————————————————————————# 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, 2006-2024 -- leonerd@leonerd.org.uk
package
IO::Async::Notifier 0.803;
use
v5.14;
use
warnings;
use
Carp;
use
IO::Async::Debug;
# Perl 5.8.4 cannot do trampolines by modiying @_ then goto &$code
=head1 NAME
C<IO::Async::Notifier> - base class for L<IO::Async> event objects
=head1 SYNOPSIS
Usually not directly used by a program, but one valid use case may be:
use IO::Async::Notifier;
use IO::Async::Stream;
use IO::Async::Signal;
use IO::Async::Loop;
my $loop = IO::Async::Loop->new;
my $notifier = IO::Async::Notifier->new;
$notifier->add_child(
IO::Async::Stream->new_for_stdin(
on_read => sub {
my $self = shift;
my ( $buffref, $eof ) = @_;
while( $$buffref =~ s/^(.*)\n// ) {
print "You said $1\n";
}
return 0;
},
)
);
$notifier->add_child(
IO::Async::Signal->new(
name => 'INT',
on_receipt => sub {
print "Goodbye!\n";
$loop->stop;
},
)
);
$loop->add( $notifier );
$loop->run;
=head1 DESCRIPTION
This object class forms the basis for all the other event objects that an
L<IO::Async> program uses. It provides the lowest level of integration with a
L<IO::Async::Loop> container, and a facility to collect Notifiers together, in
a tree structure, where any Notifier can contain a collection of children.
Normally, objects in this class would not be directly used by an end program,
as it performs no actual IO work, and generates no actual events. These are all
left to the various subclasses, such as:
=over 4
=item *
L<IO::Async::Handle> - event callbacks for a non-blocking file descriptor
=item *
L<IO::Async::Stream> - event callbacks and write bufering for a stream
filehandle
=item *
L<IO::Async::Socket> - event callbacks and send buffering for a socket
filehandle
=item *
L<IO::Async::Timer> - base class for Notifiers that use timed delays
=item *
L<IO::Async::Signal> - event callback on receipt of a POSIX signal
=item *
L<IO::Async::PID> - event callback on exit of a child process
=item *
L<IO::Async::Process> - start and manage a child process
=back
For more detail, see the SYNOPSIS section in one of the above.
One case where this object class would be used, is when a library wishes to
provide a sub-component which consists of multiple other C<Notifier>
subclasses, such as C<Handle>s and C<Timers>, but no particular object is
suitable to be the root of a tree. In this case, a plain C<Notifier> object
can be used as the tree root, and all the other notifiers added as children of
it.
=cut
=head1 AS A MIXIN
Rather than being used as a subclass this package also supports being used as
a non-principle superclass for an object, as a mix-in. It still provides
methods and satisfies an C<isa> test, even though the constructor is not
directly called. This simply requires that the object be based on a normal
blessed hash reference and include C<IO::Async::Notifier> somewhere in its
C<@ISA> list.
The methods in this class all use only keys in the hash prefixed by
C<"IO_Async_Notifier__"> for namespace purposes.
This is intended mainly for defining a subclass of some other object that is
also an C<IO::Async::Notifier>, suitable to be added to an L<IO::Async::Loop>.
package SomeEventSource::Async;
use base qw( SomeEventSource IO::Async::Notifier );
sub _add_to_loop
{
my $self = shift;
my ( $loop ) = @_;
# Code here to set up event handling on $loop that may be required
}
sub _remove_from_loop
{
my $self = shift;
my ( $loop ) = @_;
# Code here to undo the event handling set up above
}
Since all the methods documented here will be available, the implementation
may wish to use the C<configure> and C<make_event_cb> or C<invoke_event>
methods to implement its own event callbacks.
=cut
=head1 EVENTS
The following events are invoked, either using subclass methods or CODE
references in parameters:
=head2 on_error $message, $name, @details
Invoked by C<invoke_error>.
=cut
=head1 PARAMETERS
A specific subclass of C<IO::Async::Notifier> defines named parameters that
control its behaviour. These may be passed to the C<new> constructor, or to
the C<configure> method. The documentation on each specific subclass will give
details on the parameters that exist, and their uses. Some parameters may only
support being set once at construction time, or only support being changed if
the object is in a particular state.
The following parameters are supported by all Notifiers:
=over 8
=item on_error => CODE
CODE reference for event handler.
=item notifier_name => STRING
Optional string used to identify this particular Notifier. This value will be
returned by the C<notifier_name> method.
=back
=cut
=head1 CONSTRUCTOR
=cut
=head2 new
$notifier = IO::Async::Notifier->new( %params );
This function returns a new instance of a C<IO::Async::Notifier> object with
the given initial values of the named parameters.
Up until L<IO::Async> version 0.19, this module used to implement the IO
handle features now found in the L<IO::Async::Handle> subclass. Code that
needs to use any of C<handle>, C<read_handle>, C<write_handle>,
C<on_read_ready> or C<on_write_ready> should use L<IO::Async::Handle> instead.
=cut
sub
new
{
my
$class
=
shift
;
my
%params
=
@_
;
my
$self
=
bless
{},
$class
;
$self
->_init( \
%params
);
$self
->configure(
%params
);
return
$self
;
}
=head1 METHODS
=cut
=head2 configure
$notifier->configure( %params );
Adjust the named parameters of the C<Notifier> as given by the C<%params>
hash.
=cut
# for subclasses to override and call down to
sub
configure
{
my
$self
=
shift
;
my
%params
=
@_
;
foreach
(
qw( notifier_name on_error )
) {
$self
->{
"IO_Async_Notifier__$_"
} =
delete
$params
{
$_
}
if
exists
$params
{
$_
};
}
$self
->configure_unknown(
%params
)
if
keys
%params
;
}
sub
configure_unknown
{
my
$self
=
shift
;
my
%params
=
@_
;
my
$class
=
ref
$self
;
croak
"Unrecognised configuration keys for $class - "
.
join
(
" "
,
keys
%params
);
}
=head2 loop
$loop = $notifier->loop;
Returns the L<IO::Async::Loop> that this Notifier is a member of.
=cut
sub
loop
{
my
$self
=
shift
;
return
$self
->{IO_Async_Notifier__loop}
}
*get_loop
= \
&loop
;
# Only called by IO::Async::Loop, not external interface
sub
__set_loop
{
my
$self
=
shift
;
my
(
$loop
) =
@_
;
# early exit if no change
return
if
!
$loop
and !
$self
->loop or
$loop
and
$self
->loop and
$loop
==
$self
->loop;
$self
->_remove_from_loop(
$self
->loop )
if
$self
->loop;
$self
->{IO_Async_Notifier__loop} =
$loop
;
weaken(
$self
->{IO_Async_Notifier__loop} );
# To avoid a cycle
$self
->_add_to_loop(
$self
->loop )
if
$self
->loop;
}
=head2 notifier_name
$name = $notifier->notifier_name;
Returns the name to identify this Notifier. If a has not been set, it will
return the empty string. Subclasses may wish to override this behaviour to
return some more useful information, perhaps from configured parameters.
=cut
sub
notifier_name
{
my
$self
=
shift
;
return
$self
->{IO_Async_Notifier__notifier_name} ||
""
;
}
=head2 adopt_future
$f = $notifier->adopt_future( $f );
Stores a reference to the L<Future> instance within the notifier itself, so
the reference doesn't get lost. This reference will be dropped when the future
becomes ready (either by success or failure). Additionally, if the future
failed the notifier's C<invoke_error> method will be informed.
This means that if the notifier does not provide an C<on_error> handler, nor
is there one anywhere in the parent chain, this will be fatal to the caller of
C<< $f->fail >>. To avoid this being fatal if the failure is handled
elsewhere, use the C<else_done> method on the future to obtain a sequence one
that never fails.
$notifier->adopt_future( $f->else_done() );
The future itself is returned.
=cut
sub
adopt_future
{
my
$self
=
shift
;
my
(
$f
) =
@_
;
my
$fkey
=
"$f"
;
# stable stringification
$self
->{IO_Async_Notifier__futures}{
$fkey
} =
$f
;
$f
->on_ready(
$self
->_capture_weakself(
sub
{
my
$self
=
shift
;
my
(
$f
) =
@_
;
delete
$self
->{IO_Async_Notifier__futures}{
$fkey
};
$self
->invoke_error(
$f
->failure )
if
$f
->is_failed;
}));
return
$f
;
}
=head2 adopted_futures
@f = $notifier->adopted_futures;
I<Since version 0.73.>
Returns a list of all the adopted and still-pending futures, in no particular
order.
=cut
sub
adopted_futures
{
my
$self
=
shift
;
return
values
%{
$self
->{IO_Async_Notifier__futures} };
}
=head1 CHILD NOTIFIERS
During the execution of a program, it may be the case that certain IO handles
cause other handles to be created; for example, new sockets that have been
C<accept()>ed from a listening socket. To facilitate these, a notifier may
contain child notifier objects, that are automatically added to or removed
from the L<IO::Async::Loop> that manages their parent.
=cut
=head2 parent
$parent = $notifier->parent;
Returns the parent of the notifier, or C<undef> if does not have one.
=cut
sub
parent
{
my
$self
=
shift
;
return
$self
->{IO_Async_Notifier__parent};
}
=head2 children
@children = $notifier->children;
Returns a list of the child notifiers contained within this one.
=cut
sub
children
{
my
$self
=
shift
;
return
unless
$self
->{IO_Async_Notifier__children};
return
@{
$self
->{IO_Async_Notifier__children} };
}
=head2 add_child
$notifier->add_child( $child );
Adds a child notifier. This notifier will be added to the containing loop, if
the parent has one. Only a notifier that does not currently have a parent and
is not currently a member of any loop may be added as a child. If the child
itself has grandchildren, these will be recursively added to the containing
loop.
=cut
sub
add_child
{
my
$self
=
shift
;
my
(
$child
) =
@_
;
croak
"Cannot add a child that already has a parent"
if
defined
$child
->{IO_Async_Notifier__parent};
croak
"Cannot add a child that is already a member of a loop"
if
defined
$child
->loop;
if
(
defined
(
my
$loop
=
$self
->loop ) ) {
$loop
->add(
$child
);
}
push
@{
$self
->{IO_Async_Notifier__children} },
$child
;
$child
->{IO_Async_Notifier__parent} =
$self
;
weaken(
$child
->{IO_Async_Notifier__parent} );
return
;
}
=head2 remove_child
$notifier->remove_child( $child );
Removes a child notifier. The child will be removed from the containing loop,
if the parent has one. If the child itself has grandchildren, these will be
recurively removed from the loop.
=cut
sub
remove_child
{
my
$self
=
shift
;
my
(
$child
) =
@_
;
LOOP: {
my
$childrenref
=
$self
->{IO_Async_Notifier__children};
for
my
$i
( 0 ..
$#$childrenref
) {
no
warnings
'uninitialized'
;
next
unless
$childrenref
->[
$i
] ==
$child
;
splice
@$childrenref
,
$i
, 1, ();
last
LOOP;
}
croak
"Cannot remove child from a parent that doesn't contain it"
;
}
undef
$child
->{IO_Async_Notifier__parent};
if
(
defined
(
my
$loop
=
$self
->loop ) ) {
$loop
->remove(
$child
);
}
}
=head2 remove_from_parent
$notifier->remove_from_parent;
Removes this notifier object from its parent (either another notifier object
or the containing loop) if it has one. If the notifier is not a child of
another notifier nor a member of a loop, this method does nothing.
=cut
sub
remove_from_parent
{
my
$self
=
shift
;
if
(
my
$parent
=
$self
->parent ) {
$parent
->remove_child(
$self
);
}
elsif
(
my
$loop
=
$self
->loop ) {
$loop
->remove(
$self
);
}
}
=head1 SUBCLASS METHODS
C<IO::Async::Notifier> is a base class provided so that specific subclasses of
it provide more specific behaviour. The base class provides a number of
methods that subclasses may wish to override.
If a subclass implements any of these, be sure to invoke the superclass method
at some point within the code.
=cut
=head2 _init
$notifier->_init( $paramsref );
This method is called by the constructor just before calling C<configure>.
It is passed a reference to the HASH storing the constructor arguments.
This method may initialise internal details of the Notifier as required,
possibly by using parameters from the HASH. If any parameters are
construction-only they should be C<delete>d from the hash.
=cut
sub
_init
{
# empty default
}
=head2 configure
$notifier->configure( %params );
This method is called by the constructor to set the initial values of named
parameters, and by users of the object to adjust the values once constructed.
This method should C<delete> from the C<%params> hash any keys it has dealt
with, then pass the remaining ones to the C<SUPER::configure>. The base
class implementation will throw an exception if there are any unrecognised
keys remaining.
=cut
=head2 configure_unknown
$notifier->configure_unknown( %params );
This method is called by the base class C<configure> method, for any remaining
parameters that are not recognised. The default implementation throws an
exception using C<Carp> that lists the unrecognised keys. This method is
provided to allow subclasses to override the behaviour, perhaps to store
unrecognised keys, or to otherwise inspect the left-over arguments for some
other purpose.
=cut
=head2 _add_to_loop
$notifier->_add_to_loop( $loop );
This method is called when the Notifier has been added to a Loop; either
directly, or indirectly through being a child of a Notifer already in a loop.
This method may be used to perform any initial startup activity required for
the Notifier to be fully functional but which requires a Loop to do so.
=cut
sub
_add_to_loop
{
# empty default
}
=head2 _remove_from_loop
$notifier->_remove_from_loop( $loop );
This method is called when the Notifier has been removed from a Loop; either
directly, or indirectly through being a child of a Notifier removed from the
loop.
This method may be used to undo the effects of any setup that the
C<_add_to_loop> method had originally done.
=cut
sub
_remove_from_loop
{
# empty default
}
=head1 UTILITY METHODS
=cut
=head2 _capture_weakself
$mref = $notifier->_capture_weakself( $code );
Returns a new CODE ref which, when invoked, will invoke the originally-passed
ref, with additionally a reference to the Notifier as its first argument. The
Notifier reference is stored weakly in C<$mref>, so this CODE ref may be
stored in the Notifier itself without creating a cycle.
For example,
my $mref = $notifier->_capture_weakself( sub {
my ( $notifier, $arg ) = @_;
print "Notifier $notifier got argument $arg\n";
} );
$mref->( 123 );
This is provided as a utility for Notifier subclasses to use to build a
callback CODEref to pass to a Loop method, but which may also want to store
the CODE ref internally for efficiency.
The C<$code> argument may also be a plain string, which will be used as a
method name; the returned CODE ref will then invoke that method on the object.
In this case the method name is stored symbolically in the returned CODE
reference, and dynamically dispatched each time the reference is invoked. This
allows it to follow code reloading, dynamic replacement of class methods, or
other similar techniques.
If the C<$mref> CODE reference is being stored in some object other than the
one it refers to, remember that since the Notifier is only weakly captured, it
is possible that it has been destroyed by the time the code runs, and so the
reference will be passed as C<undef>. This should be protected against by the
code body.
$other_object->{on_event} = $notifier->_capture_weakself( sub {
my $notifier = shift or return;
my ( @event_args ) = @_;
...
} );
For stand-alone generic implementation of this behaviour, see also L<curry>
and C<curry::weak>.
=cut
sub
_capture_weakself
{
my
$self
=
shift
;
my
(
$code
) =
@_
;
# actually bare method names work too
if
( !
ref
$code
) {
my
$class
=
ref
$self
;
# Don't save this coderef, or it will break dynamic method dispatch,
# which means code reloading, dynamic replacement, or other funky
# techniques stop working
$self
->can(
$code
) or
croak
qq(Can't locate object method "$code" via package "$class")
;
}
weaken
$self
;
return
sub
{
my
$cv
=
ref
(
$code
) ?
$code
:
$self
->can(
$code
);
if
( HAS_BROKEN_TRAMPOLINES ) {
return
$cv
->(
$self
,
@_
);
}
else
{
unshift
@_
,
$self
;
goto
&$cv
;
}
};
}
=head2 _replace_weakself
$mref = $notifier->_replace_weakself( $code );
Returns a new CODE ref which, when invoked, will invoke the originally-passed
ref, with a reference to the Notifier replacing its first argument. The
Notifier reference is stored weakly in C<$mref>, so this CODE ref may be
stored in the Notifier itself without creating a cycle.
For example,
my $mref = $notifier->_replace_weakself( sub {
my ( $notifier, $arg ) = @_;
print "Notifier $notifier got argument $arg\n";
} );
$mref->( $object, 123 );
This is provided as a utility for Notifier subclasses to use for event
callbacks on other objects, where the delegated object is passed in the
function's arguments.
The C<$code> argument may also be a plain string, which will be used as a
method name; the returned CODE ref will then invoke that method on the object.
As with C<_capture_weakself> this is stored symbolically.
As with C<_capture_weakself>, care should be taken against Notifier
destruction if the C<$mref> CODE reference is stored in some other object.
=cut
sub
_replace_weakself
{
my
$self
=
shift
;
my
(
$code
) =
@_
;
# actually bare method names work too
if
( !
ref
$code
) {
# Don't save this coderef, see _capture_weakself for why
my
$class
=
ref
$self
;
$self
->can(
$code
) or
croak
qq(Can't locate object method "$code" via package "$class")
;
}
weaken
$self
;
return
sub
{
my
$cv
=
ref
(
$code
) ?
$code
:
$self
->can(
$code
);
if
( HAS_BROKEN_TRAMPOLINES ) {
return
$cv
->(
$self
,
@_
[1..
$#_
] );
}
else
{
# Don't assign to $_[0] directly or we will change caller's first argument
shift
@_
;
unshift
@_
,
$self
;
goto
&$cv
;
}
};
}
=head2 can_event
$code = $notifier->can_event( $event_name );
Returns a C<CODE> reference if the object can perform the given event name,
either by a configured C<CODE> reference parameter, or by implementing a
method. If the object is unable to handle this event, C<undef> is returned.
=cut
sub
can_event
{
my
$self
=
shift
;
my
(
$event_name
) =
@_
;
return
$self
->{
$event_name
} ||
$self
->can(
$event_name
);
}
=head2 make_event_cb
$callback = $notifier->make_event_cb( $event_name );
Returns a C<CODE> reference which, when invoked, will execute the given event
handler. Event handlers may either be subclass methods, or parameters given to
the C<new> or C<configure> method.
The event handler can be passed extra arguments by giving them to the C<CODE>
reference; the first parameter received will be a reference to the notifier
itself. This is stored weakly in the closure, so it is safe to store the
resulting C<CODE> reference in the object itself without causing a reference
cycle.
=cut
sub
make_event_cb
{
my
$self
=
shift
;
my
(
$event_name
) =
@_
;
my
$code
=
$self
->can_event(
$event_name
)
or croak
"$self cannot handle $event_name event"
;
my
$caller
=
caller
;
return
$self
->_capture_weakself(
!
$IO::Async::Debug::DEBUG
?
$code
:
sub
{
my
$self
=
$_
[0];
$self
->_debug_printf_event(
$caller
,
$event_name
);
goto
&$code
;
}
);
}
=head2 maybe_make_event_cb
$callback = $notifier->maybe_make_event_cb( $event_name );
Similar to C<make_event_cb> but will return C<undef> if the object cannot
handle the named event, rather than throwing an exception.
=cut
sub
maybe_make_event_cb
{
my
$self
=
shift
;
my
(
$event_name
) =
@_
;
my
$code
=
$self
->can_event(
$event_name
)
or
return
undef
;
my
$caller
=
caller
;
return
$self
->_capture_weakself(
!
$IO::Async::Debug::DEBUG
?
$code
:
sub
{
my
$self
=
$_
[0];
$self
->_debug_printf_event(
$caller
,
$event_name
);
goto
&$code
;
}
);
}
=head2 invoke_event
@ret = $notifier->invoke_event( $event_name, @args );
Invokes the given event handler, passing in the given arguments. Event
handlers may either be subclass methods, or parameters given to the C<new> or
C<configure> method. Returns whatever the underlying method or CODE reference
returned.
=cut
sub
invoke_event
{
my
$self
=
shift
;
my
(
$event_name
,
@args
) =
@_
;
my
$code
=
$self
->can_event(
$event_name
)
or croak
"$self cannot handle $event_name event"
;
$self
->_debug_printf_event(
scalar
caller
,
$event_name
)
if
$IO::Async::Debug::DEBUG
;
return
$code
->(
$self
,
@args
);
}
=head2 maybe_invoke_event
$retref = $notifier->maybe_invoke_event( $event_name, @args );
Similar to C<invoke_event> but will return C<undef> if the object cannot
handle the name event, rather than throwing an exception. In order to
distinguish this from an event-handling function that simply returned
C<undef>, if the object does handle the event, the list that it returns will
be returned in an ARRAY reference.
=cut
sub
maybe_invoke_event
{
my
$self
=
shift
;
my
(
$event_name
,
@args
) =
@_
;
my
$code
=
$self
->can_event(
$event_name
)
or
return
undef
;
$self
->_debug_printf_event(
scalar
caller
,
$event_name
)
if
$IO::Async::Debug::DEBUG
;
return
[
$code
->(
$self
,
@args
) ];
}
=head1 DEBUGGING SUPPORT
=cut
=head2 debug_printf
$notifier->debug_printf( $format, @args );
Conditionally print a debugging message to C<STDERR> if debugging is enabled.
If such a message is printed, it will be printed using C<printf> using the
given format and arguments. The message will be prefixed with a string, in
square brackets, to help identify the C<$notifier> instance. This string will
be the class name of the notifier, and any parent notifiers it is contained
by, joined by an arrow C<< <- >>. To ensure this string does not grow too
long, certain prefixes are abbreviated:
IO::Async::Protocol:: => IaP:
IO::Async:: => Ia:
Net::Async:: => Na:
Finally, each notifier that has a name defined using the C<notifier_name>
parameter has that name appended in braces.
For example, invoking
$stream->debug_printf( "EVENT on_read" );
On an L<IO::Async::Stream> instance reading and writing a file descriptor
whose C<fileno> is 4, which is a child of an L<IO::Async::Protocol::Stream>,
will produce a line of output:
[Ia:Stream{rw=4}<-IaP:Stream] EVENT on_read
=cut
sub
debug_printf
{
$IO::Async::Debug::DEBUG
or
return
;
my
$self
=
shift
;
my
(
$format
,
@args
) =
@_
;
my
@id
;
while
(
$self
) {
push
@id
,
ref
$self
;
my
$name
=
$self
->notifier_name;
$id
[-1] .=
"{$name}"
if
defined
$name
and
length
$name
;
$self
=
$self
->parent;
}
s/^IO::Async::Protocol::/IaP:/,
s/^IO::Async::/Ia:/,
s/^Net::Async::/Na:/
for
@id
;
IO::Async::Debug::logf
"[%s] $format\n"
,
join
(
"<-"
,
@id
),
@args
;
}
sub
_debug_printf_event
{
my
$self
=
shift
;
my
(
$caller
,
$event_name
) =
@_
;
my
$class
=
ref
$self
;
if
(
$IO::Async::Debug::DEBUG
> 1 or
$class
eq
$caller
) {
s/^IO::Async::Protocol::/IaP:/,
s/^IO::Async::/Ia:/,
s/^Net::Async::/Na:/
for
my
$str_caller
=
$caller
;
$self
->debug_printf(
"EVENT %s"
,
(
$class
eq
$caller
?
$event_name
:
"${str_caller}::$event_name"
)
);
}
}
=head2 invoke_error
$notifier->invoke_error( $message, $name, @details );
Invokes the stored C<on_error> event handler, passing in the given arguments.
If no handler is defined, it will be passed up to the containing parent
notifier, if one exists. If no parent exists, the error message will be thrown
as an exception by using C<die()> and this method will not return.
If a handler is found to handle this error, the method will return as normal.
However, as the expected use-case is to handle "fatal" errors that now render
the notifier unsuitable to continue, code should be careful not to perform any
further work after invoking it. Specifically, sockets may become disconnected,
or the entire notifier may now be removed from its containing loop.
The C<$name> and C<@details> list should follow similar semantics to L<Future>
failures. That is, the C<$name> should be a string giving a category of
failure, and the C<@details> list should contain additional arguments that
relate to that kind of failure.
=cut
sub
invoke_error
{
my
$self
=
shift
;
my
(
$message
,
$name
,
@details
) =
@_
;
if
(
my
$code
=
$self
->{IO_Async_Notifier__on_error} ||
$self
->can(
"on_error"
) ) {
return
$code
->(
$self
,
$message
,
$name
,
@details
);
}
if
(
my
$parent
=
$self
->parent ) {
return
$parent
->invoke_error(
@_
);
}
die
"$message\n"
;
}
=head1 AUTHOR
Paul Evans <leonerd@leonerd.org.uk>
=cut
0x55AA;