NAME¶
"IO::Async::Stream" - event callbacks and write bufering for a stream
filehandle
SYNOPSIS¶
use IO::Async::Stream;
use IO::Async::Loop;
my $loop = IO::Async::Loop->new;
my $stream = IO::Async::Stream->new(
read_handle => \*STDIN,
write_handle => \*STDOUT,
on_read => sub {
my ( $self, $buffref, $eof ) = @_;
while( $$buffref =~ s/^(.*\n)// ) {
print "Received a line $1";
}
if( $eof ) {
print "EOF; last partial line is $$buffref\n";
}
return 0;
}
);
$loop->add( $stream );
$stream->write( "An initial line here\n" );
DESCRIPTION¶
This subclass of IO::Async::Handle contains a filehandle that represents a
byte-stream. It provides buffering for both incoming and outgoing data. It
invokes the "on_read" handler when new data is read from the
filehandle. Data may be written to the filehandle by calling the
"write" method.
For implementing real network protocols that are based on messages sent over a
byte-stream (such as a TCP socket), it may be more appropriate to use a
subclass of IO::Async::Protocol::Stream.
EVENTS¶
The following events are invoked, either using subclass methods or CODE
references in parameters:
$ret = on_read \$buffer, $eof¶
Invoked when more data is available in the internal receiving buffer.
The first argument is a reference to a plain perl string. The code should
inspect and remove any data it likes, but is not required to remove all, or
indeed any of the data. Any data remaining in the buffer will be preserved for
the next call, the next time more data is received from the handle.
In this way, it is easy to implement code that reads records of some form when
completed, but ignores partially-received records, until all the data is
present. If the handler is confident no more useful data remains, it should
return 0. If not, it should return 1, and the handler will be called again.
This makes it easy to implement code that handles multiple incoming records at
the same time. See the examples at the end of this documentation for more
detail.
The second argument is a scalar indicating whether the stream has reported an
end-of-file (EOF) condition. A reference to the buffer is passed to the
handler in the usual way, so it may inspect data contained in it. Once the
handler returns a false value, it will not be called again, as the handle is
now at EOF and no more data can arrive.
The "on_read" code may also dynamically replace itself with a new
callback by returning a CODE reference instead of 0 or 1. The original
callback or method that the object first started with may be restored by
returning "undef". Whenever the callback is changed in this way, the
new code is called again; even if the read buffer is currently empty. See the
examples at the end of this documentation for more detail.
on_read_eof¶
Optional. Invoked when the read handle indicates an end-of-file (EOF) condition.
If there is any data in the buffer still to be processed, the
"on_read" event will be invoked first, before this one.
on_write_eof¶
Optional. Invoked when the write handle indicates an end-of-file (EOF)
condition. Note that this condition can only be detected after a
"write" syscall returns the "EPIPE" error. If there is no
data pending to be written then it will not be detected yet.
on_read_error $errno¶
Optional. Invoked when the "sysread" method on the read handle fails.
on_write_error $errno¶
Optional. Invoked when the "syswrite" method on the write handle
fails.
The "on_read_error" and "on_write_error" handlers are passed
the value of $! at the time the error occured. (The $! variable itself, by its
nature, may have changed from the original error by the time this handler runs
so it should always use the value passed in).
If an error occurs when the corresponding error callback is not supplied, and
there is not a handler for it, then the "close" method is called
instead.
on_outgoing_empty¶
Optional. Invoked when the writing data buffer becomes empty.
PARAMETERS¶
The following named parameters may be passed to "new" or
"configure":
- read_handle => IO
- The IO handle to read from. Must implement
"fileno" and "sysread" methods.
- write_handle => IO
- The IO handle to write to. Must implement
"fileno" and "syswrite" methods.
- handle => IO
- Shortcut to specifying the same IO handle for both of the
above.
- on_read => CODE
- on_read_error => CODE
- on_outgoing_empty => CODE
- on_write_error => CODE
- CODE references for event handlers.
- autoflush => BOOL
- Optional. If true, the "write" method will
attempt to write data to the operating system immediately, without waiting
for the loop to indicate the filehandle is write-ready. This is useful,
for example, on streams that should contain up-to-date logging or console
information.
It currently defaults to false for any file handle, but future versions of
"IO::Async" may enable this by default on STDOUT and
STDERR.
- read_len => INT
- Optional. Sets the buffer size for "read" calls.
Defaults to 8 KiBytes.
- read_all => BOOL
- Optional. If true, attempt to read as much data from the
kernel as possible when the handle becomes readable. By default this is
turned off, meaning at most one fixed-size buffer is read. If there is
still more data in the kernel's buffer, the handle will still be readable,
and will be read from again.
This behaviour allows multiple streams and sockets to be multiplexed
simultaneously, meaning that a large bulk transfer on one cannot starve
other filehandles of processing time. Turning this option on may improve
bulk data transfer rate, at the risk of delaying or stalling processing on
other filehandles.
- write_len => INT
- Optional. Sets the buffer size for "write" calls.
Defaults to 8 KiBytes.
- write_all => BOOL
- Optional. Analogous to the "read_all" option, but
for writing. When "autoflush" is enabled, this option only
affects deferred writing if the initial attempt failed due to buffer
space.
- close_on_read_eof => BOOL
- Optional. Usually true, but if set to a false value then
the stream will not be "close"d when an EOF condition occurs on
read. This is normally not useful as at that point the underlying stream
filehandle is no longer useable, but it may be useful for reading regular
files, or interacting with TTY devices.
- encoding => STRING
- If supplied, sets the name of encoding of the underlying
stream. If an encoding is set, then the "print" method will
expect to receive Unicode strings and encodes them into bytes, and
incoming bytes will be decoded into Unicode strings for the
"on_read" event.
If an encoding is not supplied then "print" and
"on_read" will work in byte strings.
IMPORTANT NOTE: in order to handle reads of UTF-8 content or other
multibyte encodings, the code implementing the "on_read" event
uses a feature of Encode; the "STOP_AT_PARTIAL" flag. While this
flag has existed for a while and is used by the ":encoding"
PerlIO layer itself for similar purposes, the flag is not officially
documented by the "Encode" module. In principle this
undocumented feature could be subject to change, in practice I believe it
to be reasonably stable.
This note applies only to the "on_read" event; data written using
the "write" method does not rely on any undocumented features of
"Encode".
If a read handle is given, it is required that either an "on_read"
callback reference is configured, or that the object provides an
"on_read" method. It is optional whether either is true for
"on_outgoing_empty"; if neither is supplied then no action will be
taken when the writing buffer becomes empty.
An "on_read" handler may be supplied even if no read handle is yet
given, to be used when a read handle is eventually provided by the
"set_handles" method.
This condition is checked at the time the object is added to a Loop; it is
allowed to create a "IO::Async::Stream" object with a read handle
but without a "on_read" handler, provided that one is later given
using "configure" before the stream is added to its containing Loop,
either directly or by being a child of another Notifier already in a Loop, or
added to one.
METHODS¶
$stream->close¶
A synonym for "close_when_empty". This should not be used when the
deferred wait behaviour is required, as the behaviour of "close" may
change in a future version of "IO::Async". Instead, call
"close_when_empty" directly.
$stream->close_when_empty¶
If the write buffer is empty, this method calls "close" on the
underlying IO handles, and removes the stream from its containing loop. If the
write buffer still contains data, then this is deferred until the buffer is
empty. This is intended for "write-then-close" one-shot streams.
$stream->write( "Here is my final data\n" );
$stream->close_when_empty;
Because of this deferred nature, it may not be suitable for error handling. See
instead the "close_now" method.
$stream->close_now¶
This method immediately closes the underlying IO handles and removes the stream
from the containing loop. It will not wait to flush the remaining data in the
write buffer.
$stream->write( $data, %params )¶
This method adds data to the outgoing data queue, or writes it immediately,
according to the "autoflush" parameter.
If the "autoflush" option is set, this method will try immediately to
write the data to the underlying filehandle. If this completes successfully
then it will have been written by the time this method returns. If it fails to
write completely, then the data is queued as if "autoflush" were not
set, and will be flushed as normal.
$data can either be a plain string, or a CODE reference. If it is a CODE
reference, it will be invoked to generate data to be written. Each time the
filehandle is ready to receive more data to it, the function is invoked, and
what it returns written to the filehandle. Once the function has finished
generating data it should return undef. The function is passed the Stream
object as its first argument.
For example, to stream the contents of an existing opened filehandle:
open my $fileh, "<", $path or die "Cannot open $path - $!";
$stream->write( sub {
my ( $stream ) = @_;
sysread $fileh, my $buffer, 8192 or return;
return $buffer;
} );
Takes the following optional named parameters in %params:
- on_flush => CODE
- A CODE reference which will be invoked once the data queued
by this "write" call has been flushed. This will be invoked even
if the buffer itself is not yet empty; if more data has been queued since
the call.
$on_flush->( $stream )
If the object is not yet a member of a loop and doesn't yet have a
"write_handle", then calls to the "write" method will
simply queue the data and return. It will be flushed when the object is added
to the loop.
If $data is a defined but empty string, the write is still queued, and the
"on_flush" continuation will be invoked, if supplied. This can be
used to obtain a marker, to invoke some code once the output queue has been
flushed up to this point.
UTILITY CONSTRUCTORS¶
$stream = IO::Async::Stream->new_for_stdin¶
$stream = IO::Async::Stream->new_for_stdout¶
$stream = IO::Async::Stream->new_for_stdio¶
Return a "IO::Async::Stream" object preconfigured with the correct
"read_handle", "write_handle" or both.
EXAMPLES¶
A line-based "on_read" method¶
The following "on_read" method accepts incoming
"\n"-terminated lines and prints them to the program's
"STDOUT" stream.
sub on_read
{
my $self = shift;
my ( $buffref, $eof ) = @_;
while( $$buffref =~ s/^(.*\n)// ) {
print "Received a line: $1";
}
return 0;
}
Because a reference to the buffer itself is passed, it is simple to use a
"s///" regular expression on the scalar it points at, to both check
if data is ready (i.e. a whole line), and to remove it from the buffer. If no
data is available then 0 is returned, to indicate it should not be tried
again. If a line was successfully extracted, then 1 is returned, to indicate
it should try again in case more lines exist in the buffer.
For implementing real network protocols that are based on lines of text it may
be more appropriate to use a subclass of IO::Async::Protocol::LineStream.
Reading binary data¶
This "on_read" method accepts incoming records in 16-byte chunks,
printing each one.
sub on_read
{
my ( $self, $buffref, $eof ) = @_;
if( length $$buffref >= 16 ) {
my $record = substr( $$buffref, 0, 16, "" );
print "Received a 16-byte record: $record\n";
return 1;
}
if( $eof and length $$buffref ) {
print "EOF: a partial record still exists\n";
}
return 0;
}
The 4-argument form of "substr()" extracts the 16-byte record from the
buffer and assigns it to the $record variable, if there was enough data in the
buffer to extract it.
A lot of protocols use a fixed-size header, followed by a variable-sized body of
data, whose size is given by one of the fields of the header. The following
"on_read" method extracts messages in such a protocol.
sub on_read
{
my ( $self, $buffref, $eof ) = @_;
return 0 unless length $$buffref >= 8; # "N n n" consumes 8 bytes
my ( $len, $x, $y ) = unpack $$buffref, "N n n";
return 0 unless length $$buffref >= 8 + $len;
substr( $$buffref, 0, 8, "" );
my $data = substr( $$buffref, 0, $len, "" );
print "A record with values x=$x y=$y\n";
return 1;
}
In this example, the header is "unpack()"ed first, to extract the body
length, and then the body is extracted. If the buffer does not have enough
data yet for a complete message then 0 is returned, and the buffer is left
unmodified for next time. Only when there are enough bytes in total does it
use "substr()" to remove them.
Dynamic replacement of "on_read"¶
Consider the following protocol (inspired by IMAP), which consists of
"\n"-terminated lines that may have an optional data block attached.
The presence of such a data block, as well as its size, is indicated by the
line prefix.
sub on_read
{
my $self = shift;
my ( $buffref, $eof ) = @_;
if( $$buffref =~ s/^DATA (\d+):(.*)\n// ) {
my $length = $1;
my $line = $2;
return sub {
my $self = shift;
my ( $buffref, $eof ) = @_;
return 0 unless length $$buffref >= $length;
# Take and remove the data from the buffer
my $data = substr( $$buffref, 0, $length, "" );
print "Received a line $line with some data ($data)\n";
return undef; # Restore the original method
}
}
elsif( $$buffref =~ s/^LINE:(.*)\n// ) {
my $line = $1;
print "Received a line $line with no data\n";
return 1;
}
else {
print STDERR "Unrecognised input\n";
# Handle it somehow
}
}
In the case where trailing data is supplied, a new temporary "on_read"
callback is provided in a closure. This closure captures the $length variable
so it knows how much data to expect. It also captures the $line variable so it
can use it in the event report. When this method has finished reading the
data, it reports the event, then restores the original method by returning
"undef".
SEE ALSO¶
- •
- IO::Handle - Supply object methods for I/O handles
AUTHOR¶
Paul Evans <leonerd@leonerd.org.uk>