File Coverage

blib/lib/Syntax/Keyword/Try.pm

Criterion	Covered	Total	%
statement	35	37	94.5
branch	14	18	77.7
condition	1	3	33.3
subroutine	10	10	100.0
pod	0	3	0.0
total	60	71	84.5

#  You may distribute under the terms of either the GNU General Public License
#  or the Artistic License (the same terms as Perl itself)
#
#  (C) Paul Evans, 2016-2023 -- [e-mail]

package Syntax::Keyword::Try 0.29;

use v5.14;
use warnings;

use Carp;

require XSLoader;
XSLoader::load( __PACKAGE__, our $VERSION );

=head1 NAME

C - a C syntax for perl

=head1 SYNOPSIS

   use Syntax::Keyword::Try;

   sub foo {
      try {
         attempt_a_thing();
         return "success";
      }
      catch ($e) {
         warn "It failed - $e";
         return "failure";
      }
   }

=head1 DESCRIPTION

This module provides a syntax plugin that implements exception-handling
semantics in a form familiar to users of other languages, being built on a
block labeled with the C keyword, followed by at least one of a C
or C block.

As well as providing a handy syntax for this useful behaviour, this module
also serves to contain a number of code examples for how to implement parser
plugins and manipulate optrees to provide new syntax and behaviours for perl
code.

Syntax similar to this module has now been added to core perl, starting at
version 5.34.0. If you are writing new code, it is suggested that you instead
use the L module instead, as that will enable the core
feature on those supported perl versions, falling back to
C on older perls.

=head1 Experimental Features

Some of the features of this module are currently marked as experimental. They
will provoke warnings in the C category, unless silenced.

You can silence this with C but then that will
silence every experimental warning, which may hide others unintentionally. For
a more fine-grained approach you can instead use the import line for this
module to only silence this module's warnings selectively:

   use Syntax::Keyword::Try qw( try :experimental(typed) );

   use Syntax::Keyword::Try qw( try :experimental );  # all of the above

Don't forget to import the main C symbol itself, to activate the syntax.

=cut

=head1 KEYWORDS

=head2 try

   try {
      STATEMENTS...
   }
   ...

A C statement provides the main body of code that will be invoked, and
must be followed by either a C statement, a C statement, or
both.

Execution of the C statement itself begins from the block given to the
statement and continues until either it throws an exception, or completes
successfully by reaching the end of the block. What will happen next depends
on the presence of a C or C statement immediately following
it.

The body of a C block may contain a C expression. If executed,
such an expression will cause the entire containing function to return with
the value provided. This is different from a plain C block, in which
circumstance only the C itself would return, not the entire function.

The body of a C block may contain loop control expressions (C,
C, C) which will have their usual effect on any loops that the
C block is contained by.

The parsing rules for the set of statements (the C block and its
associated C and C) are such that they are parsed as a self-
contained statement. Because of this, there is no need to end with a
terminating semicolon.

Even though it parses as a statement and not an expression, a C block can
still yield a value if it appears as the final statement in its containing
C or C block. For example:

   my $result = do {
      try { attempt_func() }
      catch ($e) { "Fallback Value" }
   };

Note (especially to users of L and similar) that the C
block itself does not necessarily stop exceptions thrown inside it from
propagating outside. It is the presence of a later C block which
causes this to happen. A C with only a C and no C will
still propagate exceptions up to callers as normal.

=head2 catch

   ...
   catch ($var) {
      STATEMENTS...
   }

or

   ...
   catch {
      STATEMENTS...
   }

A C statement provides a block of code to the preceding C
statement that will be invoked in the case that the main block of code throws
an exception. Optionally a new lexical variable can be provided to store the
exception in. If not provided, the C block can inspect the raised
exception by looking in C<$@> instead.

Presence of this C statement causes any exception thrown by the
preceding C block to be non-fatal to the surrounding code. If the
C block wishes to optionally handle some exceptions but not others, it
can re-raise it (or another exception) by calling C in the usual manner.

As with C, the body of a C block may also contain a C
expression, which as before, has its usual meaning, causing the entire
containing function to return with the given value. The body may also contain
loop control expressions (C, C or C) which also have their
usual effect.

If a C statement is not given, then any exceptions raised by the C
block are raised to the caller in the usual way.

=head2 catch (Typed)

   ...
   catch ($var isa Class) { ... }

   ...
   catch ($var =~ m/^Regexp match/) { ... }

I

Optionally, multiple catch statements can be provided, where each block is
given a guarding condition, to control whether or not it will catch particular
exception values. Use of this syntax will provoke an C category
warning on supporting perl versions, unless silenced by importing the
C<:experimental(typed)> tag (see above).

Two kinds of condition are supported:

=over 4

=item *

   catch ($var isa Class)

The block is invoked only if the caught exception is a blessed object, and
derives from the given package name.

On Perl version 5.32 onwards, this condition test is implemented using the
same op type that the core C<$var isa Class> syntax is provided by and works
in exactly the same way.

On older perl versions it is emulated by a compatibility function. Currently
this function does not respect a C<< ->isa >> method overload on the exception
instance. Usually this should not be a problem, as exception class types
rarely provide such a method.

=item *

   catch ($var =~ m/regexp/)

The block is invoked only if the caught exception is a string that matches
the given regexp.

=back

When an exception is caught, each condition is tested in the order they are
written in, until a matching case is found. If such a case is found the
corresponding block is invoked, and no further condition is tested. If no
contional block matched and there is a default (unconditional) block at the
end then that is invoked instead. If no such block exists, then the exception
is propagated up to the calling scope.

=head2 finally

   ...
   finally {
      STATEMENTS...
   }

A C statement provides a block of code to the preceding C
statement (or C pair) which is executed afterwards, both in the
case of a normal execution or a thrown exception. This code block may be used
to provide whatever clean-up operations might be required by preceding code.

Because it is executed during a stack cleanup operation, a C block
may not cause the containing function to return, or to alter the return value
of it. It also cannot see the containing function's C<@_> arguments array
(though as it is block scoped within the function, it will continue to share
any normal lexical variables declared up until that point). It is protected
from disturbing the value of C<$@>. If the C block code throws an
exception, this will be printed as a warning and discarded, leaving C<$@>
containing the original exception, if one existed.

=head1 OTHER MODULES

There are already quite a number of modules on CPAN that provide a
C-like syntax for Perl.

=over 2

=item *

L

=item *

L

=item *

L

=item *

L

=back

In addition, core perl itself gained a C syntax based on this
module at version 5.34.0. It is available as C.

They are compared here, by feature:

=head2 True syntax plugin

Like L and L, this module is implemented as a true
syntax plugin, allowing it to provide new parsing rules not available to
simple functions. Most notably here it means that the resulting combination
does not need to end in a semicolon.

The core C is also implemented as true native syntax in the
perl parser.

In comparison, L is plain perl and provides its functionality using
regular perl functions; as such its syntax requires the trailing semicolon.

L is a hybrid that uses L to parse the syntax tree.

=head2 C<@_> in a try or catch block

Because the C and C block code is contained in a true block rather
than an entire anonymous subroutine, invoking it does not interfere with the
C<@_> arguments array. Code inside these blocks can interact with the
containing function's array as before.

This feature is unique among these modules; none of the others listed have
this ability.

The core C also behaves in this manner.

=head2 C in a try or catch block

Like L and L, the C statement has its
usual effect within a subroutine containing syntax provided by this module.
Namely, it causes the containing C itself to return.

It also behaves this way using the core C.

In comparison, using L or L mean that a C statement
will only exit from the C block.

=head2 C/C/C in a try or catch block

The loop control keywords of C, C and C have their usual
effect on dynamically contained loops.

These also work fine when using the core C.

L documents that these do not work there. The other
modules make no statement either way.

=head2 Value Semantics

Like L and L, the syntax provided by this module
only works as a syntax-level statement and not an expression. You cannot
assign from the result of a C block. A common workaround is to wrap
the C statement inside a C block, where its final expression
can be captured and used as a value.

The same C block wrapping also works for the core C.

In comparison, the behaviour implemented by L can be used as a
valued expression, such as assigned to a variable or returned to the caller of
its containing function.

=head2 C without C

Like L, the syntax provided by this module allows a
C block to be followed by only a C block, with no C. In
this case, exceptions thrown by code contained by the C are not
suppressed, instead they propagate as normal to callers. This matches the
behaviour familiar to Java or C++ programmers.

In comparison, the code provided by L and L always suppress
exception propagation even without an actual C block.

The L module does not allow a C block not followed by C.

The core C does not implement C at all, and also
requires that every C block be followed by a C.

=head2 Typed C

L and L make no attempt to perform any kind of typed dispatch
to distinguish kinds of exception caught by C blocks.

Likewise the core C currently does not provide this ability,
though it remains an area of ongoing design work.

L and L both attempt to provide a kind of
typed dispatch where different classes of exception are caught by different
blocks of code, or propagated up entirely to callers.

This module provides such an ability, via the currently-experimental
C syntax.

The design thoughts continue on the RT ticket
L.

=cut

sub import
{
   my $pkg = shift;
   my $caller = caller;

   $pkg->import_into( $caller, @_ );
}

sub unimport
{
   my $pkg = shift;
   my $caller = caller;

   $pkg->unimport_into( $caller, @_ );
}

my @EXPERIMENTAL = qw( typed );

sub import_into   { shift->apply( sub { $^H{ $_[0] }++ },      @_ ) }
sub unimport_into { shift->apply( sub { delete $^H{ $_[0] } }, @_ ) }

sub apply
{
   my $pkg = shift;
   my ( $cb, $caller, @syms ) = @_;

   @syms or @syms = qw( try );

   my %syms = map { $_ => 1 } @syms;
   $cb->( "Syntax::Keyword::Try/try" ) if delete $syms{try};

   # Largely for Feature::Compat::Try's benefit
   $cb->( "Syntax::Keyword::Try/no_finally" )    if delete $syms{"-no_finally"};
   $cb->( "Syntax::Keyword::Try/require_catch" ) if delete $syms{"-require_catch"};
   $cb->( "Syntax::Keyword::Try/require_var" )   if delete $syms{"-require_var"};

   # stablised experiments
   delete $syms{":experimental($_)"} for qw( var );

   foreach ( @EXPERIMENTAL ) {
      $cb->( "Syntax::Keyword::Try/experimental($_)" ) if delete $syms{":experimental($_)"};
   }

   if( delete $syms{":experimental"} ) {
      $cb->( "Syntax::Keyword::Try/experimental($_)" ) for @EXPERIMENTAL;
   }

   # Ignore requests for these, as they come automatically with `try`
   delete @syms{qw( catch finally )};

   if( $syms{try_value} or $syms{":experimental(try_value)"} ) {
      croak "The 'try_value' experimental feature is now removed\n" .
            "Instead, you should use  do { try ... }  to yield a value from a try/catch statement";
   }

   croak "Unrecognised import symbols @{[ keys %syms ]}" if keys %syms;
}

=head1 WITH OTHER MODULES

=head2 Future::AsyncAwait

As of C version 0.10 and L version
0.07, cross-module integration tests assert that basic C blocks
inside an C work correctly, including those that attempt to
C from inside C.

   use Future::AsyncAwait;
   use Syntax::Keyword::Try;

   async sub attempt
   {
      try {
         await func();
         return "success";
      }
      catch {
         return "failed";
      }
   }

=head1 ISSUES

=head2 Thread-safety at load time cannot be assured before perl 5.16

On F versions 5.16 and above this module is thread-safe.

On F version 5.14 this module is thread-safe provided that it is
Cd before any additional threads are created.

However, when using 5.14 there is a race condition if this module is loaded
late in the program startup, after additional threads have been created. This
leads to the potential for it to be started up multiple times concurrently,
which creates data races when modifying internal structures and likely leads
to a segmentation fault, either during load or soon after when more code is
compiled.

As a workaround, for any such program that creates multiple threads, loads
additional code (such as dynamically-discovered plugins), and has to run on
5.14, it should make sure to

   use Syntax::Keyword::Try;

early on in startup, before it spins out any additional threads.

(See also L)

=head2 $@ is not local'ised by C before perl 5.24

On F versions 5.24 and above, or when using only control-flow statement
syntax, C<$@> is always correctly Cised.

However, when using the experimental value-yielding expression version
C on perl versions 5.22 or older, the Cisation of C<$@>
does not correctly apply around the expression. After such an expression, the
value of C<$@> will leak out if a failure happened and the C block was
invoked, overwriting any previous value that was visible there.

(See also L)

=head1 ACKNOWLEDGEMENTS

With thanks to C, C and others from C for
assisting with trickier bits of XS logic.

=head1 AUTHOR

Paul Evans

=cut

0x55AA;