Client-To-Client Protocol (CTCP)

CTCP – Bezug Internet Relay Chat

Client-To-Client Protocol (CTCP) ist ein wichtiger Bestandteil des Internet-Protokolls, der die Kommunikation zwischen verschiedenen Anwendungen auf Computern ermöglicht. Das CTCP wird oft in Internet Relay Chat (IRC)-Systemen verwendet, um direkte Kommunikation zwischen Benutzern zu ermöglichen.

CTCP funktioniert, indem es eine spezielle Nachrichtenstruktur verwendet, die es Benutzern ermöglicht, direkte Befehle an andere Benutzer zu senden. Dies kann nützlich sein, um Informationen über einen Benutzer zu erhalten, wie z.B. die Art des verwendeten Betriebssystems oder die Version der verwendeten IRC-Software. CTCP wird auch oft verwendet, um Dateien direkt zwischen Benutzern zu senden oder um Informationen über eine laufende Chat-Sitzung zu teilen.

Eine der nützlichsten Funktionen von CTCP ist die Möglichkeit, den Status eines Benutzers abzufragen, z.B. ob er gerade verfügbar oder abwesend ist. Diese Funktion ist besonders nützlich in Chat-Systemen, die von Unternehmen oder Organisationen genutzt werden, um die Verfügbarkeit von Mitarbeitern zu überwachen.

Insgesamt ist das Client-To-Client Protocol eine wichtige Technologie, die dazu beiträgt, die Kommunikation zwischen Benutzern auf verschiedenen Computern zu verbessern. Es ermöglicht es Benutzern, direkt miteinander zu interagieren, Dateien zu teilen und Informationen auszutauschen, was zu einer reibungsloseren und effizienteren Zusammenarbeit führt.

The Client-To-Client Protocol (CTCP)
Klaus Zeuge <sojge@Minsk.DoCS.UU.SE>
Troy Rollo <troy@plod.cbme.unsw.oz.au>
Ben Mesander <ben@gnu.ai.mit.edu>

The Client-To-Client Protocol is meant to be used as a way to
1/ in general send structured data (such as graphics,
voice and different font information) between users
clients, and in a more specific case:
2/ place a query to a users client and getting an answer.

*****************************************
BASISPROTOKOLL ZWISCHEN CLIENTS UND SERVER
*****************************************

Characters between an Internet Relay Chat (IRC) client and server are
8 bit bytes (also known as octets) and can have numeric values from
octal \000 to \377 inclusive (0 to 255 decimal). Some characters are
special:

CHARS ::= ‚\000‘ .. ‚\377‘ 
NUL ::= ‚\000‘
NL ::= ‚\n‘
CR ::= ‚\r‘

Note: `\‘ followed by three digits is used to denote an octal value in this
paper. `\‘ followed by an alphabetic character is used to denote a C
language style special character, and `..‘ denotes a range of characters.

A line sent to a server, or received from a server (here called „low
level messages“) consist or zero or more octets (expcept NUL, NL or
CR) with either a NL or CR appended.

L-CHARS ::= ‚\001‘ .. ‚\011‘ | ‚\013‘ | ‚\014‘ |
‚\016‘ .. ‚\377‘
L-LINE ::= L-CHARS* CR LF

Note: The `*‘ is used here to denote „zero or more of the preceding class of 
characters“, and the `|‘ is used to denote alternation.

A NUL is never sent to the server.

LOW LEVEL QUOTING

Even though messages to and from IRC servers cannot contain NUL, NL,
or CR, it still might be desirable to send ANY character (in so called
„middle level messages“) between clients. In order for this to be
possible, those three characters have to be quoted. Therefore a quote
character is needed. Of course, the quote character itself has to be
quoted too, since it is in-band.

M-QUOTE ::= ‚\020‘

(Ie a CNTRL/P).

When sending a middle level message, if there is a character in the
set { NUL, NL, CR, M-QUOTE } present in the message, that character is
replaced by a two character sequence according to the following table:

NUL –> M-QUOTE ‚0‘
NL –> M-QUOTE ’n‘
CR –> M-QUOTE ‚r‘
M-QUOTE –> M-QUOTE M-QUOTE

When receiving a low level message, if there is a M-QUOTE, look at the
next character, and replace those two according to the following table
to get the corresponding middle level message:

M-QUOTE ‚0‘ –> NUL
M-QUOTE ’n‘ –> NL
M-QUOTE ‚r‘ –> CR
M-QUOTE M-QUOTE –> M-QUOTE

If the character following M-QUOTE is not any of the listed
characters, that is an error, so drop the M-QUOTE character from the
message, optionally warning the user about it. For example, a string
‚x‘ M-QUOTE ‚y‘ ‚z‘ from a server dequotes into ‚x ‚y‘ ‚z‘.

Before low level quoting, a message to the server (and in the opposite
direction: after low level dequoting, a message from the server) looks
like:

M-LINE ::= CHARS*

TAGGED DATA

To send both extended data and query/reply pairs between clients, an
extended data format is needed. The extended data are sent in the text
part of a middle level message (and after low level quoting, in the
text part of the low level message).

To send extended data inside the middle level message, we need some
way to delimit it. This is done by starting and ending extended data
with a delimiter character, defined as:

X-DELIM ::= ‚\001‘

As both the starting and ending delimiter looks the same, the first
X-DELIM is called the odd delimiter, and the one that follows, the
even delimiter. The next one after that, an odd delimiter, then and
even, and so on.

When data are quoted (and conversely, before being dequoted) any number
of characters of any kind except X-DELIM can be used in the extended
data inside the X-DELIM pair.

X-CHR ::= ‚\000‘ | ‚\002‘ .. ‚\377‘

An extended message is either empty (nothing between the odd and even
delimiter), has one or more non-space characters (any character but
‚\040‘) or has one or more non-space characters followed by a space
followed by zero or more characters.

X-N-AS ::= ‚\000‘ | ‚\002‘ .. ‚\037‘ | ‚\041‘ .. ‚\377‘
SPC ::= ‚\040‘
X-MSG ::= | X-N-AS+ | X-N-AS+ SPC X-CHR*

Note: Here `+‘ is used to denote „one or more of the previous class of 
characters“, and `*‘ is used to denote „zero or more of the previous 
class of characters“.

The characters up until the first SPC (or if no SPC, all of the X-MSG)
is called the tag of the extended message. The tag is used to denote
what kind of extended data is used.

The tag can be *any* string of characters, and if it contains
alphabetics, it is case sensitive, so upper and lower case matters.

Extended data is only valid in PRIVMSG and NOTICE commands. If the
extended data is a reply to a query, it is sent in a NOTICE, otherwise
it is sent in a PRIVMSG. Both PRIVMSG and NOTICE to a user and to a
channel may contain extended data.

The text part of a PRIVMSG or NOTICE might contain zero or more
extended messages, intermixed with zero or more chunks of non-extended
data.

******************
CTCP LEVEL QUOTING
******************

In order to be able to send the delimiter X-DELIM inside an extended
data message, it has to be quoted. This introduces another quote
character (which differs from the low level quote character so it
won’t have to be quoted yet again).

X-QUOTE ::= ‚\134‘

(a back slash – `\‘).

When quoting on the CTCP level, only the actual CTCP message (extended
data, queries, replies) are quoted. This enables users to actually
send X-QUOTE characters at will. The following translations should be
used:

X-DELIM –> X-QUOTE ‚a‘
X-QUOTE –> X-QUOTE X-QUOTE

and when dequoting on the CTCP level, only CTCP messages are dequoted
whereby the following table is used.

X-QUOTE ‚a‘ –> X-DELIM
X-QUOTE X-QUOTE –> X-QUOTE

If an X-QUOTE is seen with a character following it other than the
ones above, that is an error and the X-QUOTE character should be
dropped. For example the CTCP-quoted string ‚x‘ X-QUOTE ‚y‘ ‚z‘
becomes after dequoting, the three character string ‚x‘ ‚y‘ ‚z‘.

If a X-DELIM is found outside a CTCP message, the message will contain
the X-DELIM. (This should only happen with the last X-DELIM when there
are an odd number of X-DELIM’s in a middle level message.)

****************
QUOTING EXAMPLES
****************

There are three levels of messages. The highest level (H) is the text
on the user-to-client level. The middle layer (M) is on the level
where CTCP quoting has been applied to the H-level message. The lowest
level (L) is on the client-to-server level, where low level quoting
has been applied to the M-level message.

The following relations are true, with lowQuote(message) being a
function doing the low level quoting, lowDequote(message) the low
level dequoting function, ctcpQuote(message) the CTCP level quoting
function, ctcpDequote(message) the CTCP level dequoting function, and
ctcpExtract(message) the function which removes all CTCP messages from
a message:

L = lowQuote(M)
M = ctcpDequote(L)

M = ctcpQuote(H)
H = ctcpDequote(ctcpExtract(M))

When sending a CTCP message embedded in normal text:

M = ctcpQuote(H1) || ‚\001‘ || ctcpQuote(X) || ‚\001‘ || ctcpQuote(H2)

Note: The operator || denotes string concatenation.

Of course, there might be zero or more normal text messages and zero
or more CTCP messages mixed.

– — Example 1 —————————————————————–

A user (called actor) wanting to send the string:

Hi there!\nHow are you?

to user victim, i.e. a message where the user has entered an inline
newline (how this is done, if at all, differs from client to client),
will result internaly in the client in the command:

PRIVMSG victim :Hi there!\nHow are you? \K?

which will be CTCP quoted into:

PRIVMSG victim :Hi there!\nHow are you? \\K?

which in turn will be low level quoted into:

PRIVMSG victim :Hi there!\020nHow are you? \\K?

and sent to the server after appending a newline at the end.

This will arrive on victim’s side as:

:actor PRIVMSG victim :Hi there!\020nHow are you? \\K?

(where the \\K would look similar to OK in SIS D47 et. al.) which after
low level dequoting becomes:

:actor PRIVMSG victim :Hi there!\nHow are you? \\K?

and after CTCP dequoting:

:actom PRIVMSG victim :Hi there!\nHow are you? \K?

How this is displayed differs from client to client, but it suggested
that a line break should occour between the words „there“ and „How“.

– — Example 2 —————————————————————–

If actor’s client wants to send the string „Emacs wins“ this might
become the string „\n\t\big\020\001\000\\:“ when being SED-encrypted
[SED is a simple encryption protocol between IRC clients implemented
with CTCP. I don’t have any reference for it — Ben] using some key,
so the client starts by CTCP-quoting this string into the string
„\n\t\big\020\\a\000\\\\:“ and builds the M-level message:

PRIVMSG victim :\001SED \n\t\big\020\\a\000\\\\:\001

which after low level quoting becomes:

PRIVMSG victim :\001SED \020n\t\big\020\020\\a\0200\\\\:\001

which will be sent to the server, with a newline tacked on.

On victim’s side, the string:

:actor PRIVMSG victim :\001SED \020n\t\big\020\020\\a\0200\\\\:\001

will be received from the server and low level dequoted into:

:actor PRIVMSG victim :\001SED \n\t\big\020\\a\000\\\\:\001

whereafter the string „\n\t\big\020\\a\000\\\\:“ will be extracted
and first CTCP dequoted into „\n\t\big\020\001\000\\:“ and then
SED decoded getting back „Emacs wins“ when using the same key.

– — Example 3 —————————————————————–

If the user actor wants to query the USERINFO of user victim, and is
in the middle of a conversation, the client may decide to tack on
USERINFO request on the end of a normal text message. Let’s say actor
wants to send the textmessage „Say hi to Ron\n\t/actor“ and the CTCP
request „USERINFO“ to victim:

PRIVMSG victim :Say hi to Ron\n\t/actor

plus:

USERINFO

which after CTCP quoting become:

PRIVMSG victim :Say hi to Ron\n\t/actor

plus:

USERINFO

which gets merged into:

PRIVMSG victim :Say hi to Ron\n\t/actor\001USERINFO\001

and after low level quoting:

PRIVMSG victim :Say hi to Ron\020n\t/actor\001USERINFO\001

and sent off to the server.

On victim’s side, the message:

:actor PRIVMSG victim :Say hi to Ron\020n\t/actor\001USERINFO\001

arrives. This gets low level dequoted into:

:actor PRIVMSG victim :Say hi to Ron\n\t/actor\001USERINFO\001

and thereafter split up into:

:actor PRIVMSG victim :Say hi to Ron\n\t/actor

plus:

USERINFO

After CTCP dequoting both, the message:

:actor PRIVMSG victim :Say hi to Ron\n\t/actor

gets displayed, while the CTCP command:

USERINFO

gets replied to. The reply might be:

USERINFO :CS student\n\001test\001

which gets CTCP quoted into:

USERINFO :CS student\n\\atest\\a

and sent in a NOTICE as it is a reply:

NOTICE actor :\001USERINFO :CS student\n\\atest\\a\001

and low level quoted into:

NOTICE actor :\001USERINFO :CS student\020n\\atest\\a\001

after which is it sent to victim’s server.

When arriving on actor’s side, the message:

:victim NOTICE actor :\001USERINFO :CS student\020n\\atest\\a\001

gets low level dequoted into:

:victim NOTICE actor :\001USERINFO :CS student\n\\atest\\a\001

At this point, all CTCP replies get extracted, giving 1 CTCP reply and
no normal NOTICE:

USERINFO :CS student\n\\atest\\a

The remaining reply gets CTCP dequoted into:

USERINFO :CS student\n\001test\001

and presumly displayed to user actor.

*******************
KNOWN EXTENDED DATA
*******************

Extended data passed between clients can be used to pass structured
information between them. Currently known extended data types are:

ACTION – Used to simulate „role playing“ on IRC.
DCC – Negotiates file transfers and direct tcp chat 
connections between clients.
SED – Used to send encrypted messages between clients.

ACTION
======
This is used by losers on IRC to simulate „role playing“ games. An
action message looks like the following:

\001ACTION barfs on the floor.\001

Clients that recieve such a message should format them to indicate the
user who did this is performing an „action“. For example, if the user
„actor“ sent the above message to the channel „#twilight_zone“, other
users clients might display the message as:

[ACTION] actor->#twilight_zone: barfs on the floor.

Presumably other users on the channel are suitably impressed.

DCC
=== 
DCC stands for something like „Direct Client Connection“. CTCP DCC
extended data messages are used to negotiate file transfers between
clients and to negotiate chat connections over tcp connections between
two clients, with no IRC server involved. Connections between clients
involve protocols other than the usual IRC protocol. Due to this
complexity, a full description of the DCC protocol is included
separately at the end of this document in Appendix A.

SED
===
SED probably stands for something like „Simple Encryption D???“. It is
used by clients to exchange encrypted messages between clients. A
message encoded by SED probably looks something like:

\001SED encrypted-text-goes-here\001

Clients which accept such messages should display them in decrypted
form. It would be nice if someone documented this, and included the
encryption scheme in an Appendix B.

*************************
KNOWN REQUEST/REPLY PAIRS
*************************

A request/reply pair is sent between the two clients in two phases.
The first phase is to send the request. This is done with a „privmsg“
command (either to a nick or to a channel — it doesn’t matter).

The second phase is to send a reply. This is done with a „notice“
command.

The known request/reply pairs are for the following commands.

FINGER – Returns the user’s full name, and idle time.
VERSION – The version and type of the client.
SOURCE – Where to obtain a copy of a client.
USERINFO – A string set by the user (never the client coder)
CLIENTINFO – Dynamic master index of what a client knows.
ERRMSG – Used when an error needs to be replied with.
PING – Used to measure the delay of the IRC network
between clients.
TIME – Gets the local date and time from other clients.

FINGER
======
This is used to get a user’s real name, and perhaps also the idle time
of the user (this usage has been obsoleted by enhancements to the IRC
protocol. The request is in a „privmsg“ and looks like

\001FINGER\001

while the reply is in a „notice“ and looks like

\001FINGER :#\001

where the # denotes contains information about the users real name,
login name at clientmachine and idle time and is of type X-N-AS.

VERSION
=======
This is used to get information about the name of the other client and
the version of it. The request in a „privmsg“ is simply

\001VERSION\001

and the reply

\001VERSION #:#:#\001

where the first # denotes the name of the client, the second # denotes
the version of the client, the third # the enviroment the client is
running in.

Using

X-N-CLN ::= ‚\000‘ .. ‚\071‘ | ‚\073‘ .. ‚\377‘

the client name is a string of type X-N-CLN saying things like „Kiwi“
or „ircII“, the version saying things like „5.2“ or „2.1.5c“, the
enviroment saying things like „GNU Emacs 18.57.19 under SunOS 4.1.1 on
Sun SLC“ or „Compiled with gcc -ansi under Ultrix 4.0 on VAX-11/730“.

SOURCE

This is used to get information about where to get a copy of the
client. The request in a „privmsg“ is simply

\001SOURCE\001

and the reply is zero or more CTCP replies of the form

\001SOURCE #:#:#\001

followed by an end marker

\001SOURCE\001

where the first # is the name of an Internet host where the client can
be gotten from with anonymous FTP the second # a directory names, and
the third # a space separated list of files to be gotten from that
directory.

Using

X-N-SPC ::= ‚\000‘ .. ‚\037‘ | ‚\041‘ .. ‚\377‘

the name of the FTP site is to be given by name like „cs.bu.edu“ or
„funic.funet.fi“.

The file name field is a directory specification optionally followed
by one or more file names, delimited by spaces. If only a directory
name is given, all files in that directory should be copied when
retrieving the clients source. If some files are given, only those
files in that directpry should be copied. Note that the spcification
allows for all characters but space in the names, this includes
allowing :. Examples are „pub/emacs/irc/“ to get all files in
directory pub/emacs/irc/, the client should be able to first login as
user „ftp“ and the give the command „CD pub/emacs/irc/“, followed by
the command „mget *“. (It of course has to take care of binary and
prompt mode too). Another example is „/pub/irc Kiwi.5.2.el.Z“ in which
case a „CD /pub/irc“ and „get Kiwi.5.2.el.Z“ is what should be done.

USERINFO
========
This is used to transmit a string which is settable by the user (and
never should be set by the client). The query is simply

\001USERINFO\001

with the reply

\001USERINFO :#\001

where the # is the value of the string the client’s user has set.

CLIENTINFO
==========
This is for client developers use to make it easier to show other
client hackers what a certain client knows when it comes to CTCP. The
replies should be fairly verbose explaining what CTCP commands are
understood, what arguments are expected of what type, and what replies
might be expected from the client.

The query is the word CLIENTINFO in a „privmsg“ optionally followed by
a colon and one or more specifying words delimited by spaces, where
the word CLIENTINFO by itself,

\001CLIENTINFO\001

should be replied to by giving a list of known tags (see above in
section TAGGED DATA). This is only intended to be read by humans.

With one argument, the reply should be a description of how to use
that tag. With two arguments, a description of how to use that
tag’s subcommand. And so on.

ERRMSG
======
This is used as a reply whenever an unknown query is seen. Also, when
used as a query, the reply should echo back the text in the query,
together with an indication that no error has happened. Should the
query form be used, it is

\001ERRMSG #\001

where # is a string containing any character, with the reply

\001ERRMSG # :#\001

where the first # is the same string as in the query and the second #
a short text notifying the user that no error has occurred.

A normal ERRMSG reply which is sent when a corrupted query or some
corrupted extended data is received, looks like

\001ERRMSG # :#\001

where the first # is the the failed query or corrupted extended data
and the second # a text explaining what the problem is, like „unknown
query“ or „failed decrypting text“.

PING
====
Ping is used to measure the time delay between clients on the IRC
network. A ping query is encoded in a privmsg, and has the form:

\001PING timestamp\001

where `timestamp‘ is the current time encoded in any form the querying
client finds convienent. The replying client sends back an identical
message inside a notice:

\001PING timestamp\001

The querying client can then subtract the recieved timestamp from the
current time to obtain the delay between clients over the IRC network.

TIME
====
Time queries are used to determine what time it is where another
user’s client is running. This can be useful to determine if someone
is probably awake or not, or what timezone they are in. A time query
has the form:

\001TIME\001

On reciept of such a query in a privmsg, clients should reply with a
notice of the form:

\001TIME :human-readable-time-string\001

For example:

\001TIME :Thu Aug 11 22:52:51 1994 CST\001

********
EXAMPLES
********

Sending

PRIVMSG victim :\001FINGER\001

might return

:victim NOTICE actor :\001FINGER :Please check my USERINFO
instead :Klaus Zeuge (sojge@mizar) 1 second has passed since
victim gave a command last.\001

(this is only one line) or why not

:victim NOTICE actor :\001FINGER :Please check my USERINFO
instead :Klaus Zeuge (sojge@mizar) 427 seconds (7 minutes and
7 seconds) have passed since victim gave a command last.\001

if Klaus Zeuge happens to be lazy? 🙂

Sending

PRIVMSG victim :\001CLIENTINFO\001

might return

:victim NOTICE actor :\001CLIENTINFO :You can request help of the
commands CLIENTINFO ERRMSG FINGER USERINFO VERSION by giving
an argument to CLIENTINFO.\001

Sending

PRIVMSG victim :\001CLIENTINFO CLIENTINFO\001

might return

:victim NOTICE actor :\001CLIENTINFO :CLIENTINFO with 0
arguments gives a list of known client query keywords. With 1
argument, a description of the client query keyword is
returned.\001

while sending

PRIVMSG victim :\001clientinfo clientinfo\001

probably will return something like

:victim NOTICE actor :\001ERRMSG clientinfo clientinfo :Query is
unknown\001

as tag „clientinfo“ isn’t known.

Sending

PRIVMSG victim :\001CLIENTINFO ERRMSG\001

might return

:victim NOTICE actor :\001CLIENTINFO :ERRMSG is the given answer
on seeing an unknown keyword. When seeing the keyword ERRMSG,
it works like an echo.\001

Sending

PRIVMSG victim :\001USERINFO\001

might return the somewhat pathetically long

:victim NOTICE actor :\001USERINFO :I’m studying computer
science in Uppsala, I’m male (somehow, that seems to be an
important matter on IRC:-) and I speak fluent swedish, decent
german, and some english.\001

Sending

PRIVMSG victim :\001VERSION\001

might return:

:victim NOTICE actor :\001VERSION Kiwi:5.2:GNU Emacs
18.57.19 under SunOS 4.1.1 on Sun
SLC:FTP.Lysator.LiU.SE:/pub/emacs Kiwi-5.2.el.Z
Kiwi.README\001

if the client is named Kiwi of version 5.2 and is used under GNU Emacs
18.57.19 running on a Sun SLCwith SunOS 4.1.1. The client claims a
copy of it can be found with anonymous FTP on FTP.Lysator.LiU.SE after
giving the FTP command „cd /pub/emacs/“. There, one should get files
Kiwi-5.2.el.Z and Kiwi.README; presumably one of the files tells how to
proceed with building the client after having gotten the files.

——————————————————————————–

**********************************************************************
Appendix A — A description of the DCC protocol
**********************************************************************

By Troy Rollo (troy@plod.cbme.unsw.oz.au)
Revised by Ben Mesander (ben@gnu.ai.mit.edu)

Troy Rollo, the original implementor of the DCC protocol, said
that the DCC protocol was never designed to be portable to clients
other than IRCII. However, time has shown that DCC is useable in
environments other than IRCII. IRC clients in diverse languages, such
as ksh, elisp, C, and perl have all had DCC implementations.

Why DCC?
========

DCC allows the user to overcome some limitations of the IRC
server network and to have a somewhat more secure chat connection
while still in an IRC-oriented protocol.

DCC uses direct TCP connections between the clients taking
part to carry data. There is no flood control, so packets can be sent
at full speed, and there is no dependance on server links (or load
imposed on them). In addition, since only the initial handshake for
DCC conections is passed through the IRC network, it makes it harder
for operators with cracked servers to spy on personal messages.

How?
====

The initial socket for a DCC connection is created
by the side that initiates (Offers) the connection. This socket
should be a TCP socket bound to INADDR_ANY, listening for
connections.

The Initiating client, on creating the socket, should
send its details to the target client using the CTCP command
DCC. This command takes the form:

DCC type argument address port [size]

type – The connection type.
argument – The connectin type dependant argument.
address – The host address of the initiator as an integer.
port – The port or the socket on which the initiator expects
to receive the connection.
size – If the connection type is „SEND“ (see below), then size
will indicate the size of the file being offered. Obsolete
IRCII clients do not send this, so be prepared if this is
not present.

The address, port, and size should be sent as ASCII representations of
the decimal integer formed by converting the values to host byte order
and treating them as an unsigned long, unsigned short, and unsigned
long respectively.

Implementations of the DCC protocol should be prepared to
accept further arguments in a CTCP DCC message. There has been some
discussion of adding another argument that would specify the type of
file being transferred – text, binary, and perhaps others if DCC is
implemented on operating systems other than UNIX. If additional
arguments are added to the protocol, they should have semantics such
that clients which ignore them will interoperate with clients that
don’t in a sensible way.

The following DCC connection types are defined:

Type Purpose Argument
CHAT To carry on a semi-secure conversation the string „chat“
SEND To send a file to the recipient the file name

Although the following subcommand is included in the IRCII DCC command,
it does _not_ transmit a DCC request via IRC, and thus is not
discussed in this document:

TALK Establishes a TALK connection

Implementation
==============

The CHAT and SEND connection types should not be
accepted automatically as this would create the potential for
terrorism. Instead, they should notify the user that an
offer has been made, and allow the user to accept it.

The recipient should have the opportunity to rename a file
offered with the DCC SEND command prior to retrieving it. It is also
desirable to ensure that the offered file will not overwrite an
existing file.

Older IRCII clients send the entire pathname of the file being
transmitted. This is annoying, and newer clients should simply send
the filename portion of the file being transmitted.

The port number should be scrutinized – if the port number is
in the UNIX reserved port range, the connection should only be
accepted with caution.

If it is not possible in the client implementation language to
handle a 32-bit integer (for instance emacs 18 elisp and ksh 88), then
it is often possible to use the hostname in the originating PRIVMSG.

The following are the steps which should occur in the clients
(this description assumes use of the BSD socket interface on a UNIX
system).

Initiator:
DCC command issued.
Create a socket, bind it to INADDR_ANY, port 0, and
make it passive (a listening socket).
Send the recipient a DCC request via CTCP supplying
the address and port of the socket. (This
is ideally taken from the address of the local
side of the socket which is connected to a
server. This is presumably the interface on
the host which is closest to the rest of
the net, and results in one less routing hop
in the case of gateway nodes).
Continue normally until a connection is received.

On a connection:
Accept the connection.
Close the original passive socket.
Conduct transaction on the new socket.

Acceptor:
CTCP DCC request received.
Record information on the DCC request and notify the user.

At this point, the USER should be able to abort (close) the
request, or accept it. The request should be accepted with
a command specifying the sender, type, and argument, or
a subset of these where no ambiguity exists.

If accepted, create a TCP socket.
Connect the new socket to the address and port supplied.
Conduct the transaction over the socket.

Type specific details.
======================

CHAT Data sent across a CHAT connection should be sent line-by-line
without any prefixes or commands. A CHAT connection ends when
one party issues the DCC CLOSE command to their clients, which
causes the socket to be closed and the information on the connection
to be discarded. The terminating character of each line is a 
newline character, ‚\n‘.

FILE Data is sent in packets, rather than dumped in a stream manner.
This allows the DCC SEND connection to survive where an FTP
connection might fail. The size of the packets is up to the
client, and may be set by the user. Smaller packets result
in a higher probability of survival over bad links.
The recipient should acknowledge each packet by transmitting
the total number of bytes received as an unsigned, 4 byte
integer in network byte order. The sender should not continue
to transmit until the recipient has acknowledged all data
already transmitted. Additionally, the sender should not
close the connection until the last byte has been
acknowledged by the recipient.

Older IRCII clients do not send the file size of the file
being transmitted via DCC. For those clients, note that it is
not possible for the recipient to tell if the entire file has
been received – only the sender has that information, although
IRCII does not report it. Users generally verify the transfer
by checking file sizes. Authors of clients are urged to use
the size feature.

Note also that no provision is made for text translation.

The original block size used by IRCII was 1024. Other clients
have adopted this. Note, however, that an implementation should accept
any blocksize. IRCII currently allows a user-settable blocksize.