IO::Stream - ease non-blocking I/O streams based on EV


This document describes IO::Stream version v2.0.3


    use EV;
    use IO::Stream;

        host        => '',
        port        => 80,
        cb          => \&client,
        wait_for    => SENT|EOF,
        in_buf_limit=> 102400,
        out_buf     => "GET / HTTP/1.0\nHost:\n\n",

    $EV::DIED = sub { warn $@; EV::unloop };

    sub client {
        my ($io, $e, $err) = @_;
        if ($err) {
            die $err;
        if ($e & SENT) {
            print "request sent, waiting for reply...\n";
        if ($e & EOF) {
            print "server reply:\n", $io->{in_buf};
            EV::unloop;         # ALL DONE


Non-blocking event-based low-level I/O is hard to get right. Code usually error-prone and complex... and it very similar in all applications. Things become much worse when you need to alter I/O stream in some way - use proxies, encryption, SSL, etc.

This module designed to give user ability to work with I/O streams on higher level, using input/output buffers (just scalars) and high-level events like CONNECTED, SENT or EOF. As same time it doesn't hide low-level things, and user still able to work on low-level without any limitations.


Architecture of this module make it ease to write plugins, which will alter I/O stream in any way - route it through proxies, encrypt, log, etc.

Here are few available plugins, you may find more on CPAN: IO::Stream::Crypt::RC4, IO::Stream::Proxy::HTTPS, IO::Stream::MatrixSSL::Client, IO::Stream::MatrixSSL::Server.

If you interested in writing own plugin, check source for "skeleton" plugins: IO::Stream::Noop and IO::Stream::NoopAlias.


This modules doesn't export any functions/methods/variables, but it exports a lot of constants. There two groups of constants: events and errors (which can be imported using tags ':Event' and ':Error'). By default all constants are exported.





Errors are similar to $! - they're dualvars, having both textual and numeric values.

NOTE: Since v2.0.0 ETORESOLVE, EDNSNXDOMAIN and EDNSNODATA are not used anymore (EDNS is used instead), but they're still exported for compatibility.


You can create IO::Stream object using any "stream" fh (file, TTY, UNIX socket, TCP socket, pipe, FIFO). Or, if you need TCP socket, you can create IO::Stream object using host+port instead of fh (in this case IO::Stream will do non-blocking host resolving, create TCP socket and do non-blocking connect).

After you created IO::Stream object, it will handle read/write on this fh, and deliver only high-level events you asked for into your callback, where you will be able to operate with in/out buffers instead of doing sysread()/syswrite() manually.

There no limitations on what you can do with fh after you've created IO::Stream object - you can even do sysread()/syswrite() (but there no reasons for you to do this anymore).

IMPORTANT! When you want to close this fh, you MUST use $io->close() method for closing fh instead of doing close($fh). This is because IO::Stream doesn't require from you to keep object returned by new(), and without call to $io->close() IO::Stream object will continue to exists and may receive/generate some events, which is not what you expect after closing fh. Also, if you keep object returned by IO::Stream->new() somewhere in your variables, you should either undef all such variables after you called $io->close(), or you should use Scalar::Util::weaken() on these variables after storing IO::Stream object. (The same is applicable for all plugin objects too.)



If you created IO::Stream object using {host}+{port} instead of {fh}, this event will be generated after resolving {host}. Resolved IP address will be stored in {ip}.


If you created IO::Stream object using {host}+{port} instead of {fh}, this event will be generated after connecting socket to {ip}:{port}.


Generated after each successful read. IO::Stream may execute several sysread() at once before generating IN event for optimization. Read data will be stored in {in_buf}, and {in_bytes} counter will be incremented by amount of bytes read.


Generated only ONCE when EOF reached (sysread() return 0). Also will set {is_eof} to true.


Generated when some data from {out_buf} was written. Written bytes either removed from {out_buf} or just increment {out_pos} by amount of bytes written (see documentation about these fields below for more details). Also increment {out_bytes} counter by amount of bytes written.

Here 'written' may be somewhat virtual, while {out_buf}/{out_pos} changes, the real data still can be in plugin buffers (if you use plugins) and real syswrite() may not be called yet. To detect when all data is really written you should use SENT event, not OUT.


Generated when all data from {out_buf} was written. It's usual and safe to call $io->close() on SENT event.


IO::Stream has 30-second timeouts for connect and write, to timeout DNS resolve it use default AnyEvent::DNS timeout. If you need to timeout other operations, you have to create own timers using EV::timer().

Current version doesn't allow you to change these timeouts.


If you need to run TCP/UNIX-server socket, then you should handle that socket manually. But you can create IO::Stream object for accept()'ed socket:

    my ($host, $port) = ('', 1234);
    socket  my $srv_sock, AF_INET, SOCK_STREAM, 0;
    setsockopt $srv_sock, SOL_SOCKET, SO_REUSEADDR, 1;
    bind       $srv_sock, sockaddr_in($port, inet_aton($host));
    listen     $srv_sock, SOMAXCONN;
    fcntl      $srv_sock, F_SETFL, O_NONBLOCK;
    $srv_w = EV::io($srv_sock, EV::READ, sub {
        if (accept my $sock, $srv_sock) {
                fh          => $sock,
                cb          => \&server,
                wait_for    => IN,
        elsif ($! != EAGAIN) {
            die "accept: $!";


IO::Stream provide only three public methods: new(), write() and close(). new() will create new object, close() will destroy it and write() must be called when you want to modify (or just modified) output buffer.

All other operations are done using IO::Stream object fields - for simplicity and performance reasons. Moreover, you can keep your own data in it. There convention on field names, to avoid conflicts:


Fields with names started with underscore are for internal use by IO::Stream, you shouldn't touch them or create your own field with such names.


Fields with names started with lower-case letter are part of IO::Stream public interface - you allowed to read/write these fields, but you should not store incorrect values in these fields. Check "PUBLIC FIELDS" below for description of available fields and their format.


You can store your own data in IO::Stream object using field names started with upper-case letter. IO::Stream will not touch these fields.

When some event arise which you're waited for, your callback will be called with 3 parameters: IO::Stream object, event mask, and error (if any):

    sub callback {
        my ($io, $e, $err) = @_;



    IO::Stream->new( \%opt );

Create and return IO::Stream object. You may not keep returned object - you will get it in your callback (in first parameter) when some interesting for your event happens, and will exists until to call method close(). See OVERVIEW for more details.

Fields of %opt become fields of created IO::Stream object. There only few fields required, but you can set any other fields too, and can also set your custom fields (with names starting from upper-case letter).

Only required fields in %opt are {cb} and either {fh} or {host}+{port}. The {wait_for} field also highly recommended to set when creating object.

If {out_buf} will be set, then new() will automatically call write() after creating object.

        fh          => \*STDIN,
        cb          => \&console,
        wait_for    => IN,



Method write() MUST be called after any modifications of {out_buf} field, to ensure data in {out_buf} will be written to {fh} as soon as it will be possible.

If {fh} available for writing when calling write(), then it will write (may be partially) {out_buf} and may immediately call your callback function delivering OUT|SENT events there. So, if you call write() from that callback (as it usually happens), keep in mind it may be called again while executing write(), and object state may significantly change (it even may be close()'d) after it return from write() into your callback.

The write($data) is just a shortcut for:

    $io->{out_buf} .= $data;



Method close() will close {fh} and destroy IO::Stream object. See OVERVIEW for more details.


If field marked *RO* that mean field is read-only and shouldn't be changed.

Some field have default values (shown after equal sign).

Some field modified on events.

method ='IO'

User callback which will be called when some listed in {wait_for} events arise or error happens.

Field {cb} should be either CODE ref or object or class name. In last two cases method named {method} will be called. Field {method} should be string.


Bitmask of events interesting for user. Can be changed at any time. For example:


When some data will be read from {fh}, {wait_for} must contain IN and/or EOF, or error EREQINEOF will be generated. So, it's better to always have IN and/or EOF in {wait_for}.

If {wait_for} contain EOF and doesn't contain IN then {in_buf_limit} must be defined or error EREQINBUFLIMIT will be generated.

fh *RO*

File handle for doing I/O. It's either provided by user to new(), or created by new() (when user provided {host}+{port} instead).

host *RO*
port *RO*

If user doesn't provide {fh} to new(), he should provide {host} and {port} instead. This way new() will create new TCP socket in {fh} and resolve {host} and connect this {fh} to resolved {ip} and {port}. Both resolving and connecting happens in non-blocking way, and will result in delivering RESOLVED and CONNECTED events into user callback (if user {wait_for} these events).

in_buf_limit =undef

Used to avoid DoS attach when user doesn't handle IN events and want his callback called only on EOF event. Must be defined if user have EOF without IN in {wait_for}.

Any value >0 will defined amount of bytes which can be read into {in_buf} before EOF happens. When size of {in_buf} become larger than {in_buf_limit}, error EINBUFLIMIT will be delivered to user callback. In this case user can either remove some data from {in_buf} to make it smaller than {in_buf_limit} or increase {in_buf_limit}, and continue reading data.

NOT RECOMMENDED! Value 0 will switch off DoS protection, so there will be no limit on amount of data to read into {in_buf} until EOF happens.

out_buf =q{} # modified on: OUT
out_pos =undef # modified on: OUT

Data from {out_buf} will be written to {fh}.

If {out_pos} not defined, then data will be written from beginning of {out_buf}, and after successful write written bytes will be removed from beginning of {out_buf}.

If {out_pos} defined, it should be >= 0. In this case data will be written from {out_pos} position in {out_buf}, and after successful write {out_pos} will be incremented by amount of bytes written. {out_buf} will not be changed!

out_bytes =0 # modified on: OUT

Each successful write will increment {out_bytes} by amount of written bytes. You can change {out_bytes} in any way, but it should always be a number.

in_buf =q{} # modified on: IN

Each successful read will concatenate read bytes to {in_buf}. You can change {in_buf} in any way, but it should always be a string.

in_bytes =0 # modified on: IN

Each successful read will increment {in_bytes} by amount of read bytes. You can change {in_bytes} in any way, but it should always be a number.

ip *RO* =undef # modified on: RESOLVED

When you call new() with {host}+{port} instead of {fh} then IP address resolved from {host} will be stored in {ip}, and event RESOLVED will be generated.

is_eof *RO* =undef # modified on: EOF

When EOF event happens {is_eof} will be set to true value. This allow you to detect is EOF already happens at any time, even if you doesn't have EOF in {wait_for}.

plugin *RO* ={}

Allow you to set list of plugins when creating object with new(), and later access these plugins.

This field is somewhat special, because when you call new() you should set plugin to ARRAY ref, but in IO::Stream object {plugin} is HASH ref:

    my $io = IO::Stream->new({
        host        => '',
        port        => 443,
        cb          => \&google,
        wait_for    => EOF,
        in_buf_limit=> 102400,
        out_buf     => "GET / HTTP/1.0\nHost:\n\n",
        plugin      => [    # <------ it's ARRAY, but looks like HASH
            ssl         => IO::Stream::MatrixSSL::Client->new(),
            proxy       => IO::Stream::Proxy::HTTPS->new({
                host        => '',
                port        => 3218,
                user        => 'me',
                pass        => 'my pass',
        MyField1    => 'my data1',
        MyField2    => \%mydata2,

    # access the "proxy" plugin:

This is because when calling new() it's important to keep plugins in order, but later it's easier to access them using names.


Exceptions may be thrown only in new(). All other errors will be delivered to user's callback in last parameter.

usage: IO::Stream->new({ cb=>, wait_for=>, [fh=>, | host=>, port=>,] ... })

You called new() with wrong parameters.

socket: %s
fcntl: %s

Error happens while creating new socket. Usually this happens because you run out of file descriptors.

can't get file descriptor

Failed to get fileno() for your fh. Either fh doesn't open, or this fh type is not supported (directory handle), or fh is not file handle at all.

can't create second object for same fh

You can't have more than one IO::Stream object for same fh.

IO::Stream keep all objects created by new() until $io->close() will be called. Probably you've closed fh in some way without calling $io->close(), then new fh was created with same file descriptor number, and you've tried to create IO::Stream object using new fh.




Bugs / Feature Requests

Please report any bugs or feature requests through the issue tracker at You will be notified automatically of any progress on your issue.

Source Code

This is open source software. The code repository is available for public review and contribution under the terms of the license. Feel free to fork the repository and submit pull requests.

    git clone



Alex Efros <>


This software is Copyright (c) 2008- by Alex Efros <>.

This is free software, licensed under:

  The MIT (X11) License