WHO documentation, updated on 02 Jan 1999.

Since ircu2.10.02 the WHO command had been changed from what described in
RFC1459, while still keeping backward compatibility, actually it has been
changed again in u2.10.05 so that since this release the format of the who
query is now:

[:source] WHO <mask1> [<options> [<mask2>]]

<mask2> is optional, if mask2 is present it's used for matching and mask1 is
ignored, otherwise mask1 is used for matching, since mask2 is the last
parameter it *can* contain a space and this can help when trying to match a
"realname".

When matching IP numbers the <mask> can be in 3 forms:

- The old and well known IRC masks using * and ? as wanted
- The IPmask form a.b.c.d/e.f.g.h as used in most firewalls and
  system configurations, where what is before the / are the bits we expect
  in the IP number and what is after the / is the "filter mask" telling wich
  bits whould be considered and wich should be ignored.
- The IPmask form a.b.c.d/bitcount where bitcount is an integer between 0
  and 31 (inclusive), the matching will be for the IPs whose first
  "bitcount" bits are equal to those in a.b.c.d

Note that:
. The bitcount must be between 0 and 31, 32 is NOT good (and
  makes no sense to use it... just match against the static IP a.b.c.d)
. The missing pieces of both the bitmask and the ipnumber in the forms
  ipnumber/bitmask and ipnumber/bitcount default to zero from right to left,
  this is NOT what inet_aton and most tools do but makes more sense here
  IMO, in example /who 194.243/16 is taken as /who 194.243.0.0/255.255.0.0
  (inet_aton whould take 194.243 as 194.0.0.243).
. For the above reason and specified validity limits 1.2.3.4/31 becomes
  1.2.3.4/255.255.255.254 while 1.2.3.4/32 becomes 1.2.3.4/32.0.0.0 :)

For all the other fields th match happens as has always been, i.e. it's only
considered the IRC mask with * and ? (that is: don't expect to catch an user
with "realname" = "1.2.3.4" when doing "/who 1.2/16 h" :)

For both the masks and the options (and thus for all flags) case is NOT
significative (so "/who <any> o" is exactly the same as "/who <ANY> O".

The "options2" part can be as follows:

 [<flags>][%[<fields>[,<querytype>]]]

in which:

 <flags>: can be a sequence of field matching flags, use mode matching flags
          and special purpose flags

   Field matching flags, when one of these is specified the field in
   question is matched against the mask, otherwise it's not matched.

   n   Nick (in nick!user@host)
   u   Username (in nick!user@host)
   h   Hostname (in nick!user@host)
   i   Numeric IP (the unresolved host)
   s   Servername (the canonic name of the server the guy is on)
   r   Info text (formerly "Realname")
   a   Account name

   If no field-matching flags are specified they default to what old servers
   used to do: nuhsr (= everything except the numeric IP)

   User mode matching flags (specifying one of these means that only clients
   with that umode are considered, what is not specified is always matched):

   o   Irc operator
       [In the future more flags will be supported, basically all
        usermodes plus the +/- specificators to revert the filtering]

   Special purpose flags:

   x   If this is specified the extended visibility of information for opers
       is applied, what this means depends on the fact that you are local or
       global operator and on how the admin configured the server (global
       and eventually local irc opers might be allowed with this flag to see
       +i local users, to see all +i users, to see users into +p and/or +s
       channels, and so on). Using the 'x' flag while not being an irc
       operator is meaningless (it will be ignored), using it while oper'd
       means that the query is almost certainly logged and the admin might
       (rightfully) ask you an explanation on why you did.

   The rest, what follows the %, that is [%[fields[,<querytype>]]], is as it
   has always been since the first who.patch, the <fields> part specifies
   wich fields to include in the output as:

   c : Include (first) channel name
   d : Include "distance" in hops (hopcount)
   f : Include flags (all of them)
   h : Include hostname
   i : Include IP
   l : Include idle time (0 for remote users) [2.10.11+]
   n : Include nick
   r : Include real name
   s : Include server name
   t : Include the querytype in the reply
   u : Include userID with eventual ~
   a : Include account name

And the ,<querytype> final option can be used to specify what you want the
server to say in the querytype field of the output, useful to filter the
output in scripts that do a kind of "on 354 ..."

If no %fields are specified the reply is _exactly_ the same as has always
been, numeric 352, same fields, same order.

If one or more %fields are specified the reply uses a new numeric, since an
out-of-standard 352 crashes EPIC and confuses several other clients. I used
354.

:"source" 354 "target" ["querytype"] ["channel"] ["user"] 
                       ["IP"] ["host"] ["server"] ["nick"] 
                       ["flags"] ["hops"] ["idle"] ["account"]
                       [:"realname"]

Where only the fields specified in the %fields options are present.

"querytype" is the same value passed in the /who command, it is provided to
simplify scripting, in example one could pass a certain value in the query
and have that value "signal" back what is to be done with those replies.

The number of lines in the reply is still limited to avoid self-flooding and
sooner or later another limitation will be added: you will be forced to do
no more than one /who query every 'n' seconds where 'n' depends on the
number of fields you actually match (the field-match flags specified before
% in the option, defaulting to 6 if you don't specify an option at all),
infact matching against many fields as the default query does severely
affects the CPU usage of the server and is *much* better to specify with the
field-matching flags what you are looking for, in example when you are
looking for all french users a "/who *.fr h" is A LOT better than just "/who
*.fr" (and actually you want users that have the
_hostname_ matching *.fr, you wouldn't want to match a japanese user
that has the realname "ku fung-kay aj.fr" in example...)

Note that:

- An user doing a "/who whatever" or a "/who whatever o"
  will not see any change (except for the anti-flood limit and sooner or
  later the CPU usage limit)

- An user doing a "/who #wasteland %n" will get just a list of nicks (lame,
  very lame way of doing it :-)

- An user doing a "/who 0 o%nuhs" will get a list of the opers with Nick,
  userID, server and hostname like:
  
:Amst* 354 Nemesi #wasteland nbakker pc73.a.sn.no Oslo*.org Niels

- An user doing a "/who 0 o%tnuhs,166" will get a list of the opers
  with Nick, userID, server and hostname like the above but with a
  request type field of 166 like:

  :Amst* 354 Nemesi 166 #wasteland nbakker pc73.a.sn.no 
         Oslo-R.NO.EU.Undernet.org Niels
  
  So that he can have in example a script that does 
  on ^354 "% 166" display "There is an oper ..."

- The client will have to sort/format the fields by itself,
  the _order_ in which flags are passed is not significant, the fields in the
  reply will always have the same order.

- The maximum number of _lines_ reported as reply for a query
  is 2048/(n+4) where 'n' is the number of flags "enabled" that is the
  number of fields included in each reply.

  Actually:   1 field  returned = maximum 409 replies
              2 fields returned = maximum 341 replies
              3 fields returned = maximum 292 replies
              4 fields returned = maximum 256 replies
              5 fields returned = maximum 227 replies
              6 fields returned = maximum 204 replies
              7 fields returned = maximum 186 replies (default query)
              8 fields returned = maximum 170 replies
              9 fields returned = maximum 157 replies
             10 fields returned = maximum 146 replies

  If the limit is reached before completing the query the reply is truncated
  and a new numeric error is issued after the "End of WHO", anyway the "end
  of" numeric is _always_ sent (otherwise some scripts and clients go
  crazy).

The actual "mask" to match can have one of the two following forms:

- A comma-separated list of elements: in this case each element
  is treated as a flat channel or nick name and is not matched to the other
  elements. Nicks do count in the limit of output lines (they should not be
  that many anyway), channels count if who asks the query is not on the
  channel. (That is: a /who #channel gives unlimited output if you are in
  there).

- A _single_ mask: in this case (no commas, only one element) the mask is
  first checked to be a full channel or nickname, then it is matched against
  all relevant fiels as already known.  These happens in different steps
  with replicates-removal so that if one has (?) something like "#wasteland"
  as "real name" or is on a channel named "#***MyChan***" it all works
  nicely.

Miscellaneous bug fixes / "undocumented feature" changes:

- /who NickName did not show the user with nick = NickName  when it was
  invisible, even if the nick was given completely (without wildchars) now
  it does, since one could always see him as /whois NickName.  It does not
  report him twice if he also has in example the userID == NickName and is
  -i.

- ":source WHO :The Black Hacker" did not report an user having "The Black
  Hacker" as real name, now it does. Since this can only be done without the
  flags/format specificator because that would become the "last parameter"
  an escape has been provided: if you pass to m_who _3_ parameters the first
  one will be ignored and the last one used for matching, like in example
  ":source WHO foo %nuh :*Black Hacker*" where "foo" will not be used and
  the match will happen on "*Black Hacker*".  (It was passed through
  clean_channelname() that prevented the mask from containing spaces and
  such...)

- When one user was umode -i he was shown or not depending on the 
  fact he was on a +p or +s channel... since we are doing a lookup on the
  _user_ this makes no sense to me, example: 
  Neme1 : umode -i, on no channels, was SEEN with a /who 0
  Neme2 : umode -i, on channel #p with chmode +p, was NOT SEEN by /who 0
  Neme3 : umode -i, on channel #s with chmode +s, was NOT SEEN by /who 0

  Now all users "-i" are matched with a "/who mask", the +i users instead
  must be on a _common_ channel to be seen.

  Basically being on "one" +s|p channel "forced" a +i status while one might
  want to be on #secret (mode +s) and have nobody know that he is in there
  but on the other side stay -i so others can find him.  Of course a +s|p
  channel is never shown in the reply unless who asks the query is in there,
  if no "visible" channels are available for a -i user he is shown on
  "channel *".

- When one user is +i is shown _only_ if there is a common channel,
  the first common channel found is shown in the reply.

- As requested by many persons an escape has been provided for opers,
  when #defined SHOW_ALL_CHANNELS opers can /who #channel from outside
  and see users in there even if the channel is +s|+p
  Each admin decides locally if this feature is enabled to his opers.

- As requested by many admins an escape from the query-size limit
  has been provided for opers, by #defining UNLIMIT_OPER_QUERY opers
  can do unlimited sized /who-s (until they get disconnected by max
  SendQ exceeded ;)
  Again admins will decide if enable or not this feature.

- A /who a,c,b,d,e,f used to return as many ** END OF WHO as there 
  were elements in the list, since now the command is supposed to be
  _efficient_ for /who nick1,nick2,nick3 .. I return a _single_ end
  of query message.

- /who did not work for a channel named in example #**StarWars**  
  now it does handle it properly (the mask was passed through
  collapse() and then.. did not find that channel, fixed).

- "/who #John" did not report an user having '#John' as "Real name",
  now it does (and does NOT report him twice if he is ALSO on a
  channel named #John, strange but true: this can happen).

- "/who a,b,c,d" where a b c and d are channelnames/nicks now uses an hash
  lookup and therefore is extremely efficient, if _only_ one field is
  specified it is looked in all the fields; who really wants _only_ users on
  a specific channel or a single nick (without looking for a match in the
  other fields) can force the server to consider the parameter as a list
  adding a comma somewhere, like:

  "/who #Italia," or "/who ,Nemesi"

  Or even better to avoid misbehaviour with other servers:
  "/who #Italia %... #Italia,"   or "/who Nemesi %... Nemesi,"

  This will make old servers act properly and new ones and should be the
  recomended way for GUI based clients to get a channel's userlist and all
  the infos they want about users on the channel.

- If you use the new numeric, flags will contain all the information about
  a user on a channel.  @ for op'd, + for voiced, and ! for zombie. eg:
  Isomer #coder-com H@+, where the old behavor of just displaying one of
  them has been preserved for the old numeric.  [2.10.11+]

Regards, Andrea aka Nemesi