs and scalars
-have been modified to treat only immutable globs as globs. (C<tie>,
-C<tied> and C<untie> has been left as they are for compatibility's sake,
+copy. This allows operators that make a distinction between globs and
+scalars to be modified to treat only immutable globs as globs. (C<tie>,
+C<tied> and C<untie> have been left as they are for compatibility's sake,
but will warn. See L</Deprecations>.)
This causes an incompatible change in code that assigns a glob to the
glob is made an alias to C<*bar>. Then it is discarded. So the second
assignment has no effect.
-The upside to this incompatible change is that bugs [perl #77496],
-[perl #77502], [perl #77508], [perl #77688], and [perl #77812],
-and maybe others, too, have been fixed.
-
See L<http://rt.perl.org/rt3/Public/Bug/Display.html?id=77810> for even
more detail.
This will change the behavior of Policy.sh if you happen to have been
accidentally relying on the Policy.sh incorrect behavior.
-=head2 C API changes
-
-=head3 Check API compatibility when loading XS modules
-
-When perl's API changes in incompatible ways (which usually happens between
-major releases), XS modules compiled for previous versions of perl will not
-work anymore. They will need to be recompiled against the new perl.
-
-In order to ensure that modules are recompiled, and to prevent users from
-accidentally loading modules compiled for old perls into newer ones, the
-C<XS_APIVERSION_BOOTCHECK> macro has been added. That macro, which is called
-when loading every newly compiled extension, compares the API version of the
-running perl with the version a module has been compiled for and raises an
-exception if they don't match.
-
-=head3 Perl_fetch_cop_label
-
-The first argument of the C API function C<Perl_fetch_cop_label> has changed
-from C<struct refcounted he *> to C<COP *>, to better insulate the user from
-implementation details.
-
-This API function was marked as "may change", and likely isn't in use outside
-the core. (Neither an unpacked CPAN, nor Google's codesearch, finds any other
-references to it.)
-
-=head3 GvCV() and GvGP() are no longer lvalues
+=head3 Perl source code is read in text mode on Windows
-The new GvCV_set() and GvGP_set() macros are now provided to replace
-assignment to those two macros.
+Perl scripts used to be read in binary mode on Windows for the benefit
+of the ByteLoader module (which is no longer part of core Perl). This
+had the side effect of breaking various operations on the DATA filehandle,
+including seek()/tell(), and even simply reading from DATA after file handles
+have been flushed by a call to system(), backticks, fork() etc.
-This allows a future commit to eliminate some backref magic between GV
-and CVs, which will require complete control over assignment to the
-gp_cv slot.
+The default build options for Windows have been changed to read Perl source
+code on Windows in text mode now. Hopefully ByteLoader will be updated on
+CPAN to automatically handle this situation [perl #28106].
=head1 Deprecations
+See also L</Deprecated C APIs>.
+
=head2 Omitting a space between a regular expression and subsequent word
Omitting a space between a regular expression operator or
example, C<< m/foo/sand $bar >> will still be parsed
as C<< m/foo/s and $bar >> but will issue a warning.
+=head2 C<\cI<X>>
+
+The backslash-c construct was designed as a way of specifying
+non-printable characters, but there were no restrictions (on ASCII
+platforms) on what the character following the C<c> could be. Now,
+a deprecation warning is raised if that character isn't an ASCII character.
+Also, a deprecation warning is raised for C<"\c{"> (which is the same
+as simply saying C<";">).
+
+=head2 C<"\b{"> and C<"\B{">
+
+In regular expressions, a literal C<"{"> immediately following a C<"\b">
+(not in a bracketed character class) or a C<"\B{"> is now deprecated
+to allow for its future use by Perl itself.
+
=head2 Deprecation warning added for deprecated-in-core .pl libs
This is a mandatory warning, not obeying -X or lexical warning bits.
This feature is being deprecated due to its many issues, as documented in
L<perlunicode/User-Defined Case Mappings (for serious hackers only)>.
-It is planned to remove this feature in Perl 5.16. A CPAN module
-providing improved functionality is being prepared for release by the
-time 5.14 is.
-
-XXX What module is that?
+It is planned to remove this feature in Perl 5.16. Instead use the CPAN module
+L<Unicode::Casing>, which provides improved functionality.
=head2 Deprecated modules
=back
-=head2 Deprecated C APIs
-
-=over
-
-=item C<Perl_ptr_table_clear>
-
-C<Perl_ptr_table_clear> is no longer part of Perl's public API. Calling it now
-generates a deprecation warning, and it will be removed in a future
-release.
-
-=item C<sv_compile_2op>
-
-The C<sv_compile_2op()> API function is now deprecated. Searches suggest
-that nothing on CPAN is using it, so this should have zero impact.
-
-It attempted to provide an API to compile code down to an optree, but failed
-to bind correctly to lexicals in the enclosing scope. It's not possible to
-fix this problem within the constraints of its parameters and return value.
-
-=back
-
=head1 Performance Enhancements
=head2 "Safe signals" optimisation
=item *
-C<HTTP::Tiny> 0.010 has been added as a dual-life module. It is a very
+C<HTTP::Tiny> 0.011 has been added as a dual-life module. It is a very
small, simple HTTP/1.1 client designed for simple GET requests and file
mirroring. It has has been added to enable CPAN.pm and CPANPLUS to
"bootstrap" HTTP access to CPAN using pure Perl without relying on external
=item *
-C<Module::Metadata> 1.000003 has been added as a dual-life module. It gathers
+C<Module::Metadata> 1.000004 has been added as a dual-life module. It gathers
package and POD information from Perl module files. It is a standalone module
based on Module::Build::ModuleInfo for use by other module installation
toolchain components. Module::Build::ModuleInfo has been deprecated in
=back
-=head2 Updated Modules and Pragmata
+=head2 Updated Modules and Pragma
=over 4
=item *
-C<ptar> has a new option to allow safe
+The C<ptar> utility has a new option to allow safe
creation of tarballs without world-writable files on Windows, allowing those
archives to be uploaded to CPAN.
The deparsing of a nextstate op has changed when it has both a
change of package (relative to the previous nextstate), or a change of
C<%^H> or other state, and a label. Previously the label was emitted
-first, but now the label is emitted last.
+first, but now the label is emitted last (5.12.1).
-The C<no 5.13.2> or similar form is now correctly handled by B::Deparse.
+The C<no 5.13.2> or similar form is now correctly handled by B::Deparse
+(5.12.3).
B::Deparse now properly handles the code that applies a conditional
pattern match against implicit C<$_> as it was fixed in [perl #20444].
=item *
-C<Carp> has been upgraded from version 1.18 to 1.19.
+C<Carp> has been upgraded from version 1.15 to 1.19.
L<Carp> now detects incomplete L<caller()|perlfunc/"caller EXPR"> overrides and
-avoids using bogus C<@DB::args>. To provide backtraces, Carp relies on
-particular behaviour of the caller built-in. Carp now detects if other code has
+avoids using bogus C<@DB::args>. To provide backtraces,
+Carp relies on particular behaviour of the C<caller>
+built-in. Carp now detects if other code has
overridden this with an incomplete implementation, and modifies its backtrace
-accordingly. Previously incomplete overrides would cause incorrect values in
-backtraces (best case), or obscure fatal errors (worst case)
+accordingly. Previously incomplete overrides would cause incorrect values
+in backtraces (best case), or obscure fatal errors (worst case).
This fixes certain cases of C<Bizarre copy of ARRAY> caused by modules
-overriding C<caller()> incorrectly.
+overriding C<caller()> incorrectly (5.12.2).
It now also avoids using regular expressions that cause perl to
load its Unicode tables, in order to avoid the 'BEGIN not safe after
=item *
-C<CPAN> has been upgraded from version 1.94_56 to 1.9600.-
+C<CPAN> has been upgraded from version 1.94_56 to 1.9600.
Major highlights:
=item *
-C<CPANPLUS> has been upgraded from version 0.90 to 0.9102.
+C<CPANPLUS> has been upgraded from version 0.90 to 0.9103.
+
+A change to F<cpanp-run-perl>
+resolves L<RT #55964|http://rt.cpan.org/Public/Bug/Display.html?id=55964>
+and L<RT #57106|http://rt.cpan.org/Public/Bug/Display.html?id=57106>, both
+of which related to failures to install distributions that use
+C<Module::Install::DSL> (5.12.2).
A dependency on Config was not recognised as a
core module dependency. This has been fixed.
=item *
+C<File::Spec> has been upgraded from version 3.31 to 3.33.
+
+Several portability fixes were made in C<File::Spec::VMS>: a colon is now
+recognized as a delimiter in native filespecs; caret-escaped delimiters are
+recognized for better handling of extended filespecs; C<catpath()> returns
+an empty directory rather than the current directory if the input directory
+name is empty; C<abs2rel()> properly handles Unix-style input (5.12.2).
+
+=item *
+
C<File::stat> has been upgraded from 1.02 to 1.04.
The C<-x> and C<-X> file test operators now work correctly under the root
=item *
-C<IPC::Cmd> has been upgraded from version 0.54 to 0.68.
+C<IPC::Cmd> has been upgraded from version 0.54 to 0.70.
Resolves an issue with splitting Win32 command lines. An argument
consisting of the single character "0" used to be omitted (CPAN RT #62961).
=item *
-C<Locale::Maketext> has been upgraded from version 1.14 to 1.17.
+C<IPC::SysV> has been upgraded from version 2.01 to 2.03.
+
+=item *
+
+C<lib> has been upgraded from version 0.62 to 0.63.
+
+=item *
+
+C<Locale::Maketext> has been upgraded from version 1.14 to 1.19.
Locale::Maketext now supports external caches.
=item *
+C<Log::Message> has been upgraded from version 0.02 to 0.04.
+
+=item *
+
+C<Log::Message::Simple> has been upgraded from version 0.06 to 0.08.
+
+=item *
+
C<Math::BigInt> has been upgraded from version 1.89_01 to 1.994.
This fixes, among other things, incorrect results when computing binomial
=item *
+C<Math::BigInt::FastCalc> has been upgraded from version 0.19 to 0.28.
+
+=item *
+
+C<Math::BigRat> has been upgraded from version 0.24 to 0.26_02.
+
+=item *
+
+C<Memoize> has been upgraded from version 1.01_03 to 1.02.
+
+=item *
+
C<MIME::Base64> has been upgraded from 3.08 to 3.13.
Includes new functions to calculate the length of encoded and decoded
=item *
-C<Module::CoreList> has been upgraded from version 2.29 to XXX.
+C<Module::CoreList> has been upgraded from version 2.29 to 2.47.
Besides listing the updated core modules of this release, it also stops listing
the C<Filespec> module. That module never existed in core. The scripts
=item *
-C<NDBM_File> and C<ODBM_File> have been upgraded from 1.08 to 1.11, and
-from 1.08 to 1.09, respectively.
+C<Module::Load> has been upgraded from version 0.16 to 0.18.
+
+=item *
+
+C<Module::Load::Conditional> has been upgraded from version 0.34 to 0.44.
+
+=item *
+
+C<NDBM_File> has been upgraded from version 1.08 to 1.11.
+
+This fixes a memory leak when DBM filters are used.
+
+=item *
+
+C<Net::Ping> has been upgraded from version 2.36 to 2.38.
+
+=item *
+
+C<NEXT> has been upgraded from version 0.64 to 0.65.
+
+=item *
+
+C<Object::Accessor> has been upgraded from version 0.36 to 0.38.
+
+=item *
+
+C<ODBM_File> have been upgraded from version 1.07 to 1.10.
This fixes a memory leak when DBM filters are used.
=item *
-C<overload> has been upgraded from 1.11 to 1.12.
+C<overload> has been upgraded from 1.10 to 1.12.
C<overload::Method> can now handle subroutines that are themselves blessed
into overloaded classes [perl #71998].
=item *
+C<Params::Check> has been upgraded from version 0.26 to 0.28.
+
+=item *
+
+C<parent> has been upgraded from version 0.223 to 0.225.
+
+=item *
+
C<Parse::CPAN::Meta> has been upgraded from version 1.40 to 1.4401.
The latest Parse::CPAN::Meta can now read YAML and JSON files using
=item *
-C<POSIX> has been upgraded from 1.19 to 1.23.
+C<Pod::LaTeX> has been upgraded from version 0.58 to 0.59.
+
+=item *
+
+C<Pod::Perldoc> has been upgraded from version 3.15_02 to 3.15_03.
+
+=item *
+
+C<Pod::Simple> has been upgraded from version 3.13 to 3.16.
+
+=item *
+
+C<POSIX> has been upgraded from 1.19 to 1.24.
It now includes constants for POSIX signal constants.
C<Safe> has been upgraded from version 2.25 to 2.29.
+Coderefs returned by C<reval()> and C<rdo()> are now wrapped via
+C<wrap_code_refs> (5.12.1).
+
This fixes a possible infinite loop when looking for coderefs.
-It adds C<&version::vxs::VCMP> to the default share.
+It adds several version::vxs::* routines to the default share.
=item *
=item *
+C<Term::ANSIColor> has been upgraded from version 2.02 to 3.00.
+
+=item *
+
+C<Term::UI> has been upgraded from version 0.20 to 0.26.
+
+=item *
+
+C<Test::Harness> has been upgraded from version 3.17 to 3.23.
+
+=item *
+
C<Test::Simple> has been upgraded from version 0.94 to 0.98.
Among many other things, subtests without a C<plan> or C<no_plan> now have an
=item *
+C<Thread::Queue> has been upgraded from version 2.11 to 2.12.
+
+=item *
+
+C<threads> has been upgraded from version 1.75 to 1.83.
+
+=item *
+
+C<threads::shared> has been upgraded from version 1.32 to 1.36.
+
+=item *
+
C<Tie::Hash> has been upgraded from version 1.03 to 1.04.
Calling C<< Tie::Hash-E<gt>TIEHASH() >> used to loop forever. Now it C<croak>s.
=item *
+C<Tie::RefHash> has been upgraded from version 1.38 to 1.39.
+
+=item *
+
+C<Time::HiRes> has been upgraded from version 1.9719 to 1.9721_01.
+
+=item *
+
+C<Time::Local> has been upgraded from version 1.1901_01 to 1.2000.
+
+=item *
+
+C<Time::Piece> has been upgraded from version 1.15_01 to 1.20_01.
+
+=item *
+
C<Unicode::Collate> has been upgraded from version 0.52_01 to 0.73.
Unicode::Collate has been updated to use Unicode 6.0.0.
=item *
+C<Unicode::Normalize> has been upgraded from version 1.03 to 1.10.
+
+=item *
+
C<Unicode::UCD> has been upgraded from version 0.27 to 0.32.
A new function, C<Unicode::UCD::num()>, has been added. This function
C<version> has been upgraded from 0.82 to 0.88.
Due to a bug, now fixed, the C<is_strict> and C<is_lax> functions did not
-work when exported.
+work when exported (5.12.1).
=item *
-C<warnings> and C<warnings::register> have been upgraded from version 1.09
-to 1.11 and from version 1.01 to 1.02 respectively.
+C<warnings> has been upgraded from version 1.09 to 1.12.
Calling C<use warnings> without arguments is now significantly more efficient.
+=item *
+
+C<warnings::register> have been upgraded from version 1.01 to 1.02.
+
It is now possible to register warning categories other than the names of
packages using C<warnings::register>. See L<perllexwarn> for more information.
=item *
+C<XSLoader> has been upgraded from version 0.10 to 0.13.
+
+=item *
+
C<VMS::DCLsym> has been upgraded from version 1.03 to 1.05.
Two bugs have been fixed [perl #84086]:
=over
-=item C<Class::ISA>
+=item *
+
+C<Class::ISA> has been removed from the Perl core. Prior version was 0.36.
+
+=item *
+
+C<Pod::Plainer> has been removed from the Perl core. Prior version was 1.02.
-=item C<Pod::Plainer>
+=item *
-=item C<Switch>
+C<Switch> has been removed from the Perl core. Prior version was 2.16.
=back
=head3 L<perlgpl>
L<perlgpl> has been updated to contain GPL version 1, as is included in the
-F<README> distributed with perl.
+F<README> distributed with perl (5.12.1).
=head3 Perl 5.12.x delta files
The perlmodlib page that came with Perl 5.12.0 was missing a lot of
modules, due to a bug in the script that generates the list. This has been
-fixed [perl #74332].
+fixed [perl #74332] (5.12.1).
=head3 Replace wrong tr/// table in L<perlebcdic>
=head3 INSTALL explicitly states the requirement for C89
-This was already true but it's now Officially Stated For The Record.
+This was already true but it's now Officially Stated For The Record
+(5.12.2).
=head3 Explanation of C<\xI<HH>> and C<\oI<OOO>> escapes
=head3 C<-0I<NNN>> switch
In L<perlrun>, the behavior of the C<-0NNN> switch for C<-0400> or higher
-has been clarified.
+has been clarified (5.12.2).
+
+=head3 Maintenance policy
+
+L<perlpolicy> now contains the policy on what patches are acceptable for
+maintenance branches (5.12.1).
=head3 Deprecation policy
L<perlpolicy> now contains the policy on compatibility and deprecation
-along with definitions of terms like "deprecation"
+along with definitions of terms like "deprecation" (5.12.2).
=head3 New descriptions in L<perldiag>
not resolve. So the user's address is now passed to sendmail so it's
less likely to get stuck in a mail queue somewhere [perl #82996].
+=item *
+
+L<perlbug> now always gives the reporter a chance to change the email
+address it guesses for them (5.12.2).
+
+=item *
+
+L<perlbug> should no longer warn about uninitialized values when using the C<-d>
+and C<-v> options (5.12.2).
+
=back
=head3 L<perl5db.pl>
build. The special-case handling for this situation got broken in Perl
5.11.0, and has now been repaired.
-=back
-
-=head1 Testing
-
-XXX This section has not been cleaned up yet. Do we really need this
-section? How many people are going to find a list of new test files useful?
-Anyone hacking on the core can use
-‘git diff v5.12.0..v5.14.0 MANIFEST|grep '^[+-]t/'’.
-
-XXX Any significant changes to the testing of a freshly built perl should be
-listed here. Changes which create B<new> files in F<t/> go here as do any
-large changes to the testing harness (e.g. when parallel testing was added).
-Changes to existing files in F<t/> aren't worth summarising, although the bugs
-that they represent may be covered elsewhere.
+=item *
-=over 4
-
-=item *
-
-F<t/harness> clears PERL5LIB, PERLLIB, PERL5OPT as t/TEST does (a2d3de1)
-
-=item *
-
-Many common testing routines were refactored into t/lib/common.pl
-
-=item *
-
-Several test files have been modernized to use Test::More
-
-=item *
-
-F<t/op/print.t> has been added to test implicit printing of C<$_>.
-
-=item *
-
-F<t/io/errnosig.t> has been added to test for restoration of of C<$!> when
-leaving signal handlers.
-
-=item *
-
-F<t/op/tie_fetch_count.t> has been added to see if C<FETCH> is only called once
-on tied variables.
-
-=item *
-
-F<lib/Tie/ExtraHash.t> has been added to make sure the, previously untested,
-L<Tie::ExtraHash> keeps working.
-
-=item *
-
-F<t/re/overload.t> has been added to test against string corruption in pattern
-matches on overloaded objects. This is a TODO test.
-
-=item *
-
-The new F<t/lib/universal.t> script tests the Internal::* functions and other
-things in F<universal.c>.
-
-=item *
-
-A rare race condition in F<t/op/while_readdir.t> has been fixed, stopping it
-from failing randomly when running tests in parallel.
-
-=item *
-
-The new F<t/op/leaky-magic.t> script tests that magic applied to variables in
-the main packages does not affect other packages.
-
-=item *
-
-The script F<t/op/threads-dirh.t> has been added, which tests interaction
-of threads and directory handles.
-
-=item *
-
-The new F<t/mro/isa_aliases.t> has been added, which tests that
-C<*Foo::ISA = *Bar::ISA> works properly.
-
-=item *
-
-F<t/mro/isarev.t> has been added, which tests that C<PL_isarev> (accessible
-at the Perl level via C<mro::get_isarev>) is updated properly.
-
-=item *
-
-F<t/run/switchd-78586.t> has been added, which tests that [perl #78586]
-has been fixed (related to line numbers in the debugger).
-
-=item *
-
-C<lib/File/DosGlob.t> has been modernized and now uses C<Test::More>.
-
-=item *
-
-A new test script, C<t/porting/filenames.t>, makes sure that filenames and
-paths are reasonably portable.
-
-=item *
-
-C<t/porting/diag.t> is now several orders of magnitude faster.
-
-=item *
-
-C<t/porting/buildtoc.t> now tests that the documentation TOC file is current and well-formed.
-
-=item *
-
-C<t/base/while.t> now tests the basics of a while loop with minimal dependencies.
-
-=item *
-
-C<t/cmd/while.t> now uses F<test.pl> for better maintainability.
-
-=item *
-
-C<t/op/split.t> now tests calls to C<split> without any pattern specified.
-
-=item *
-
-F<porting/FindExt.t> now skips all tests on a static (-Uusedl) build
-of perl.
-
-=item *
-
-F<porting/FindExt.t> now passes on non-Win32 platforms when some
-extensions are built statically.
-
-=item *
-
-The tests for C<split /\s/> and Unicode have been moved from
-F<t/op/split.t> to the new F<t/op/split_unicode.t>.
+The previous default size of a PerlIO buffer (4096 bytes) has been increased
+to the larger of 8192 bytes and your local BUFSIZ. Benchmarks show that doubling
+this decade-old default increases read and write performance in the neighborhood
+of 25% to 50% when using the default layers of perlio on top of unix. To choose
+a non-default size, such as to get back the old value or to obtain an even
+larger value, configure with:
-=item *
+ ./Configure -Accflags=-DPERLIOBUF_DEFAULT_BUFSIZ=N
-F<t/re/re.t> has been moved to F<ext/re/t/re_funcs_u.t>.
+where N is the desired size in bytes; it should probably be a multiple of
+your page size.
=item *
-The tests for [perl #72922] have been moved from F<t/re/qr.t> to the new
-F<t/re/qr-72922.t>.
+An "incompatible operand types" error in ternary expressions when building
+with C<clang> has been fixed (5.12.2).
=item *
-F<t/re/reg_unsafe.t> has been deleted and its only test moved to
-F<t/re/pat_advanced.t>.
+Perl now skips setuid C<File::Copy> tests on partitions it detects to be mounted
+as C<nosuid> (5.12.2).
=back
=item AIX
-Perl now builds on AIX 4.2.
+Perl now builds on AIX 4.2 (5.12.1).
=back
=head2 Platform-Specific Notes
+=head3 AIX
+
+=over
+
+=item *
+
+F<README.aix> has been updated with information about the XL C/C++ V11 compiler
+suite (5.12.2).
+
+=back
+
+=head3 ARM
+
+=over
+
+=item *
+
+The C<d_u32align> configuration probe on ARM has been fixed (5.12.2).
+
+=back
+
=head3 Cygwin
=over 4
=back
+=head3 FreeBSD 7
+
+=over
+
+=item *
+
+FreeBSD 7 no longer contains F</usr/bin/objformat>. At build time,
+Perl now skips the F<objformat> check for versions 7 and higher and
+assumes ELF (5.12.1).
+
+=back
+
+=head3 HP-UX
+
+=over
+
+=item *
+
+Perl now allows -Duse64bitint without promoting to use64bitall on HP-UX
+(5.12.1).
+
+=back
+
=head3 IRIX
Conversion of strings to floating-point numbers is now more accurate on
=head3 OpenVOS
perl now builds again with OpenVOS (formerly known as Stratus VOS)
-[perl #78132].
+[perl #78132] (5.12.3).
=head3 Solaris
DTrace is now supported on Solaris. There used to be build failures, but
-these have been fixed [perl #73630].
+these have been fixed [perl #73630] (5.12.3).
=head3 VMS
=item *
+It's now possible to build extensions on older (pre 7.3-2) VMS systems.
+
+DCL symbol length was limited to 1K up until about seven years or
+so ago, but there was no particularly deep reason to prevent those
+older systems from configuring and building Perl (5.12.1).
+
+=item *
+
+We fixed the previously-broken C<-Uuseperlio> build on VMS.
+
+We were checking a variable that doesn't exist in the non-default
+case of disabling perlio. Now we only look at it when it exists (5.12.1).
+
+=item *
+
+We fixed the -Uuseperlio command-line option in configure.com.
+
+Formerly it only worked if you went through all the questions
+interactively and explicitly answered no (5.12.1).
+
+=item *
+
C<PerlIOUnix_open> now honours the default permissions on VMS.
When C<perlio> became the default and C<unixio> became the default bottom layer,
To avoid this, C<0777> is now passed as the permissions to C<open()>. In the
VMS CRTL, C<0777> has a special meaning over and above intersecting with the
current umask; specifically, it allows Unix syscalls to preserve native default
-permissions.
+permissions (5.12.3).
+
+=item *
+
+Spurious record boundaries are no longer
+introduced by the PerlIO layer during output (5.12.3).
=item *
opened for write by the perlio layer will now be line-buffered to prevent the
introduction of spurious line breaks whenever the perlio buffer fills up.
+=item *
+
+F<git_version.h> is now installed on VMS. This was an oversight in v5.12.0 which
+caused some extensions to fail to build (5.12.2).
+
+=item *
+
+Several memory leaks in L<stat()|perlfunc/"stat FILEHANDLE"> have been fixed (5.12.2).
+
+=item *
+
+A memory leak in C<Perl_rename()> due to a double allocation has been
+fixed (5.12.2).
+
+=item *
+
+A memory leak in C<vms_fid_to_name()> (used by C<realpath()> and
+C<realname()>) has been fixed (5.12.2).
+
=back
=head3 Windows
+See also L</"fork() emulation will not wait for signalled children"> and
+L</"Perl source code is read in text mode on Windows">, above.
+
=over 4
=item *
=item *
+Compilation with Visual Studio 2010 is now supported.
+
+=item *
+
When using old 32-bit compilers, the define C<_USE_32BIT_TIME_T> will now
be set in C<$Config{ccflags}>. This improves portability when compiling
XS extensions using new compilers, but for a perl compiled with old 32-bit
=item *
+When building Perl with the mingw64 x64 cross-compiler C<incpath>,
+C<libpth>, C<ldflags>, C<lddlflags> and C<ldflags_nolargefiles> values
+in F<Config.pm> and F<Config_heavy.pl> were not previously being set
+correctly because, with that compiler, the include and lib directories
+are not immediately below C<$(CCHOME)> (5.12.2).
+
+=item *
+
The build process proceeds more smoothly with mingw and dmake when
F<C:\MSYS\bin> is in the PATH, due to a C<Cwd> fix.
=head1 Internal Changes
-See also L</New C APIs>, L</C API Changes> and L</Deprecated C APIs>, above.
+=head2 New APIs
-=over 4
+=head3 CLONE_PARAMS structure added to ease correct thread creation
-=item *
+Modules that create threads should now create C<CLONE_PARAMS> structures
+by calling the new function C<Perl_clone_params_new()>, and free them with
+C<Perl_clone_params_del()>. This will ensure compatibility with any future
+changes to the internals of the C<CLONE_PARAMS> structure layout, and that
+it is correctly allocated and initialised.
-The protocol for unwinding the C stack at the last stage of a C<die>
-has changed how it identifies the target stack frame. This now uses
-a separate variable C<PL_restartjmpenv>, where previously it relied on
-the C<blk_eval.cur_top_env> pointer in the C<eval> context frame that
-has nominally just been discarded. This change means that code running
-during various stages of Perl-level unwinding no longer needs to take
-care to avoid destroying the ghost frame.
+=head3 New parsing functions
-=item *
+Several functions have been added for parsing statements or multiple
+statements:
-The format of entries on the scope stack has been changed, resulting in a
-reduction of memory usage of about 10%. In particular, the memory used by
-the scope stack to record each active lexical variable has been halved.
+=over
=item *
-Memory allocation for pointer tables has been changed. Previously
-C<Perl_ptr_table_store> allocated memory from the same arena system as C<SV>
-bodies and C<HE>s, with freed memory remaining bound to those arenas until
-interpreter exit. Now it allocates memory from arenas private to the specific
-pointer table, and that memory is returned to the system when
-C<Perl_ptr_table_free> is called. Additionally, allocation and release are both
-less CPU intensive.
+C<parse_fullstmt> parses a complete Perl statement.
=item *
-A new function, Perl_magic_methcall has been added that wraps the setup needed
-to call a magic method like FETCH (the existing S_magic_methcall function has
-been renamed S_magic_methcall1).
+C<parse_stmtseq> parses a sequence of statements, up
+to closing brace or EOF.
=item *
-The implementation of sv_dup_inc() has changed from a macro to a function.
+C<parse_block> parses a block [perl #78222].
=item *
-The C<find_rundefsvoffset> function has been deprecated. It appeared that
-its design was insufficient to reliably get the lexical C<$_> at run-time.
-
-Use the new C<find_rundefsv> function or the C<UNDERBAR> macro instead.
-They directly return the right SV representing C<$_>, whether it's lexical
-or dynamic (789bd8 .. 03d5bc).
+C<parse_barestmt> parses a statement
+without a label.
=item *
-The following new functions or macros have been added to the public API:
-C<SvNV_nomg>, C<sv_2nv_flags>, C<find_rundefsv>.
-
-=item *
+C<parse_label> parses a statement label, separate from statements.
-The C<UNDERBAR> macro now calls C<find_rundefsv>. C<dUNDERBAR> is now a
-noop but should still be used to ensure past and future compatibility.
+=back
-=item *
+The
+L<C<parse_fullexpr()>|perlapi/parse_fullexpr>,
+L<C<parse_listexpr()>|perlapi/parse_listexpr>,
+L<C<parse_termexpr()>|perlapi/parse_termexpr>, and
+L<C<parse_arithexpr()>|perlapi/parse_arithexpr>
+functions have been added to the API. They perform
+recursive-descent parsing of expressions at various precedence levels.
+They are expected to be used by syntax plugins.
-The ibcmp_* functions have been renamed and are now called foldEQ,
-foldEQ_locale and foldEQ_utf8 (e6226b).
+See L<perlapi> for details.
-=item *
+=head3 Hints hash API
-Under some circumstances, the C<CvGV()> field of a CV is now reference
-counted. To ensure consistent behaviour, direct assignment to it, for
-example C<CvGV(cv) = gv> is now a compile-time error. A new macro,
-C<CvGV_set(cv,gv)> has been introduced to perform this operation safely.
-Note that modification of this field is not part of of the public API,
-regardless of this new macro. This change caused some
-L<issues|/"Known Problems"> in modules that used the private C<GvGV()>
-field.
+A new C API for introspecting the hinthash C<%^H> at runtime has been
+added. See C<cop_hints_2hv>, C<cop_hints_fetchpvn>, C<cop_hints_fetchpvs>,
+C<cop_hints_fetchsv>, and C<hv_copy_hints_hv> in L<perlapi> for details.
-=item *
+A new, experimental API has been added for accessing the internal
+structure that Perl uses for C<%^H>. See the functions beginning with
+C<cophh_> in L<perlapi>.
-It is now possible for XS code to hook into Perl's lexical scope
-mechanism at compile time, using the new C<Perl_blockhook_register>
-function. See L<perlguts/"Compile-time scope hooks">.
+=head3 C interface to C<caller()>
-=item *
+The C<caller_cx> function has been added as an XSUB-writer's equivalent of
+C<caller()>. See L<perlapi> for details.
-Added C<Perl_croak_no_modify()> to implement
-C<Perl_croak("%s", PL_no_modify)> (6ad8f25)
+=head3 Custom per-subroutine check hooks
-=item *
+XS code in an extension module can now annotate a subroutine (whether
+implemented in XS or in Perl) so that nominated XS code will be called
+at compile time (specifically as part of op checking) to change the op
+tree of that subroutine. The compile-time check function (supplied by
+the extension module) can implement argument processing that can't be
+expressed as a prototype, generate customised compile-time warnings,
+perform constant folding for a pure function, inline a subroutine
+consisting of sufficiently simple ops, replace the whole call with a
+custom op, and so on. This was previously all possible by hooking the
+C<entersub> op checker, but the new mechanism makes it easy to tie the
+hook to a specific subroutine. See L<perlapi/cv_set_call_checker>.
-Added prototypes for C<tie()> and C<untie()> to allow overloading (RT#75902)
-(1db4d19)
+To help in writing custom check hooks, several subtasks within standard
+C<entersub> op checking have been separated out and exposed in the API.
-=item *
+=head3 Improved support for custom OPs
-Adds C<my_[l]stat_flags()> to replace C<my_[l]stat()>. C<my_stat()> and
-C<my_lstat()> call get magic on the stack arg, so create C<_flags()>
-variants that allow us to control this. (0d7d409)
+Custom ops can now be registered with the new C<custom_op_register> C
+function and the C<XOP> structure. This will make it easier to add new
+properties of custom ops in the future. Two new properties have been added
+already, C<xop_class> and C<xop_peep>.
-=item *
+C<xop_class> is one of the OA_*OP constants, and allows L<B> and other
+introspection mechanisms to work with custom ops
+that aren't BASEOPs. C<xop_peep> is a pointer to
+a function that will be called for ops of this
+type from C<Perl_rpeep>.
-Removed C<PERL_POLLUTE>
+See L<perlguts/Custom Operators> and L<perlapi/Custom Operators> for more
+detail.
-The option to define C<PERL_POLLUTE> to expose older 5.005 symbols for backwards
-compatibility has been removed. It's use was always discouraged, and MakeMaker
-contains a more specific escape hatch:
+The old C<PL_custom_op_names>/C<PL_custom_op_descs> interface is still
+supported but discouraged.
- perl Makefile.PL POLLUTE=1
+=head3 Scope hooks
-This can be used for modules that have not been upgraded to 5.6 naming
-conventions (and really should be completely obsolete by now).
+It is now possible for XS code to hook into Perl's lexical scope
+mechanism at compile time, using the new C<Perl_blockhook_register>
+function. See L<perlguts/"Compile-time scope hooks">.
-=item *
+=head3 The recursive part of the peephole optimizer is now hookable
-Added C<PERL_STATIC_INLINE>
+In addition to C<PL_peepp>, for hooking into the toplevel peephole optimizer, a
+C<PL_rpeepp> is now available to hook into the optimizer recursing into
+side-chains of the optree.
-The C<PERL_STATIC_INLINE> define has been added to provide the best-guess
-incantation to use for static inline functions, if the C compiler supports
-C99-style static inline. If it doesn't, it'll give a plain C<static>.
+=head3 New non-magical variants of existing functions
-C<HAS_STATIC_INLINE> can be used to check if the compiler actually supports
-inline functions.
+The following functions/macros have been added to the API. The C<*_nomg>
+macros are equivalent to their non-_nomg variants, except that they ignore
+get-magic. Those ending in C<_flags> allow one to specify whether
+get-magic is processed.
-=item *
+ sv_2bool_flags
+ SvTRUE_nomg
+ sv_2nv_flags
+ SvNV_nomg
+ sv_cmp_flags
+ sv_cmp_locale_flags
+ sv_eq_flags
+ sv_collxfrm_flags
-C<CALL_FPTR> and C<CPERLscope> have been deprecated.
+In some of these cases, the non-_flags functions have
+been replaced with wrappers around the new functions.
-Those are left from an old implementation of C<MULTIPLICITY> using C++ objects,
-which was removed in Perl 5.8. Nowadays these macros do exactly nothing, so
-they shouldn't be used anymore.
+=head3 pv/pvs/sv versions of existing functions
-For compatibility, they are still defined for external C<XS> code. Only
-extensions defining C<PERL_CORE> must be updated now.
+Many functions ending with pvn now have equivalent pv/pvs/sv versions.
-=item *
+=head3 List op-building functions
-C<lex_stuff_pvs()> has been added as a convenience macro wrapping
-C<lex_stuff_pvn()> for literal strings.
+List op-building functions have been added to the
+API. See L<op_append_elem|perlapi/op_append_elem>,
+L<op_append_list|perlapi/op_append_list>, and
+L<op_prepend_elem|perlapi/op_prepend_elem> in L<perlapi>.
-=item *
+=head3 C<LINKLIST>
-The recursive part of the peephole optimizer is now hookable.
+The L<LINKLIST|perlapi/LINKLIST> macro, part of op building that
+constructs the execution-order op chain, has been added to the API.
-In addition to C<PL_peepp>, for hooking into the toplevel peephole optimizer, a
-C<PL_rpeepp> is now available to hook into the optimizer recursing into
-side-chains of the optree.
+=head3 Localisation functions
-=item *
+The C<save_freeop>, C<save_op>, C<save_pushi32ptr> and C<save_pushptrptr>
+functions have been added to the API.
-See L</Regular expressions retain their localeness when interpolated>,
-above.
+=head3 Stash names
-=item *
+A stash can now have a list of effective names in addition to its usual
+name. The first effective name can be accessed via the C<HvENAME> macro,
+which is now the recommended name to use in MRO linearisations (C<HvNAME>
+being a fallback if there is no C<HvENAME>).
-The C<sv_cmp_flags>, C<sv_cmp_locale_flags>, C<sv_eq_flags> and
-C<sv_collxfrm_flags> functions have been added. These are like their
-non-_flags counterparts, but allow one to specify whether get-magic is
-processed.
+These names are added and deleted via C<hv_ename_add> and
+C<hv_ename_delete>. These two functions are I<not> part of the API.
-The C<sv_cmp>, C<sv_cmp_locale>, C<sv_eq> and C<sv_collxfrm> functions have
-been replaced with wrappers around the new functions.
+=head3 New functions for finding and removing magic
-=item *
+The L<C<mg_findext()>|perlapi/mg_findext> and
+L<C<sv_unmagicext()>|perlapi/sv_unmagicext>
+functions have been added to the API.
+They allow extension authors to find and remove magic attached to
+scalars based on both the magic type and the magic virtual table, similar to how
+C<sv_magicext()> attaches magic of a certain type and with a given virtual table
+to a scalar. This eliminates the need for extensions to walk the list of
+C<MAGIC> pointers of an C<SV> to find the magic that belongs to them.
-A new C<sv_2bool_flags> function has been added.
+=head3 C<find_rundefsv>
-This is like C<sv_2bool>, but it lets the calling code decide whether
-get-magic is handled. C<sv_2bool> is now a macro that calls the new
-function.
+This function returns the SV representing C<$_>, whether it's lexical
+or dynamic.
-=item *
+=head3 C<Perl_croak_no_modify>
-A new macro, C<SvTRUE_nomg>, has been added.
+C<Perl_croak_no_modify()> is short-hand for
+C<Perl_croak("%s", PL_no_modify)>.
-This is like C<SvTRUE>, except that it does not process magic. It uses the
-new C<sv_2bool_flags> function.
+=head3 C<PERL_STATIC_INLINE> define
-=item *
+The C<PERL_STATIC_INLINE> define has been added to provide the best-guess
+incantation to use for static inline functions, if the C compiler supports
+C99-style static inline. If it doesn't, it'll give a plain C<static>.
-C<sv_catsv_flags> no longer calls C<mg_get> on its second argument (the
-source string) if the flags passed to it do not include SV_GMAGIC. So it
-now matches the documentation.
+C<HAS_STATIC_INLINE> can be used to check if the compiler actually supports
+inline functions.
-=item *
+=head3 New C<pv_escape> option for hexadecimal escapes
-A new interface has been added for custom check hooks on subroutines. See
-L</Custom per-subroutine check hooks>, above.
+A new option, C<PERL_PV_ESCAPE_NONASCII>, has been added to C<pv_escape> to
+dump all characters above ASCII in hexadecimal. Before, one could get all
+characters as hexadecimal or the Latin1 non-ASCII as octal.
-=item *
+=head3 C<lex_start>
-List op building functions have been added to the
-API. See L<op_append_elem|perlapi/op_append_elem>,
-L<op_append_list|perlapi/op_append_list>, and
-L<op_prepend_elem|perlapi/op_prepend_elem>.
+C<lex_start> has been added to the API, but is considered experimental.
-=item *
+=head3 C<op_scope()> and C<op_lvalue()>
-The L<LINKLIST|perlapi/LINKLIST> macro, part of op building that
-constructs the execution-order op chain, has been added to the API.
+The C<op_scope()> and C<op_lvalue()> functions have been added to the API,
+but are considered experimental.
-=item *
+=head2 C API Changes
-Many functions ending with pvn now have equivalent pv/pvs/sv versions.
+=head3 C<PERL_POLLUTE> has been removed
-=item *
+The option to define C<PERL_POLLUTE> to expose older 5.005 symbols for
+backwards compatibility has been removed. It's use was always discouraged,
+and MakeMaker contains a more specific escape hatch:
-The C<save_freeop>, C<save_op>, C<save_pushi32ptr> and C<save_pushptrptr>
-functions have been added to the API.
+ perl Makefile.PL POLLUTE=1
-=item *
+This can be used for modules that have not been upgraded to 5.6 naming
+conventions (and really should be completely obsolete by now).
-The new API function C<parse_stmtseq()> parses a sequence of statements, up
-to closing brace or EOF.
+=head3 Check API compatibility when loading XS modules
-=item *
+When perl's API changes in incompatible ways (which usually happens between
+major releases), XS modules compiled for previous versions of perl will not
+work anymore. They will need to be recompiled against the new perl.
-C<lex_start> has been added to the API, but is considered experimental.
+In order to ensure that modules are recompiled, and to prevent users from
+accidentally loading modules compiled for old perls into newer ones, the
+C<XS_APIVERSION_BOOTCHECK> macro has been added. That macro, which is
+called when loading every newly compiled extension, compares the API
+version of the running perl with the version a module has been compiled for
+and raises an exception if they don't match.
-=item *
+=head3 Perl_fetch_cop_label
-A new C<parse_block> function has been added to the API [perl #78222].
+The first argument of the C API function C<Perl_fetch_cop_label> has changed
+from C<struct refcounted he *> to C<COP *>, to insulate the user from
+implementation details.
-=item *
+This API function was marked as "may change", and likely isn't in use outside
+the core. (Neither an unpacked CPAN, nor Google's codesearch, finds any other
+references to it.)
-A new, experimental API has been added for accessing the internal
-structure that Perl uses for C<%^H>. See the functions beginning with
-C<cophh_> in L<perlapi>.
+=head3 GvCV() and GvGP() are no longer lvalues
-=item *
+The new GvCV_set() and GvGP_set() macros are now provided to replace
+assignment to those two macros.
-A stash can now have a list of effective names in addition to its usual
-name. The first effective name can be accessed via the C<HvENAME> macro,
-which is now the recommended name to use in MRO linearisations (C<HvNAME>
-being a fallback if there is no C<HvENAME>).
+This allows a future commit to eliminate some backref magic between GV
+and CVs, which will require complete control over assignment to the
+gp_cv slot.
-These names are added and deleted via C<hv_ename_add> and
-C<hv_ename_delete>. These two functions are I<not> part of the API.
+=head3 CvGV() is no longer an lvalue
-=item *
+Under some circumstances, the C<CvGV()> field of a CV is now
+reference-counted. To ensure consistent behaviour, direct assignment to
+it, for example C<CvGV(cv) = gv> is now a compile-time error. A new macro,
+C<CvGV_set(cv,gv)> has been introduced to perform this operation
+safely. Note that modification of this field is not part of the public
+API, regardless of this new macro (and despite its being listed in this section).
+
+=head3 CvSTASH() is no longer an lvalue
+
+The C<CvSTASH()> macro can now only be used as an rvalue. C<CvSTASH_set()>
+has been added to replace assignment to C<CvSTASH()>. This is to ensure
+that backreferences are handled properly. These macros are not part of the
+API.
+
+=head3 Calling conventions for C<newFOROP> and C<newWHILEOP>
The way the parser handles labels has been cleaned up and refactored. As a
result, the C<newFOROP()> constructor function no longer takes a parameter
stating what label is to go in the state op.
-=item *
-
The C<newWHILEOP()> and C<newFOROP()> functions no longer accept a line
number as a parameter.
-=item *
-
-A new C<parse_barestmt()> function has been added, for parsing a statement
-without a label.
-
-=item *
-
-A new C<parse_label()> function has been added, that parses a statement
-label, separate from statements.
-
-=item *
-
-The C<CvSTASH()> macro can now only be used as an rvalue. C<CvSTASH_set()>
-has been added to replace assignment to C<CvSTASH()>. This is to ensure
-that backreferences are handled properly. These macros are not part of the
-API.
-
-=item *
-
-The C<op_scope()> and C<op_lvalue()> functions have been added to the API,
-but are considered experimental.
-
-=item *
-
-The L<C<mg_findext()>|perlapi/mg_findext> and
-L<C<sv_unmagicext()>|perlapi/sv_unmagicext>
-functions have been added to the API.
-They allow extension authors to find and remove magic attached to
-scalars based on both the magic type and the magic virtual table, similar to how
-C<sv_magicext()> attaches magic of a certain type and with a given virtual table
-to a scalar. This eliminates the need for extensions to walk the list of
-C<MAGIC> pointers of an C<SV> to find the magic that belongs to them.
-
-=item *
-
-The
-L<C<parse_fullexpr()>|perlapi/parse_fullexpr>,
-L<C<parse_listexpr()>|perlapi/parse_listexpr>,
-L<C<parse_termexpr()>|perlapi/parse_termexpr>, and
-L<C<parse_arithexpr()>|perlapi/parse_arithexpr>
-functions have been added to the API. They perform
-recursive-descent parsing of expressions at various precedence levels.
-They are expected to be used by syntax plugins.
-
-=item *
-
-The opcode bodies for C<chop> and C<chomp> and for C<schop> and C<schomp> have
-been merged. The implementation functions C<Perl_do_chop()> and
-C<Perl_do_chomp()>, never part of the public API, have been merged and moved to
-a static function in F<pp.c>. This shrinks the perl binary slightly, and should
-not affect any code outside the core (unless it is relying on the order of side
-effects when C<chomp> is passed a I<list> of values).
-
-=item *
+=head3 Flags passed to C<uvuni_to_utf8_flags> and C<utf8n_to_uvuni>
-Some of the flags parameters to the uvuni_to_utf8_flags() and
-utf8n_to_uvuni() have changed. This is a result of Perl now allowing
+Some of the flags parameters to uvuni_to_utf8_flags() and
+utf8n_to_uvuni() have changed. This is a result of Perl's now allowing
internal storage and manipulation of code points that are problematic
in some situations. Hence, the default actions for these functions has
been complemented to allow these code points. The new flags are
documented in L<perlapi>. Code that requires the problematic code
-points to be rejected needs to change to use these flags. Some flag
+points to be rejected needs to change to use the new flags. Some flag
names are retained for backward source compatibility, though they do
nothing, as they are now the default. However the flags
C<UNICODE_ALLOW_FDD0>, C<UNICODE_ALLOW_FFFF>, C<UNICODE_ILLEGAL>, and
should be handled, which is now described in
L<perlunicode/Non-character code points>. See also L</Selected Bug Fixes>.
-=item *
+XXX Which bugs in particular? Selected Bug Fixes is too long for this link
+to be meaningful right now
+I don't see the bugs in that section currently -- khw
-Certain shared flags in the C<pmop.op_pmflags> and C<regexp.extflags>
-structures have been removed. These are: C<Rxf_Pmf_LOCALE>,
-C<Rxf_Pmf_UNICODE>, and C<PMf_LOCALE>. Instead there are encodes and
-three static in-line functions for accessing the information:
-C<get_regex_charset()>, C<set_regex_charset()>, and C<get_regex_charset_name()>,
-which are defined in the places where the original flags were.
+=head2 Deprecated C APIs
-=item *
+=over
-A new option has been added to C<pv_escape> to dump all characters above
-ASCII in hexadecimal. Before, one could get all characters as hexadecimal
-or the Latin1 non-ASCII as octal
+=item C<Perl_ptr_table_clear>
+C<Perl_ptr_table_clear> is no longer part of Perl's public API. Calling it
+now generates a deprecation warning, and it will be removed in a future
+release.
-=item *
+=item C<sv_compile_2op>
-Generate pp_* prototypes in pp_proto.h, and remove pp.sym
+The C<sv_compile_2op()> API function is now deprecated. Searches suggest
+that nothing on CPAN is using it, so this should have zero impact.
-Eliminate the #define pp_foo Perl_pp_foo(pTHX) macros, and update the 13
-locations that relied on them.
+It attempted to provide an API to compile code down to an optree, but failed
+to bind correctly to lexicals in the enclosing scope. It's not possible to
+fix this problem within the constraints of its parameters and return value.
-regen/opcode.pl now generates prototypes for the PP functions directly, into
-pp_proto.h. It no longer writes pp.sym, and regen/embed.pl no longer reads
-this, removing the only ordering dependency in the regen scripts. opcode.pl
-is now responsible for prototypes for pp_* functions. (embed.pl remains
-responsible for ck_* functions, reading from regen/opcodes)
+=item C<find_rundefsvoffset>
-=item *
+The C<find_rundefsvoffset> function has been deprecated. It appeared that
+its design was insufficient for reliably getting the lexical C<$_> at
+run-time.
-Fix harmless invalid read in Perl_re_compile() (f6d9469)
+Use the new C<find_rundefsv> function or the C<UNDERBAR> macro
+instead. They directly return the right SV representing C<$_>, whether it's
+lexical or dynamic.
-[perl #2460] described a case where electric fence reported an invalid
-read. This could be reproduced under valgrind with blead and -e'/x/',
-but only on a non-debugging build.
+=item C<CALL_FPTR> and C<CPERLscope>
-This was because it was checking for certain pairs of nodes (e.g. BOL + END)
-and wasn't allowing for EXACT nodes, which have the string at the next
-node position when using a naive NEXTOPER(first). In the non-debugging
-build, the nodes aren't initialised to zero, and a 1-char EXACT node isn't
-long enough to spill into the type field of the "next node".
+Those are left from an old implementation of C<MULTIPLICITY> using C++ objects,
+which was removed in Perl 5.8. Nowadays these macros do exactly nothing, so
+they shouldn't be used anymore.
-Fix this by only using NEXTOPER(first) when we know the first node is
-kosher.
+For compatibility, they are still defined for external C<XS> code. Only
+extensions defining C<PERL_CORE> must be updated now.
-=item *
+=back
-Break out the generated function Perl_keywords() into F<keywords.c>, a new file. (26ea9e1)
+=head2 Other Internal Changes
-As it and Perl_yylex() both need FEATURE_IS_ENABLED, feature_is_enabled() is
-no longer static, and the two macro definitions move from toke.c to perl.h
+=head3 Stack unwinding
-Previously, one had to cut and paste the output of perl_keywords.pl into the
-middle of toke.c, and it was not clear that it was generated code.
+The protocol for unwinding the C stack at the last stage of a C<die>
+has changed how it identifies the target stack frame. This now uses
+a separate variable C<PL_restartjmpenv>, where previously it relied on
+the C<blk_eval.cur_top_env> pointer in the C<eval> context frame that
+has nominally just been discarded. This change means that code running
+during various stages of Perl-level unwinding no longer needs to take
+care to avoid destroying the ghost frame.
-=item *
+=head3 Scope stack entries
-A lot of tests have been ported from Test to Test::More, e.g. in
-3842ad6.
+The format of entries on the scope stack has been changed, resulting in a
+reduction of memory usage of about 10%. In particular, the memory used by
+the scope stack to record each active lexical variable has been halved.
-=item *
+=head3 Memory allocation for pointer tables
-Increase default PerlIO buffer size. (b83080d)
+Memory allocation for pointer tables has been changed. Previously
+C<Perl_ptr_table_store> allocated memory from the same arena system as
+C<SV> bodies and C<HE>s, with freed memory remaining bound to those arenas
+until interpreter exit. Now it allocates memory from arenas private to the
+specific pointer table, and that memory is returned to the system when
+C<Perl_ptr_table_free> is called. Additionally, allocation and release are
+both less CPU intensive.
-The previous default size of a PerlIO buffer (4096 bytes) has been increased
-to the larger of 8192 bytes and your local BUFSIZ. Benchmarks show that doubling
-this decade-old default increases read and write performance in the neighborhood
-of 25% to 50% when using the default layers of perlio on top of unix. To choose
-a non-default size, such as to get back the old value or to obtain and even
-larger value, configure with:
+=head3 C<UNDERBAR>
- ./Configure -Accflags=-DPERLIOBUF_DEFAULT_BUFSIZ=N
+The C<UNDERBAR> macro now calls C<find_rundefsv>. C<dUNDERBAR> is now a
+noop but should still be used to ensure past and future compatibility.
-where N is the desired size in bytes; it should probably be a multiple of
-your page size.
+=head3 String comparison routines renamed
-=back
+The ibcmp_* functions have been renamed and are now called foldEQ,
+foldEQ_locale and foldEQ_utf8. The old names are still available as
+macros.
+
+=head3 C<chop> and C<chomp> implementations merged
+
+The opcode bodies for C<chop> and C<chomp> and for C<schop> and C<schomp>
+have been merged. The implementation functions C<Perl_do_chop()> and
+C<Perl_do_chomp()>, never part of the public API, have been merged and
+moved to a static function in F<pp.c>. This shrinks the perl binary
+slightly, and should not affect any code outside the core (unless it is
+relying on the order of side effects when C<chomp> is passed a I<list> of
+values).
=head1 Selected Bug Fixes
+=head2 I/O
+
=over 4
=item *
-C<binmode $fh, ':scalar'> no longer warns (8250589)
-
-Perl will now no longer produce this warning:
+Perl no longer produces this warning:
$ perl -we 'open my $f, ">", \my $x; binmode $f, "scalar"'
Use of uninitialized value in binmode at -e line 1.
=item *
-C<when(scalar){...}> no longer crashes, but produces a syntax error
-[perl #74114].
+Opening a glob reference via C<< open $fh, "E<gt>", \*glob >> will no longer
+cause the glob to be corrupted when the filehandle is printed to. This would
+cause perl to crash whenever the glob's contents were accessed
+[perl #77492].
=item *
-The C-level C<lex_stuff_pvn> function would sometimes cause a spurious
-syntax error on the last line of the file if it lacked a final semicolon
-[perl #74006].
+PerlIO no longer crashes when called recursively, e.g., from a signal
+handler. Now it just leaks memory [perl #75556].
=item *
-The regular expression engine no longer loops when matching
-C<"\N{LATIN SMALL LIGATURE FF}" =~ /f+/i> and similar expressions
-[perl #72998].
+Most I/O functions were not warning for unopened handles unless the
+'closed' and 'unopened' warnings categories were both enabled. Now only
+C<use warnings 'unopened'> is necessary to trigger these warnings (as was
+always meant to be the case).
=item *
-A label right before a string eval (C<foo: eval $string>) no longer causes
-the label to be associated also with the first statement inside the eval
-[perl #74290] (5.12.1).
+There have been several fixes to PerlIO layers:
-=item *
+When C<binmode FH, ":crlf"> pushes the C<:crlf> layer on top of the stack,
+it no longer enables crlf layers lower in the stack, to avoid unexpected
+results [perl #38456].
+
+Opening a file in C<:raw> mode now does what it advertises to do (first
+open the file, then binmode it), instead of simply leaving off the top
+layer [perl #80764].
+
+The three layers C<:pop>, C<:utf8> and C<:bytes> didn't allow stacking when
+opening a file. For example
+this:
+
+ open FH, '>:pop:perlio', 'some.file' or die $!;
+
+Would throw an error: "Invalid argument". This has been fixed in this
+release [perl #82484].
+
+=back
-Naming a deprecated character in \N{...} will not leak memory.
+=head2 Regular Expression Bug Fixes
+
+=over
=item *
-FETCH is no longer called needlessly on some tied variables.
+The regular expression engine no longer loops when matching
+C<"\N{LATIN SMALL LIGATURE FF}" =~ /f+/i> and similar expressions
+[perl #72998] (5.12.1).
=item *
=item *
-Timely cleanup of SVs that are cloned into a new thread but then
-discovered to be orphaned (i.e. their owners are -not- cloned) (e42956)
+Syntax errors in C<< (?{...}) >> blocks no longer cause panic messages
+[perl #2353].
=item *
-Don't accidentally clone lexicals in scope within active stack frames in
-the parent when creating a child thread (RT #73086) (05d04d).
+A pattern like C<(?:(o){2})?> no longer causes a "panic" error
+[perl #39233].
=item *
-Avoid loading feature.pm when 'no 5.13.2;' or similar is
-encountered (faee19).
+A fatal error in regular expressions containing C<(.*?)> when processing
+UTF-8 data has been fixed [perl #75680] (5.12.2).
=item *
-Trap invalid use of SvIVX on SVt_REGEXP when assertions are on
-(e77da3)
+An erroneous regular expression engine optimisation that caused regex verbs like
+C<*COMMIT> sometimes to be ignored has been removed.
=item *
-Don't stamp on $DB::single, $DB::trace and $DB::signal if they
-already have values when $^P is assigned to (RT #72422) (4c0f30).
+The regular expression bracketed character class C<[\8\9]> was effectively the
+same as C<[89\000]>, incorrectly matching a NULL character. It also gave
+incorrect warnings that the C<8> and C<9> were ignored. Now C<[\8\9]> is the
+same as C<[89]> and gives legitimate warnings that C<\8> and C<\9> are
+unrecognized escape sequences, passed-through.
=item *
-chop now correctly handles perl's extended UTF-8 (RT #73246) (65ab92)
+A regular expression match in the right-hand side of a global substitution
+(C<s///g>) that is in the same scope will no longer cause match variables
+to have the wrong values on subsequent iterations. This can happen when an
+array or hash subscript is interpolated in the right-hand side, as in
+C<s|(.)|@a{ print($1), /./ }|g> [perl #19078].
=item *
-Defer signal handling when shared SV locks are held to avoid
-deadlocks (RT #74868) (65c742).
+Several cases in which characters in the Latin-1 non-ASCII range (0x80 to
+0xFF) used not to match themselves or used to match both a character class
+and its complement have been fixed. For instance, U+00E2 could match both
+C<\w> and C<\W> [perl #78464] [perl #18281] [perl #60156].
=item *
-glob() no longer crashes when %File::Glob:: is empty and
-CORE::GLOBAL::glob isn't present (4984aa).
+Matching a Unicode character against an alternation containing characters
+that happened to match continuation bytes in the former's UTF8
+representation (C<qq{\x{30ab}} =~ /\xab|\xa9/>) would cause erroneous
+warnings [perl #70998].
=item *
-perlbug now always permits the sender address to be changed
-before sending - if you were having trouble sending bug reports before
-now, this should fix it, we hope (e6eb90).
+The trie optimisation was not taking empty groups into account, preventing
+'foo' from matching C</\A(?:(?:)foo|bar|zot)\z/> [perl #78356].
=item *
-Overloading now works properly in conjunction with tied
-variables. What formerly happened was that most ops checked their
-arguments for overloading I<before> checking for magic, so for example
-an overloaded object returned by a tied array access would usually be
-treated as not overloaded (RT #57012) (6f1401, ed3b9b, 6a5f8c .. 24328f).
+A pattern containing a C<+> inside a lookahead would sometimes cause an
+incorrect match failure in a global match (e.g., C</(?=(\S+))/g>)
+[perl #68564].
=item *
-Independently, a bug was fixed that prevented $tied-E<gt>() from
-always calling FETCH correctly (RT #8438) (7c7501)
+A regular expression optimisation would sometimes cause a match with a
+C<{n,m}> quantifier to fail when it should match [perl #79152].
=item *
-Some work has been done on the internal pointers that link between symbol
-tables (stashes), typeglobs and subroutines. This has the effect that
-various edge cases related to deleting stashes or stash entries (e.g.
-<%FOO:: = ()>), and complex typeglob or code reference aliasing, will no
-longer crash the interpreter.
+Case insensitive matching in regular expressions compiled under C<use
+locale> now works much more sanely when the pattern or
+target string is encoded internally in
+UTF8. Previously, under these conditions the localeness
+was completely lost. Now, code points above 255 are treated as Unicode,
+but code points between 0 and 255 are treated using the current locale
+rules, regardless of whether the pattern or the string is encoded in UTF8.
+The few case-insensitive matches that cross the 255/256 boundary are not
+allowed. For example, 0xFF does not caselessly match the character at
+0x178, LATIN CAPITAL LETTER Y WITH DIAERESIS, because 0xFF may not be
+LATIN SMALL LETTER Y in the current locale, and Perl has no way of
+knowing if that character even exists in the locale, much less what code
+point it is.
=item *
-Fixed readline() when interrupted by signals so it no longer returns
-the "same thing" as before or random memory
+The C<(?|...)> regular expression construct no longer crashes if the final
+branch has more sets of capturing parentheses than any other branch. This
+was fixed in Perl 5.10.1 for the case of a single branch, but that fix did
+not take multiple branches into account [perl #84746].
=item *
-Fixed a regression of kill() when a match variable is used for the
-process ID to kill (RT#75812) (8af710e)
+A bug has been fixed in the implementation of C<{...}> quantifiers in
+regular expressions that prevented the code block in
+C</((\w+)(?{ print $2 })){2}/> from seeing the C<$2> sometimes
+[perl #84294].
-=item *
+=back
+
+=head2 Syntax/Parsing Bugs
-Fixed several subtle bugs in sort() when @_ is accessed within a subroutine
-used for sorting (RT#72334) (8f443ca)
+=over
=item *
-Catch yyparse() exceptions in C<< (?{...}) >> (RT#2353) (634d691)
+C<when(scalar){...}> no longer crashes, but produces a syntax error
+[perl #74114] (5.12.1).
=item *
-Avoid UTF-8 cache panics with offsets beyond a string (RT #75898) (3e2d381)
+A label right before a string eval (C<foo: eval $string>) no longer causes
+the label to be associated also with the first statement inside the eval
+[perl #74290] (5.12.1).
=item *
-Fixed POSIX::strftime memory leak (RT#73520) (c4bc4aa)
+The C<no 5.13.2;> form of C<no> no longer tries to turn on features or
+pragmata (i.e., strict) [perl #70075] (5.12.2).
=item *
-Doesn't set strict with C<no VERSION> if C<VERSION> is greater than 5.12
-(da8fb5d)
+C<BEGIN {require 5.12.0}> now behaves as documented, rather than behaving
+identically to C<use 5.12.0;>. Previously, C<require> in a C<BEGIN> block
+was erroneously executing the C<use feature ':5.12.0'> and
+C<use strict;> behaviour, which only C<use> was documented to
+provide [perl #69050].
=item *
-Avoids multiple FETCH/stringify on filetest ops (40c852d)
+A regression introduced in Perl 5.12.0, making
+C<< my $x = 3; $x = length(undef) >> result in C<$x> set to C<3> has been
+fixed. C<$x> will now be C<undef> [perl #85508] (5.12.2).
=item *
-Fixed issue with string C<eval> not detecting taint of overloaded/tied
-arguments (RT #75716) (895b760)
+When strict 'refs' mode is off, C<%{...}> in rvalue context returns
+C<undef> if its argument is undefined. An optimisation introduced in perl
+5.12.0 to make C<keys %{...}> faster when used as a boolean did not take
+this into account, causing C<keys %{+undef}> (and C<keys %$foo> when
+C<$foo> is undefined) to be an error, which it should only be in strict
+mode [perl #81750].
=item *
-Fix potential crashes of string C<eval> when evaluating a object with
-overloaded stringification by creating a stringified copy when necessary
-(3e5c018)
+Constant-folding used to cause
-=item *
+ $text =~ ( 1 ? /phoo/ : /bear/)
-Fixed bug where overloaded stringification could remove tainting
-(RT #75716) (a02ec77)
+to turn into
-=item *
+ $text =~ /phoo/
-Plugs more memory leaks in vms.c. (9e2bec0)
+at compile time. Now it correctly matches against C<$_> [perl #20444].
=item *
-Fix pthread include error for Time::Piece (e9f284c)
+Parsing Perl code (either with string C<eval> or by loading modules) from
+within a C<UNITCHECK> block no longer causes the interpreter to crash
+[perl #70614].
=item *
-A possible memory leak when using L<caller()|perlfunc/"caller EXPR"> to set
-C<@DB::args> has been fixed.
+String evals no longer fail after 2 billion scopes have been
+compiled [perl #83364].
=item *
-Several memory leaks when loading XS modules were fixed.
+The parser no longer hangs when encountering certain Unicode characters,
+such as U+387 [perl #74022].
=item *
-A panic in the regular expression optimizer has been fixed (RT#75762).
+Several contexts no longer allow a Unicode character to begin a word
+that should never begin words, for an example an accent that must follow
+another character previously could precede all other characters.
=item *
-Assignments to lvalue subroutines now honor copy-on-write behavior again, which
-has been broken since version 5.10.0 (RT#75656).
+Defining a constant with the same name as one of perl's special blocks
+(e.g., INIT) stopped working in 5.12.0, but has now been fixed
+[perl #78634].
=item *
-Assignments to glob copies now behave just like assignments to regular globs
-(RT#1804).
+A reference to a literal value used as a hash key (C<$hash{\"foo"}>) used
+to be stringified, even if the hash was tied [perl #79178].
=item *
-Within signal handlers, C<$!> is now implicitly localized.
+A closure containing an C<if> statement followed by a constant or variable
+is no longer treated as a constant [perl #63540].
=item *
-L<readline|perlfunc/"readline EXPR"> now honors C<< <> >> overloading on tied
-arguments.
+C<state> can now be used with attributes. It used to mean the same thing as
+C<my> if attributes were present [perl #68658].
=item *
-L<substr()|perlfunc/"substr EXPR,OFFSET,LENGTH,REPLACEMENT">,
-L<pos()|perlfunc/"index STR,SUBSTR,POSITION">, L<keys()|perlfunc/"keys HASH">,
-and L<vec()|perlfunc/"vec EXPR,OFFSET,BITS"> could, when used in combination
-with lvalues, result in leaking the scalar value they operate on, and cause its
-destruction to happen too late. This has now been fixed.
+Expressions like C<< @$a > 3 >> no longer cause C<$a> to be mentioned in
+the "Use of uninitialized value in numeric gt" warning when C<$a> is
+undefined (since it is not part of the C<E<gt>> expression, but the operand
+of the C<@>) [perl #72090].
=item *
-Building with C<PERL_GLOBAL_STRUCT>, which has been broken accidentally in
-5.13.3, now works again.
+Accessing an element of a package array with a hard-coded number (as
+opposed to an arbitrary expression) would crash if the array did not exist.
+Usually the array would be autovivified during compilation, but typeglob
+manipulation could remove it, as in these two cases which used to crash:
+
+ *d = *a; print $d[0];
+ undef *d; print $d[0];
=item *
-A regression introduced in Perl 5.12.0, making
-C<< my $x = 3; $x = length(undef) >> result in C<$x> set to C<3> has been
-fixed. C<$x> will now be C<undef>.
+The C<-C> command line option, when used on the shebang line, can now be
+followed by other options [perl #72434].
=item *
-A fatal error in regular expressions when processing UTF-8 data has been fixed [perl #75680].
+The C<B> module was returning C<B::OP>s instead of C<B::LOGOP>s for C<entertry> [perl #80622].
+This was due to a bug in the perl core, not in C<B> itself.
+
+=back
-=item *
+=head2 Stashes, Globs and Method Lookup
-An erroneous regular expression engine optimisation that caused regex verbs like
-C<*COMMIT> to sometimes be ignored has been removed.
+Perl 5.10.0 introduced a new internal mechanism for caching MROs (method
+resolution orders, or lists of parent classes; aka "isa" caches) to make
+method lookup faster (so @ISA arrays would not have to be searched
+repeatedly). Unfortunately, this brought with it quite a few bugs. Almost
+all of these have been fixed now, along with a few MRO-related bugs that
+existed before 5.10.0:
+
+=over
=item *
-The Perl debugger now also works in taint mode [perl #76872].
+The following used to have erratic effects on method resolution, because
+the "isa" caches were not reset or otherwise ended up listing the wrong
+classes. These have been fixed.
-=item *
+=over
-Several memory leaks in cloning and freeing threaded Perl interpreters have been
-fixed [perl #77352].
+=item Aliasing packages by assigning to globs [perl #77358]
+
+=item Deleting packages by deleting their containing stash elements
+
+=item Undefining the glob containing a package (C<undef *Foo::>)
+
+=item Undefining an ISA glob (C<undef *Foo::ISA>)
+
+=item Deleting an ISA stash element (C<delete $Foo::{ISA}>)
+
+=item Sharing @ISA arrays between classes (via C<*Foo::ISA = \@Bar::ISA> or
+C<*Foo::ISA = *Bar::ISA>) [perl #77238]
+
+=back
+
+C<undef *Foo::ISA> would even stop a new C<@Foo::ISA> array from updating
+caches.
=item *
-A possible string corruption when doing regular expression matches on overloaded
-objects has been fixed [perl #77084].
+Typeglob assignments would crash if the glob's stash no longer existed, so
+long as the glob assigned to was named 'ISA' or the glob on either side of
+the assignment contained a subroutine.
=item *
-Magic applied to variables in the main package no longer affects other packages.
-See L</Magic variables outside the main package> above [perl #76138].
+C<PL_isarev>, which is accessible to Perl via C<mro::get_isarev> is now
+updated properly when packages are deleted or removed from the C<@ISA> of
+other classes. This allows many packages to be created and deleted without
+causing a memory leak [perl #75176].
+
+=back
+
+In addition, various other bugs related to typeglobs and stashes have been
+fixed:
+
+=over
=item *
-Opening a glob reference via C<< open $fh, "E<gt>", \*glob >> will no longer
-cause the glob to be corrupted when the filehandle is printed to. This would
-cause perl to crash whenever the glob's contents were accessed
-[perl #77492].
+Some work has been done on the internal pointers that link between symbol
+tables (stashes), typeglobs and subroutines. This has the effect that
+various edge cases related to deleting stashes or stash entries (e.g.
+<%FOO:: = ()>), and complex typeglob or code reference aliasing, will no
+longer crash the interpreter.
=item *
-The postincrement and postdecrement operators, C<++> and C<-->, used to cause
-leaks when being used on references. This has now been fixed.
+Assigning a reference to a glob copy now assigns to a glob slot instead of
+overwriting the glob with a scalar [perl #1804] [perl #77508].
=item *
=item *
-Perl would segfault if the undocumented C<Internals> functions that used
-reference prototypes were called with the C<&foo()> syntax, e.g.
-C<&Internals::SvREADONLY(undef)> [perl #77776].
-
-These functions now call C<SvROK> on their arguments before dereferencing them
-with C<SvRV>, and we test for this case in F<t/lib/universal.t>.
+Assigning a glob to a PVLV used to convert it to a plain string. Now it
+works correctly, and a PVLV can hold a glob. This would happen when a
+nonexistent hash or array element was passed to a subroutine:
-=item *
+ sub { $_[0] = *foo }->($hash{key});
+ # $_[0] would have been the string "*main::foo"
-When assigning a list with duplicated keys to a hash, the assignment used to
-return garbage and/or freed values:
+It also happened when a glob was assigned to, or returned from, an element
+of a tied array or hash [perl #36051].
- @a = %h = (list with some duplicate keys);
+=item *
-This has now been fixed [perl #31865].
+When trying to report C<Use of uninitialized value $Foo::BAR>, crashes could
+occur if the glob holding the global variable in question had been detached
+from its original stash by, for example, C<delete $::{'Foo::'}>. This has
+been fixed by disabling the reporting of variable names in those
+cases.
=item *
-An earlier release of the 5.13 series of Perl changed the semantics of opening a
-reference to a copy of a glob:
-
- my $var = *STDOUT;
- open my $fh, '>', \$var;
+During the restoration of a localised typeglob on scope exit, any
+destructors called as a result would be able to see the typeglob in an
+inconsistent state, containing freed entries, which could result in a
+crash. This would affect code like this:
-This was a mistake, and the previous behaviour from Perl 5.10 and 5.12, which is
-to treat \$var as a scalar reference, has now been restored.
+ local *@;
+ eval { die bless [] }; # puts an object in $@
+ sub DESTROY {
+ local $@; # boom
+ }
-=item *
+Now the glob entries are cleared before any destructors are called. This
+also means that destructors can vivify entries in the glob. So perl tries
+again and, if the entries are re-created too many times, dies with a
+'panic: gp_free...' error message.
-The regular expression bracketed character class C<[\8\9]> was effectively the
-same as C<[89\000]>, incorrectly matching a NULL character. It also gave
-incorrect warnings that the C<8> and C<9> were ignored. Now C<[\8\9]> is the
-same as C<[89]> and gives legitimate warnings that C<\8> and C<\9> are
-unrecognized escape sequences, passed-through.
+=back
-=item *
+=head2 Unicode
-C<warn()> and C<die()> now respect utf8-encoded scalars [perl #45549].
+=over
=item *
-A regular expression match in the right-hand side of a global substitution
-(C<s///g>) that is in the same scope will no longer cause match variables
-to have the wrong values on subsequent iterations. This can happen when an
-array or hash subscript is interpolated in the right-hand side, as in
-C<s|(.)|@a{ print($1), /./ }|g> [perl #19078].
+What has become known as the "Unicode Bug" is mostly resolved in this release.
+Under C<use feature 'unicode_strings'>, the internal storage format of a
+string no longer affects the external semantics. There are two known
+exceptions. User-defined case changing functions, which are planned to
+be deprecated in 5.14, require utf8-encoded strings to function; and the
+character C<LATIN SMALL LETTER SHARP S> in regular expression
+case-insensitive matching has a somewhat different set of bugs depending
+on the internal storage format. Case-insensitive matching of all
+characters that have multi-character matches, as this one does, is
+problematical in Perl [perl #58182].
=item *
-Constant-folding used to cause
-
- $text =~ ( 1 ? /phoo/ : /bear/)
-
-to turn into
+The handling of Unicode non-characters has changed.
+Previously they were mostly considered illegal, except that only one of
+the 66 of them was known about in places. The Unicode standard
+considers them legal, but forbids the "open interchange" of them.
+This is part of the change to allow the internal use of any code point
+(see L</Core Enhancements>). Together, these changes resolve
+[perl #38722], [perl #51918], [perl #51936], [perl #63446].
- $text =~ /phoo/
+=item *
-at compile time. Now it correctly matches against C<$_> [perl #20444].
+Naming a deprecated character in \N{...} no longer leaks memory.
=item *
-Parsing Perl code (either with string C<eval> or by loading modules) from
-within a C<UNITCHECK> block no longer causes the interpreter to crash
-[perl #70614].
+We fixed a bug that could cause \N{} constructs followed by a single . to
+be parsed incorrectly [perl #74978] (5.12.1).
=item *
-When C<-d> is used on the shebang (C<#!>) line, the debugger now has access
-to the lines of the main program. In the past, this sometimes worked and
-sometimes did not, depending on what order things happened to be arranged
-in memory [perl #71806].
+C<chop> now correctly handles characters above "\x{7fffffff}"
+[perl #73246].
=item *
-The C<y///> or C<tr///> operator now calls get-magic (e.g., the C<FETCH>
-method of a tie) on its left-hand side just once, not twice [perl #76814].
+Passing to C<index> an offset beyond the end of the string when the string
+is encoded internally in UTF8 no longer causes panics [perl #75898].
=item *
-String comparison (C<eq>, C<ne>, C<lt>, C<gt>, C<le>, C<ge> and
-C<cmp>) and logical not (C<not> and C<!>) operators no longer call magic
-(e.g., tie methods) twice on their operands [perl #76814].
-
-This bug was introduced in an earlier 5.13 release, and does not affect
-perl 5.12.
+C<warn()> and C<die()> now respect utf8-encoded scalars [perl #45549].
=item *
-When a tied (or other magic) variable is used as, or in, a regular
-expression, it no longer has its C<FETCH> method called twice
-[perl #76814].
+Sometimes the UTF8 length cache would not be reset on a value
+returned by substr, causing C<length(substr($uni_string,...))> to give
+wrong answers. With C<${^UTF8CACHE}> set to -1, it would produce a 'panic'
+error message, too [perl #77692].
-This bug was introduced in an earlier 5.13 release, and does not affect
-perl 5.12.
+=back
-=item *
+=head2 Ties, Overloading and Other Magic
-The C<-C> command line option can now be followed by other options
-[perl #72434].
+=over
=item *
-Assigning a glob to a PVLV used to convert it to a plain string. Now it
-works correctly, and a PVLV can hold a glob. This would happen when a
-nonexistent hash or array element was passed to a subroutine:
-
- sub { $_[0] = *foo }->($hash{key});
- # $_[0] would have been the string "*main::foo"
-
-It also happened when a glob was assigned to, or returned from, an element
-of a tied array or hash [perl #36051].
+Overloading now works properly in conjunction with tied
+variables. What formerly happened was that most ops checked their
+arguments for overloading I<before> checking for magic, so for example
+an overloaded object returned by a tied array access would usually be
+treated as not overloaded [RT #57012].
=item *
-Creating a new thread when directory handles were open used to cause a
-crash, because the handles were not cloned, but simply passed to the new
-thread, resulting in a double free.
+Various cases of magic (e.g., tie methods) being called on tied variables
+too many or too few times have been fixed:
-Now directory handles are cloned properly, on systems that have a C<fchdir>
-function. On other systems, new threads simply do not inherit directory
-handles from their parent threads [perl #75154].
+=over
=item *
-The regular expression parser no longer hangs when parsing C<\18> and
-C<\88>.
-
-This bug was introduced in version 5.13.5 and did not affect earlier
-versions [perl #78058].
+FETCH is no longer called on tied variables in void context.
=item *
-Subroutine redefinition works once more in the debugger [perl #48332].
+C<$tied-E<gt>()> did not always call FETCH [perl #8438].
=item *
-The C<&> C<|> C<^> bitwise operators no longer coerce read-only arguments
-[perl #20661].
+Filetest operators and C<y///> and C<tr///> were calling FETCH too
+many times.
=item *
-Stringifying a scalar containing -0.0 no longer has the affect of turning
-false into true [perl #45133].
+The C<=> operator used to ignore magic on its right-hand side if the
+scalar happened to hold a typeglob (if a typeglob was the last thing
+returned from or assigned to a tied scalar) [perl #77498].
=item *
-Aliasing packages by assigning to globs or deleting packages by deleting
-their containing stash elements used to have erratic effects on method
-resolution, because the internal 'isa' caches were not reset. This has been
-fixed.
+Dereference operators used to ignore magic if the argument was a
+reference already (e.g., from a previous FETCH) [perl #72144].
=item *
-C<sort> with a custom sort routine could crash if too many nested
-subroutine calls occurred from within the sort routine [perl #77930].
-
-This bug was introduced in an earlier 5.13 release, and did not affect
-perl 5.12.
+C<splice> now calls set-magic (so changes made
+by C<splice @ISA> are respected by method calls) [perl #78400].
=item *
-The C<eval_sv> and C<eval_pv> C functions now set C<$@> correctly when
-there is a syntax error and no C<G_KEEPERR> flag, and never set it if the
-C<G_KEEPERR> flag is present [perl #3719].
+In-memory files created by C<open $fh, 'E<gt>' \$buffer> were not calling
+FETCH/STORE at all [perl #43789] (5.12.2).
+
+=back
=item *
-Nested C<map> and C<grep> blocks no longer leak memory when processing
-large lists [perl #48004].
+String C<eval> now detects taintedness of overloaded or tied
+arguments [perl #75716].
=item *
-Malformed C<version> objects no longer cause crashes [perl #78286].
+String C<eval> and regular expression matches against objects with string
+overloading no longer cause memory corruption or crashes [perl 77084].
=item *
-The interpreter no longer crashes when freeing deeply-nested arrays of
-arrays. Hashes have not been fixed yet [perl #44225].
+L<readline|perlfunc/"readline EXPR"> now honors C<< <> >> overloading on tied
+arguments.
=item *
-The mechanism for freeing objects in globs used to leave dangling
-pointers to freed SVs, meaning Perl users could see corrupted state
-during destruction.
+C<< E<lt>exprE<gt> >> always respects overloading now if the expression is
+overloaded.
-Perl now only frees the affected slots of the GV, rather than freeing
-the GV itself. This makes sure that there are no dangling refs or
-corrupted state during destruction.
+Due to the way that 'E<lt>E<gt> as glob' was parsed differently from
+'E<lt>E<gt> as filehandle' from 5.6 onwards, something like C<< E<lt>$foo[0]E<gt> >> did
+not handle overloading, even if C<$foo[0]> was an overloaded object. This
+was contrary to the documentation for overload, and meant that C<< E<lt>E<gt> >>
+could not be used as a general overloaded iterator operator.
=item *
-The typeglob C<*,>, which holds the scalar variable C<$,> (output field
-separator), had the wrong reference count in child threads.
+The fallback behaviour of overloading on binary operators was asymmetric
+[perl #71286].
=item *
-C<splice> now calls set-magic. This means that, for instance, changes made
-by C<splice @ISA> are respected by method calls [perl #78400].
+Magic applied to variables in the main package no longer affects other packages.
+See L</Magic variables outside the main package> above [perl #76138].
=item *
-C<use v5.8> no longer leaks memory [perl #78436].
+Sometimes magic (ties, taintedness, etc.) attached to variables could cause
+an object to last longer than it should, or cause a crash if a tied
+variable were freed from within a tie method. These have been fixed
+[perl #81230].
=item *
-The XS multicall API no longer causes subroutines to lose reference counts
-if called via the multicall interface from within those very subroutines.
-This affects modules like List::Util. Calling one of its functions with an
-active subroutine as the first argument could cause a crash [perl #78070].
+DESTROY methods of objects implementing ties are no longer able to crash by
+accessing the tied variable through a weak reference [perl #86328].
=item *
-The C<parse_stmt> C function added in earlier in the 5.13.x series has been
-fixed to work with statements ending with C<}> [perl #78222].
+Fixed a regression of kill() when a match variable is used for the
+process ID to kill [perl #75812].
=item *
-The C<parse_fullstmt> C function added in 5.13.5 has been fixed to work
-when called while an expression is being parsed.
+C<$AUTOLOAD> used to remain tainted forever if it ever became tainted. Now
+it is correctly untainted if an autoloaded method is called and the method
+name was not tainted.
=item *
-Characters in the Latin-1 non-ASCII range (0x80 to 0xFF) used not to match
-themselves if the string happened to be UTF8-encoded internally, the
-regular expression was not, and the character in the regular expression was
-inside a repeated group (e.g.,
-C<Encode::decode_utf8("\303\200") =~ /(\xc0)+/>) [perl #78464].
+C<sprintf> now dies when passed a tainted scalar for the format. It did
+already die for arbitrary expressions, but not for simple scalars
+[perl #82250].
=item *
-The C<(?d)> regular expression construct now overrides a previous C<(?u)>
-or C<use feature "unicode_string"> [perl #78508].
-
-=item *
+utf8::is_utf8 now respects get-magic (e.g. $1) (5.12.1).
-A memory leak in C<do "file">, introduced in perl 5.13.6, has been fixed
-[perl #78488].
+=back
-=item *
+=head2 The Debugger
-Various bugs related to typeglob dereferencing have been fixed. See
-L</Dereferencing typeglobs>, above.
+=over
=item *
-The C<SvPVbyte> function available to XS modules now calls magic before
-downgrading the SV, to avoid warnings about wide characters [perl #72398].
+The Perl debugger now also works in taint mode [perl #76872].
=item *
-The C<=> operator used to ignore magic (e.g., tie methods) on its
-right-hand side if the scalar happened to hold a typeglob. This could
-happen if a typeglob was the last thing returned from or assigned to a tied
-scalar [perl #77498].
+Subroutine redefinition works once more in the debugger [perl #48332].
=item *
-C<sprintf> was ignoring locales when called with constant arguments
-[perl #78632].
+When C<-d> is used on the shebang (C<#!>) line, the debugger now has access
+to the lines of the main program. In the past, this sometimes worked and
+sometimes did not, depending on what order things happened to be arranged
+in memory [perl #71806].
=item *
-A non-ASCII character in the Latin-1 range could match both a Posix
-class, such as C<[[:alnum:]]>, and its inverse C<[[:^alnum:]]>. This is
-now fixed for regular expressions compiled under the C<"u"> modifier.
-See L</C<use feature "unicode_strings"> now applies to more regex matching>. [perl #18281].
+A possible memory leak when using L<caller()|perlfunc/"caller EXPR"> to set
+C<@DB::args> has been fixed (5.12.2).
=item *
-Concatenating long strings under C<use encoding> no longer causes perl to
-crash [perl #78674].
+Perl no longer stomps on $DB::single, $DB::trace and $DB::signal if they
+already have values when $^P is assigned to [perl #72422].
=item *
-Typeglob assignments would crash if the glob's stash no longer existed, so
-long as the glob assigned to was named 'ISA' or the glob on either side of
-the assignment contained a subroutine.
-
-=item *
-
-Calling C<< ->import >> on a class lacking an import method could corrupt
-the stack, resulting in strange behaviour. For instance,
-
- push @a, "foo", $b = bar->import;
+C<#line> directives in string evals were not properly updating the arrays
+of lines of code (C<< @{"_<..."} >>) that the debugger (or any debugging or
+profiling module) uses. In threaded builds, they were not being updated at
+all. In non-threaded builds, the line number was ignored, so any change to
+the existing line number would cause the lines to be misnumbered
+[perl #79442].
-would assign 'foo' to C<$b> [perl #63790].
+=back
-=item *
+=head2 Threads
-Creating an alias to a package when that package had been detached from the
-symbol table would result in corrupted isa caches [perl #77358].
+=over
=item *
-C<.=> followed by C<< <> >> or C<readline> would leak memory if C<$/>
-contained characters beyond the octet range and the scalar assigned to
-happened to be encoded as UTF8 internally [perl #72246].
+Perl no longer accidentally clones lexicals in scope within active stack
+frames in the parent when creating a child thread [perl #73086].
=item *
-The C<recv> function could crash when called with the MSG_TRUNC flag
-[perl #75082].
+Several memory leaks in cloning and freeing threaded Perl interpreters have been
+fixed [perl #77352].
=item *
-Evaluating a simple glob (like C<*a>) was calling get-magic on the glob,
-even when its contents were not being used [perl #78580].
+Creating a new thread when directory handles were open used to cause a
+crash, because the handles were not cloned, but simply passed to the new
+thread, resulting in a double free.
-This bug was introduced in 5.13.2 and did not affect earlier perl versions.
+Now directory handles are cloned properly, on systems that have a C<fchdir>
+function. On other systems, new threads simply do not inherit directory
+handles from their parent threads [perl #75154].
=item *
-Matching a Unicode character against an alternation containing characters
-that happened to match continuation bytes in the former's UTF8
-representation (C<qq{\x{30ab}} =~ /\xab|\xa9/>) would cause erroneous
-warnings [perl #70998].
+The typeglob C<*,>, which holds the scalar variable C<$,> (output field
+separator), had the wrong reference count in child threads.
=item *
-C<s///r> (added in 5.13.2) no longer leaks.
+[perl #78494] When pipes are shared between threads, the C<close> function
+(and any implicit close, such as on thread exit) no longer blocks.
=item *
-The trie optimisation was not taking empty groups into account, preventing
-'foo' from matching C</\A(?:(?:)foo|bar|zot)\z/> [perl #78356].
-
-=item *
+Perl now does a timely cleanup of SVs that are cloned into a new thread but
+then discovered to be orphaned (i.e., their owners are I<not> cloned). This
+eliminates several "scalars leaked" warnings when joining threads.
-A pattern containing a C<+> inside a lookahead would sometimes cause an
-incorrect match failure in a global match (e.g., C</(?=(\S+))/g>)
-[perl #68564].
+=back
-=item *
+=head2 Scoping and Subroutines
-Iterating with C<foreach> over an array returned by an lvalue sub now works
-[perl #23790].
+=over
=item *
-C<$@> is now localised during calls to C<binmode> to prevent action at a
-distance [perl #78844].
+Lvalue subroutines are again able to return copy-on-write scalars. This
+had been broken since version 5.10.0 [perl #75656] (5.12.3).
=item *
-C<PL_isarev>, which is accessible to Perl via C<mro::get_isarev> is now
-updated properly when packages are deleted or removed from the C<@ISA> of
-other classes. This allows many packages to be created and deleted without
-causing a memory leak [perl #75176].
+C<require> no longer causes C<caller> to return the wrong file name for
+the scope that called C<require> and other scopes higher up that had the
+same file name [perl #68712].
=item *
-C<undef *Foo::> and C<undef *Foo::ISA> and C<delete $package::{ISA}>
-used not to update the internal isa caches if the
-stash or C<@ISA> array had a reference elsewhere. In
-fact, C<undef *Foo::ISA> would stop a new C<@Foo::ISA> array from updating
-caches.
+C<sort> with a ($$)-prototyped comparison routine used to cause the value
+of @_ to leak out of the sort. Taking a reference to @_ within the
+sorting routine could cause a crash [perl #72334].
=item *
-C<@ISA> arrays can now be shared between classes via
-C<*Foo::ISA = \@Bar::ISA> or C<*Foo::ISA = *Bar::ISA> [perl #77238].
+Match variables (e.g., C<$1>) no longer persist between calls to a sort
+subroutine [perl #76026].
=item *
-The parser no longer hangs when encountering certain Unicode characters,
-such as U+387 [perl #74022].
+Iterating with C<foreach> over an array returned by an lvalue sub now works
+[perl #23790].
=item *
-C<formline> no longer crashes when passed a tainted format picture. It also
-taints C<$^A> now if its arguments are tainted [perl #79138].
+C<$@> is now localised during calls to C<binmode> to prevent action at a
+distance [perl #78844].
=item *
-A signal handler called within a signal handler could cause leaks or
-double-frees. Now fixed. [perl #76248].
+Calling a closure prototype (what is passed to an attribute handler for a
+closure) now results in a "Closure prototype called" error message instead
+of a crash [perl #68560].
=item *
-When trying to report C<Use of uninitialized value $Foo::BAR>, crashes could
-occur if the GLOB of the global variable causing the warning has been detached
-from its original stash by, for example C<delete $::{'Foo::'}>. This has been
-fixed by disabling the reporting of variable names in the warning in those
-cases.
-
-=item *
+Mentioning a read-only lexical variable from the enclosing scope in a
+string C<eval> no longer causes the variable to become writable
+[perl #19135].
-C<BEGIN {require 5.12.0}> now behaves as documented, rather than behaving
-identically to C<use 5.12.0;>. Previously, C<require> in a C<BEGIN> block
-was erroneously executing the C<use feature ':5.12.0'> and
-C<use strict; use warnings;> behaviour, which only C<use> was documented to
-provide [perl #69050].
+=back
-=item *
+=head2 Signals
-C<use 5.42> [perl #69050],
-C<use 6> and C<no 5> no longer leak memory.
+=over
=item *
-C<eval "BEGIN{die}"> no longer leaks memory on non-threaded builds.
+Within signal handlers, C<$!> is now implicitly localized.
=item *
-PerlIO no longer crashes when called recursively, e.g., from a signal
-handler. Now it just leaks memory [perl #75556].
+CHLD signals are no longer unblocked after a signal handler is called if
+they were blocked before by C<POSIX::sigprocmask> [perl #82040].
=item *
-Defining a constant with the same name as one of perl's special blocks
-(e.g., INIT) stopped working in 5.12.0, but has now been fixed
-[perl #78634].
-
-=item *
+A signal handler called within a signal handler could cause leaks or
+double-frees. Now fixed. [perl #76248].
-A reference to a literal value used as a hash key (C<$hash{\"foo"}>) used
-to be stringified, even if the hash was tied [perl #79178].
+=back
-=item *
+=head2 Miscellaneous Memory Leaks
-A closure containing an C<if> statement followed by a constant or variable
-is no longer treated as a constant [perl #63540].
+=over
=item *
-Calling a closure prototype (what is passed to an attribute handler for a
-closure) now results in a "Closure prototype called" error message instead
-of a crash [perl #68560].
+Several memory leaks when loading XS modules were fixed (5.12.2).
=item *
-A regular expression optimisation would sometimes cause a match with a
-C<{n,m}> quantifier to fail when it should match [perl #79152].
+L<substr()|perlfunc/"substr EXPR,OFFSET,LENGTH,REPLACEMENT">,
+L<pos()|perlfunc/"index STR,SUBSTR,POSITION">, L<keys()|perlfunc/"keys HASH">,
+and L<vec()|perlfunc/"vec EXPR,OFFSET,BITS"> could, when used in combination
+with lvalues, result in leaking the scalar value they operate on, and cause its
+destruction to happen too late. This has now been fixed.
=item *
-What has become known as the "Unicode Bug" is mostly resolved in this release.
-Under C<use feature 'unicode_strings'>, the internal storage format of a
-string no longer affects the external semantics. There are two known
-exceptions. User-defined case changing functions, which are planned to
-be deprecated in 5.14, require utf8-encoded strings to function; and the
-character C<LATIN SMALL LETTER SHARP S> in regular expression
-case-insensitive matching has a somewhat different set of bugs depending
-on the internal storage format. Case-insensitive matching of all
-characters that have multi-character matches, as this one does, is
-problematical in Perl [perl #58182].
+The postincrement and postdecrement operators, C<++> and C<-->, used to cause
+leaks when being used on references. This has now been fixed.
=item *
-Mentioning a read-only lexical variable from the enclosing scope in a
-string C<eval> no longer causes the variable to become writable
-[perl #19135].
+Nested C<map> and C<grep> blocks no longer leak memory when processing
+large lists [perl #48004].
=item *
-C<state> can now be used with attributes. It used to mean the same thing as
-C<my> if attributes were present [perl #68658].
+C<use I<VERSION>> and C<no I<VERSION>> no longer leak memory [perl #78436]
+[perl #69050].
=item *
-Expressions like C<< @$a > 3 >> no longer cause C<$a> to be mentioned in
-the "Use of uninitialized value in numeric gt" warning when C<$a> is
-undefined (since it is not part of the C<E<gt>> expression, but the operand
-of the C<@>) [perl #72090].
+C<.=> followed by C<< <> >> or C<readline> would leak memory if C<$/>
+contained characters beyond the octet range and the scalar assigned to
+happened to be encoded as UTF8 internally [perl #72246].
=item *
-C<require> no longer causes C<caller> to return the wrong file name for
-the scope that called C<require> and other scopes higher up that had the
-same file name [perl #68712].
-
-=item *
+C<eval "BEGIN{die}"> no longer leaks memory on non-threaded builds.
-The ref types in the typemap for XS bindings now support magical variables
-[perl #72684].
+=back
-=item *
+=head2 Memory Corruption and Crashes
-Match variables (e.g., C<$1>) no longer persist between calls to a sort
-subroutine [perl #76026].
+=over
=item *
-The C<B> module was returning C<B::OP>s instead of C<B::LOGOP>s for C<entertry> [perl #80622].
-This was due to a bug in the perl core, not in C<B> itself.
+glob() no longer crashes when %File::Glob:: is empty and
+CORE::GLOBAL::glob isn't present [perl #75464] (5.12.2).
=item *
-Some numeric operators were converting integers to floating point,
-resulting in loss of precision on 64-bit platforms [perl #77456].
+readline() has been fixed when interrupted by signals so it no longer
+returns the "same thing" as before or random memory.
=item *
-The fallback behaviour of overloading on binary operators was asymmetric
-[perl #71286].
+When assigning a list with duplicated keys to a hash, the assignment used to
+return garbage and/or freed values:
-=item *
+ @a = %h = (list with some duplicate keys);
-The handling of Unicode non-characters has changed.
-Previously they were mostly considered illegal, except that only one of
-the 66 of them was known about in places. The Unicode standard
-considers them legal, but forbids the "open interchange" of them.
-This is part of the change to allow the internal use of any code point
-(see L</Core Enhancements>). Together, these changes resolve
-[perl #38722], [perl #51918], [perl #51936], [perl #63446]
+This has now been fixed [perl #31865].
=item *
-Sometimes magic (ties, tainted, etc.) attached to variables could cause an
-object to last longer than it should, or cause a crash if a tied variable
-were freed from within a tie method. These have been fixed [perl #81230].
+The mechanism for freeing objects in globs used to leave dangling
+pointers to freed SVs, meaning Perl users could see corrupted state
+during destruction.
+
+Perl now only frees the affected slots of the GV, rather than freeing
+the GV itself. This makes sure that there are no dangling refs or
+corrupted state during destruction.
=item *
-Most I/O functions were not warning for unopened handles unless the
-'closed' and 'unopened' warnings categories were both enabled. Now only
-C<use warnings 'unopened'> is necessary to trigger these warnings (as was
-always meant to be the case.
+The interpreter no longer crashes when freeing deeply-nested arrays of
+arrays. Hashes have not been fixed yet [perl #44225].
=item *
-C<< E<lt>exprE<gt> >> always respects overloading now if the expression is
-overloaded.
-
-Due to the way that 'E<lt>E<gt> as glob' was parsed differently from
-'E<lt>E<gt> as filehandle' from 5.6 onwards, something like C<< E<lt>$foo[0]E<gt> >> did
-not handle overloading, even if C<$foo[0]> was an overloaded object. This
-was contrary to the documentation for overload, and meant that C<< E<lt>E<gt> >>
-could not be used as a general overloaded iterator operator.
+Concatenating long strings under C<use encoding> no longer causes perl to
+crash [perl #78674].
=item *
-Destructors on objects were not called during global destruction on objects
-that were not referenced by any scalars. This could happen if an array
-element were blessed (e.g., C<bless \$a[0]>) or if a closure referenced a
-blessed variable (C<bless \my @a; sub foo { @a }>).
-
-Now there is an extra pass during global destruction to fire destructors on
-any objects that might be left after the usual passes that check for
-objects referenced by scalars
-[perl #36347].
+Calling C<< ->import >> on a class lacking an import method could corrupt
+the stack, resulting in strange behaviour. For instance,
-=item *
+ push @a, "foo", $b = bar->import;
-A long standing bug has now been fully fixed (partial fixes came in
-earlier releases), in which some Latin-1 non-ASCII characters on
-ASCII-platforms would match both a character class and its complement,
-such as U+00E2 being both in C<\w> and C<\W>, depending on the
-UTF-8-ness of the regular expression pattern and target string.
-Fixing this did expose some bugs in various modules and tests that
-relied on the previous behavior of C<[[:alpha:]]> not ever matching
-U+00FF, "LATIN SMALL LETTER Y WITH DIAERESIS", even when it should, in
-Unicode mode; now it does match when appropriate.
-[perl #60156].
+would assign 'foo' to C<$b> [perl #63790].
=item *
-A Unicode C<\p{}> property match in a regular expression pattern will
-now force Unicode rules for the rest of the regular expression
+The C<recv> function could crash when called with the MSG_TRUNC flag
+[perl #75082].
=item *
-[perl #38456] binmode FH, ":crlf" only modifies top crlf layer (7826b36)
-
-When pushed on top of the stack, crlf will no longer enable crlf layers
-lower in the stack. This will prevent unexpected results.
+C<formline> no longer crashes when passed a tainted format picture. It also
+taints C<$^A> now if its arguments are tainted [perl #79138].
=item *
-Fix 'raw' layer for RT #80764 (ecfd064)
-
-Made a ':raw' open do what it advertises to do (first open the file,
-then binmode it), instead of leaving off the top layer.
+A bug in how we process filetest operations could cause a segfault.
+Filetests don't always expect an op on the stack, so we now use
+TOPs only if we're sure that we're not stat'ing the _ filehandle.
+This is indicated by OPf_KIDS (as checked in ck_ftst) [perl #74542]
+(5.12.1).
=item *
-Use PerlIOBase_open for pop, utf8 and bytes layers (c0888ac)
+C<unpack()> now handles scalar context correctly for C<%32H> and C<%32u>,
+fixing a potential crash. C<split()> would crash because the third item
+on the stack wasn't the regular expression it expected. C<unpack("%2H",
+...)> would return both the unpacked result and the checksum on the stack,
+as would C<unpack("%2u", ...)> [perl #73814] (5.12.2).
-Three of Perl's builtin PerlIO layers (C<:pop>, C<:utf8> and
-C<:bytes>) didn't allow stacking when opening a file. For example
-this:
+=back
- open FH, '>:pop:perlio', 'some.file' or die $!;
+=head2 Fixes to Various Perl Operators
-Would throw an error: "Invalid argument". This has been fixed in this
-release.
+=over
=item *
-An issue present since 5.13.1, where s/A/B/ with A utf8 and B
-non-utf8, could cause corruption or segfaults has been
-fixed. (c95ca9b)
+The C<&> C<|> C<^> bitwise operators no longer coerce read-only arguments
+[perl #20661].
=item *
-String evals will no longer fail after 2 billion scopes have been
-compiled (d1bfb64, 2df5bdd, 0d311cd and 6012dc8)
+Stringifying a scalar containing -0.0 no longer has the affect of turning
+false into true [perl #45133].
=item *
-[perl #81750] When strict 'refs' mode is off,
-C<%{...}> in rvalue context returns C<undef> if
-its argument is undefined. An optimisation introduced in perl 5.12.0 to
-make C<keys %{...}> faster when used as a boolean did not take this into
-account, causing C<keys %{+undef}> (and C<keys %$foo> when C<$foo> is
-undefined) to be an error, which it should only be in strict mode.
+Some numeric operators were converting integers to floating point,
+resulting in loss of precision on 64-bit platforms [perl #77456].
=item *
-[perl #83194] Combining the vector (%v) flag and dynamic precision would
-cause sprintf to confuse the order of its arguments, making it treat the
-string as the precision and vice versa.
+C<sprintf> was ignoring locales when called with constant arguments
+[perl #78632].
=item *
-[perl #77692] Sometimes the UTF8 length cache would not be reset on a value
-returned by substr, causing C<length(substr($uni_string,...))> to give
-wrong answers. With C<${^UTF8CACHE}> set to -1, it would produce a 'panic'
-error message, too.
-
-=item *
+Combining the vector (%v) flag and dynamic precision would
+cause sprintf to confuse the order of its arguments, making it treat the
+string as the precision and vice versa [perl #83194].
-During the restoration of a localised typeglob on scope exit, any
-destructors called as a result would be able to see the typeglob in an
-inconsistent state, containing freed entries, which could result in a
-crash. This would affect code like this:
+=back
- local *@;
- eval { die bless [] }; # puts an object in $@
- sub DESTROY {
- local $@; # boom
- }
+=head2 Bugs Relating to the C API
-Now the glob entries are cleared before any destructors are called. This
-also means that destructors can vivify entries in the glob. So perl tries
-again and, if the entries are re-created too many times, dies with a
-'panic: gp_free...' error message.
+=over
=item *
-[perl #78494] When pipes are shared between threads, the C<close> function
-(and any implicit close, such as on thread exit) no longer blocks.
+The C-level C<lex_stuff_pvn> function would sometimes cause a spurious
+syntax error on the last line of the file if it lacked a final semicolon
+[perl #74006] (5.12.1).
=item *
-Several contexts no longer allow a Unicode character to begin a word
-that should never begin words, for an example an accent that must follow
-another character previously could precede all other characters.
+The C<eval_sv> and C<eval_pv> C functions now set C<$@> correctly when
+there is a syntax error and no C<G_KEEPERR> flag, and never set it if the
+C<G_KEEPERR> flag is present [perl #3719].
=item *
-Case insensitive matching in regular expressions compiled under C<use
-locale> now works much more sanely when the pattern and/or target string
-are encoded in UTF-8. Previously, under these conditions the localeness
-was completely lost. Now, code points above 255 are treated as Unicode,
-but code points between 0 and 255 are treated using the current locale
-rules, regardless of whether the pattern or string are encoded in UTF-8.
-The few case insensitive matches that cross the 255/256 boundary are not
-allowed. For example, 0xFF does not caselessly match the character at
-0x178, LATIN CAPITAL LETTER Y WITH DIAERESIS, because 0xFF may not be
-LATIN SMALL LETTER Y in the current locale, and Perl has no way of
-knowing if that character even exists in the locale, much less what code
-point it is.
+The XS multicall API no longer causes subroutines to lose reference counts
+if called via the multicall interface from within those very subroutines.
+This affects modules like List::Util. Calling one of its functions with an
+active subroutine as the first argument could cause a crash [perl #78070].
=item *
-A fix for a bug in C<length(undef)> in 5.13.4 introduced a regression that
-meant C<print length undef> did not warn when warnings were enabled. It now
-correctly warns [perl #85508].
+The C<SvPVbyte> function available to XS modules now calls magic before
+downgrading the SV, to avoid warnings about wide characters [perl #72398].
=item *
-The C<(?|...)> regular expression construct no longer crashes if the final
-branch has more sets of capturing parentheses than any other branch. This
-was fixed in Perl 5.10.1 for the case of a single branch, but that fix did
-not take multiple branches into account [perl #84746].
+The ref types in the typemap for XS bindings now support magical variables
+[perl #72684].
=item *
-Accessing an element of a package array with a hard-coded number (as
-opposed to an arbitrary expression) would crash if the array did not exist.
-Usually the array would be autovivified during compilation, but typeglob
-manipulation could remove it, as in these two cases which used to crash:
+C<sv_catsv_flags> no longer calls C<mg_get> on its second argument (the
+source string) if the flags passed to it do not include SV_GMAGIC. So it
+now matches the documentation.
- *d = *a; print $d[0];
- undef *d; print $d[0];
+=item *
+
+C<my_strftime> no longer leaks memory. This fixes a memory leak in
+C<POSIX::strftime> [perl #73520].
=item *
-C<#line> directives in string evals were not properly updating the arrays
-of lines of code (C<< @{"_<..."} >>) that the debugger (or any debugging or
-profiling module) uses. In threaded builds, they were not being updated at
-all. In non-threaded builds, the line number was ignored, so any change to
-the existing line number would cause the lines to be misnumbered
-[perl #79442].
+XSUB.h now correctly redefines fgets under PERL_IMPLICIT_SYS [perl #55049]
+(5.12.1).
=item *
-C<$AUTOLOAD> used to remain tainted forever if it ever became tainted. Now
-it is correctly untainted if an autoloaded method is called and the method
-name was not tainted.
+XS code using C<fputc()> or C<fputs()>: on Windows could cause an error
+due to their arguments being swapped [perl #72704] (5.12.1).
=item *
-A bug has been fixed in the implementation of C<{...}> quantifiers in
-regular expressions that prevented the code block in
-C</((\w+)(?{ print $2 })){2}/> from seeing the C<$2> sometimes
-[perl #84294].
+A possible segfault in the C<T_PRTOBJ> default typemap has been fixed
+(5.12.2).
=item *
-C<sprintf> now dies when passed a tainted scalar for the format. It did
-already die for arbitrary expressions, but not for simple scalars
-[perl #82250].
+A bug that could cause "Unknown error" messages when
+C<call_sv(code, G_EVAL)> is called from an XS destructor has been fixed
+(5.12.2).
=back
=head1 Known Problems
+XXX Many of these have probably already been solved. There are also
+unresolved BBC articles linked to #77718 that are awaiting CPAN
+releases. These may need to be listed here.
+See also #84444. Enbugger may also need to be listed if there is no new
+release in time (see #82152).
+JJORE/overload-eval-0.08.tar.gz appears to be broken, too. See
+http://www.nntp.perl.org/group/perl.perl5.porters/2010/11/msg165773.html
+
=over 4
=item *
-Bug fixes involving CvGV reference counting break Sub::Name. A
-patch has been sent upstream to the maintainer
+C<List::Util::first> misbehaves in the presence of a lexical C<$_>
+(typically introduced by C<my $_> or implicitly by C<given>). The variable
+which gets set for each iteration is the package variable C<$_>, not the
+lexical C<$_>.
+
+A similar issue may occur in other modules that provide functions which
+take a block as their first argument, like
+
+ foo { ... $_ ...} list
+
+See also: L<http://rt.perl.org/rt3/Public/Bug/Display.html?id=67694>
=item *
=head1 Errata
-=head2 C<keys>, C<values> work on arrays
+=head2 C<keys>, C<values> and C<each> work on arrays
You can now use the C<keys>, C<values>, C<each> builtin functions on arrays
(previously you could only use them on hashes). See L<perlfunc> for details.
https://perl5.git.perl.org / perl5.git / blobdiff