Belatedly document batch file restrictions

[empserver] / doc / clients-howto

This file contains material useful for client writers.

Protocol
========

Empire's client server protocol is plain text.  It is simple enough
that you could play using nothing more than telnet.  That's a feature.

A session uses either ASCII or UTF-8, controlled by session option
utf-8.  See below for session options.  The session always starts in
ASCII.

Client-server communication is line-oriented.

The server sends lines of output.  Each output line starts with an
identification string, followed by a space, then arbitrary text, then
a line feed.

Identification strings encode small integers called output ids as base
36 numbers.  Characters '0' to '9' represent digits 0 to 9, and 'a' to
'z' represent 10..35, as do 'A' to 'Z'.  Symbolic names for ids are
defined in proto.h.

emp_client versions before version 4.3.11 parse large output ids
incorrectly.  Such ids do not currently occur.

Clients shall be able to safely handle output lines of arbitrary
length.  Naturally, a client may not be able to handle a sufficiently
large line without loss of functionality, but safety must be ensured.

When using ASCII, the most significant bit in text characters
indicates highlighting.

The following control characters have a defined meaning:

  ASCII name         decimal  meaning
  ----------------------------------------------------
  horizontal tab         9    move to next tab stop(1)
  line feed             10    end of line
  shift out             14    begin highlighting(2)
  shift in              15    end highlighting(2)

  (1) Tab stops are every eight columns.
  (2) Only if session uses UTF-8

Other ASCII control characters should not occur and may be ignored by
clients.  Likewise, overlong or malformed UTF-8 sequences should not
occur and may be ignored.

The server prompts for input.  Each prompt `consumes' one line of
input (except for C_EXECUTE, and when the update aborts a command, as
described below).  Input lines are arbitrary text, terminated by line
feed, which is optionally preceded by carriage return (decimal 13).
Lines should not contain ASCII control characters other than
horizontal tab.  Clients should not send overlong or malformed UTF-8
sequences.

A client is called synchronous if it waits for a prompt before it
sends another line of input.  Else it is called asynchronous.

Asynchronous clients must take care to avoid blocking on sending
input.  If the client process blocks that way, it can't receive server
output until the server reads more input.  That may never happen,
because the server may well block on output, which then deadlocks the
session.

An Empire session consists of two phases: login and playing.
emp_client is synchronous during the former and asynchronous during
the latter.  Versions before 4.3.11 could deadlock as described above.

Login phase
-----------

In the login phase, the server prompts for login commands.

The server starts with a C_INIT prompt.  The login phase ends when the
server sends another C_INIT prompt, which starts the playing phase, or
when it closes the connection.

The server replies to a login command with another prompt.  Except as
noted below, the server replies C_BADCMD for syntax errors, C_CMDERR
for other errors, and C_CMDOK on success.  In any case, the prompt
text contains further information intended for humans.

Login commands are:

* client CLIENT-ID...

  Identify the client.  This is optional.  If given, version
  information should be included.

* coun COUNTRY

  Set country name to COUNTRY.

* kill

  If another connection is open for this country, forcibly close it,
  else do nothing.  Country must be authenticated.

  Reply is C_EXIT regardless of success.

* options OPTION[=VALUE]...

  Negotiate session options.  Each argument requests setting OPTION to
  VALUE.  The string VALUE is interpreted in an option-specific way.
  The form without the `=' sets it to an option-specific implied
  value.

  The server accepts the request by sending C_CMDOK.  If the server
  encounters an unknown option, it rejects the request by sending
  C_BADCMD.  It rejects unsupported values by sending C_CMDERR.  It
  may or may not process valid parts of rejected requests.

  If there are no arguments, the server lists its options.  For each
  option, it sends a C_DATA line with OPTION=VALUE as text, and
  finally a C_CMDOK.  If it supports no options at all, it may reply
  with C_BADCMD instead.

  See below for supported session options.

* pass PASSWORD

  Authenticate.  Country name must be set already.

* play [USER COUNTRY PASSWORD]

  Start playing.

  If no arguments are given, the country must be authenticated
  already.

  Else, argument USER sets the user name, COUNTRY sets the country
  name, and PASSWORD authenticates.

  Some error conditions result in a C_EXIT reply.  Clients should
  treat it just like C_CMDERR.

  On success, the server sends C_INIT.  The text is the protocol
  version number in decimal, currently 2.  The client should terminate
  on protocol versions it doesn't know how to handle.

  The protocol version is not expected to change often.  In fact, it
  hasn't changed since the oldest known versions of Empire.

  Unlike the first C_INIT, the second one is not a prompt, i.e the
  server does not consume a line of input for it.

  The session then proceeds to the playing phase.

* quit

  Terminate the session.  The server replies with C_EXIT and closes
  the connection.

* sanc

  List all countries that are still in sanctuary.  The output consists
  of a number of C_DATA lines with human-readable text.

  This command is only recognized if option BLITZ is enabled.

* user NAME

  Set the user name.  This is optional and defaults to "".

Playing phase
-------------

In the playing phase, the server sends data, control and prompt lines.

Clients signal `EOF' by sending a line "ctld\n", and `interrupt' by
sending "aborted\n".  The server treats these conditions specially, as
described below.  Anything else is either a command or input for a
command, depending on how the server prompts for the line.

emp_client signals `EOF' when it encounters an end-of-file condition
while reading player input.  It signals `interrupt' when it catches
SIGINT, which is normally triggered by Ctrl-C.

The following ids occur:

* Command prompt C_PROMPT

  The server consumes a line of input.  On EOF, the server terminates
  the session.  Interrupt is ignored.  Anything else is interpreted as
  Empire command.

  Text is minutes used, space, BTUs left.  Both numbers are in
  decimal.  Clients shall ignore additional text separated by another
  space.

  emp_client prints this prompt using format "[%d:%d] Command : ".
  Clients with a tty-like user interface are advised to use a similar
  format, to minimize differences to the examples in info.

* Argument prompt C_FLUSH

  The server consumes a line of input and passes it to the currently
  executing command.  Commands usually fail on interrupt.  The server
  terminates the session on EOF (but see C_EXECUTE for an exception).

  If an update runs while the server waits for the line of input to
  arrive, the current command is aborted.  Whether the server consumes
  a line of input for this argument prompt is unpredictable.  Any
  argument prompts it may send before the next command prompt do not
  consume input.

  Text is a human-readable prompt supplied by the command.

  emp_client prints the text verbatim.

* Data C_DATA

  Text is human-readable server output.

  emp_client prints the text verbatim.

* Control C_EXECUTE

  Request execution of a batch file.  The text is whatever was on the
  line of input after the execute command.  Its syntax and semantics
  are left to the client.

  emp_client interprets the first word (sequence of non-space
  characters) in the text as file name, and sends the contents of that
  file.

  The security considerations on C_PIPE (below) apply to C_EXECUTE as
  well.

  Note that servers before 4.3.11 sent a copy of the execute command's
  first argument as text.  This made it hard for clients to ensure
  that the text is identical to what was sent, because the server
  strips funny characters and interprets and strips '"' characters
  when splitting input lines into command and arguments.

  emp_client gets confused when old servers mangle the text that way.

  The client shall mark the end of the batch file by signalling EOF as
  described above.  This does not terminate the session.  It may
  signal interrupt if it is unable or unwilling to send the complete
  batch file.

  While executing the batch file, the server rejects redirections and
  execute commands, and sends no C_PROMPT command prompts.  It still
  sends C_FLUSH argument prompts.

  Protocol flaw: not sending C_PROMPT here screws up redirections:
  they'd apply until the next C_PROMPT, i.e. from start of redirected
  command until end of containing batch file.  That's why the server
  rejects redirections in batch files.

  Certain bad failures make the server ignore the rest of the batch
  file file.  This feature is too hard to predict to be really useful.

  Protocol flaw: strictly asynchronous clients cannot support
  C_EXECUTE correctly.  By the time C_EXECUTE arrives, the client may
  have sent more input already.  That input `overtakes' the contents
  of the batch file in the input stream, and is interpreted as part of
  the batch file.  Because this is almost certain to happen when the
  execute comes from a batch file, the server rejects execute commands
  there.

  emp_client has this problem.

  Clients are not required to support C_EXECUTE.  Clients are
  encouraged to offer alternative means for scripting.

* Control C_EXIT

  End of session.  The server is about to close the connection.  Text
  is a human-readable farewell.

  emp_client prints this text prepended with "Exit: ".

* Control C_FLASH

  Asynchronous message.  The client should display the text
  immediately.

  emp_client prints the text verbatim, prepended by a line feed.  This
  is clearly sub-optimal, because it can be inserted in the middle of
  user input.  Clients wishing to to display asynchronous messages
  together with normal I/O should insert them before the current
  prompt.

  Although asynchronous messages can be switched off with the toggle
  command, client support for C_FLASH is not really optional, because
  a C_FLASH could arrive before the client manages to switch them off.
  And if the client lets the user send arbitrary commands, the user
  can switch it back on at any time.  A session option controlling
  C_FLASH would make more sense.  Since all popular clients support
  C_FLASH, it's not worth the trouble.

* Control C_INFORM

  Notification that the number of unread telegrams changed.  The text
  is the notification in human-readable form.

  emp_client prints the last received C_INFORM text right before each
  prompt.  It also repeats the last prompt when a C_INFORM arrives.
  This is sub-optimal just like its treatment of C_FLASH.

  The user can switch these off with the toggle command.  Client
  support is not really optional for the same reasons as for C_FLASH.

* Control C_PIPE

  When a command is redirected to a pipeline, its output is preceded
  by a C_PIPE line.  The text is a copy of the redirection, starting
  with '|'.  Syntax and semantics of the text after the '|' are left
  to the client.

  emp_client executes text after '|' as shell command, with standard
  input connected to the Empire command's output.

  The redirection applies to a single command, i.e. until the next
  C_PROMPT.

  For obvious security reasons, clients supporting pipes shall ensure
  that the text is identical to whatever was sent to the server.  Note
  that the server recognizes redirection only in command lines, not
  argument lines.  Asynchronous clients cannot distinguish the two.

  emp_client prepares for redirections being recognized in any line,
  and copes with only some of them being actually recognized.

* Control C_REDIR

  When a command is redirected to a file, its output is preceded by a
  C_REDIR line.  The text is a copy of the redirection, starting with
  '>', optionally followed by '>' or '!'.  Syntax and semantics of the
  remaining text are left to the client.

  emp_client interprets the first word (sequence of non-space
  characters) in the remaining text as file name, and redirects the
  command's output to that file.  The use of the first word is
  unfortunate, as it arbitrarily limits the user's choice of file
  names.  If the text starts with '>!', it silently overwrites any
  existing file, with '>>' it appends to the file, and with just '>'
  it refuses to overwrite an existing file.

  The redirection applies to a single command, i.e. until the next
  C_PROMPT.

  The security considerations on C_PIPE apply to C_REDIR as well.

* Other ids

  Other ids do not occur currently.  Clients shall deal gracefully
  with ids they don't know.

  emp_client treats unknown ids like C_DATA.  Versions before 4.3.11
  prepend "Aborted\n" to C_ABORT lines, and "Error; " to C_CMDERR and
  C_BADCMD lines for historical reasons.


Session Options
===============

Session options control client preferences.  They are not persistent;
each session starts with the same default session options.

The only session option so far is utf-8.  It controls whether to use
ASCII (disabled) or UTF-8 (enabled).  Initial value is disabled.
Setting it to 0 disables, setting it to 1 enables, and the implied
value is enabled.

Session options should not be confused with command toggle, which
controls player preferences.  Clients should leave those accessible to
their users.


Commands Useful for Clients
===========================

Traditional dumps
-----------------

A number of commands are available for clients that wish to maintain
their own game state.  These commands are called dumps.

      dump  - Dump sector information
      ldump - Dump land unit information
      lost  - Report lost items
      ndump - Dump nuclear stockpile information
      pdump - Dump plane information
      sdump - Dump ship information

See the various info pages on these for complete documentation on how
they work and how you can use them to help improve your clients.

Each of the above commands prints a timestamp, which is a decimal
number.  This together with the timestamp selector enables you to dump
incrementally, i.e.  retrieve only what changed since another dump.
For instance, if `dump *' printed timestamp 855544597, `dump *
?timestamp>855544596' dumps everything changed since.

Note that the condition compares with the timestamp value minus one.
That's because timestamps have a noticeable granularity: things may
change between a dump and the next timestamp increase.

Timestamp values are currently seconds since the epoch, but this might
change, and clients are advised not to rely on it.

Extended dump
-------------

Traditional dumps have a number of shortcomings.  They cover only the
most important part of the game state (sectors, ships, planes, land
units, nukes), but not game configuration, loans, news, and so forth.
They are not quite complete even for what they attempt to cover.
Finally, their output is harder to parse than necessary.

The new `xdump' command is designed to overcome these deficiencies.
See doc/xdump for the full story.


Advice on parsing human-readable command output
===============================================

To be written.