If sess is given, this becomes the name of the session which can be used to refer to multiple different TCP connections. If sess is not given, the function will invent a numeric name value (note this is not the same as the file descriptor to which the session is attached). It is recommended that session names not include `funny' characters, where funny characters are not well-defined but certainly do not include alphanumerics or underscores, and certainly do include whitespace.

In the second case, one or more sessions to be opened are given by name. A single session name is given after -s and a comma-separated list after -l; both options may be repeated as many times as necessary. A failure to open any session causes tcp_open to abort. The host and port are read from the file .ztcp_sessions in the same directory as the user's zsh initialisation files, i.e. usually the home directory, but $ZDOTDIR if that is set. The file consists of lines each giving a session name and the corresponding host and port, in that order (note the session name comes first, not last), separated by whitespace.

The third form allows passive and fake TCP connections. If the option -a is used, its argument is a file descriptor open for listening for connections. No function front-end is provided to open such a file descriptor, but a call to `ztcp -l port' will create one with the file descriptor stored in the parameter $REPLY. The listening port can be closed with `ztcp -c fd'. A call to `tcp_open -a fd' will block until a remote TCP connection is made to port on the local machine. At this point, a session is created in the usual way and is largely indistinguishable from an active connection created with one of the first two forms.

If the option -f is used, its argument is a file descriptor which is used directly as if it were a TCP session. How well the remainder of the TCP function system copes with this depends on what actually underlies this file descriptor. A regular file is likely to be unusable; a FIFO (pipe) of some sort will work better, but note that it is not a good idea for two different sessions to attempt to read from the same FIFO at once.

If the option -q is given with any of the three forms, tcp_open will not print informational messages, although it will in any case exit with an appropriate status.

If the line editor (zle) is in use, which is typically the case if the shell is interactive, tcp_open installs a handler inside zle which will check for new data at the same time as it checks for keyboard input. This is convenient as the shell consumes no CPU time while waiting; the test is performed by the operating system. Giving the option -z to any of the forms of tcp_open prevents the handler from being installed, so data must be read explicitly. Note, however, this is not necessary for executing complete sets of send and read commands from a function, as zle is not active at this point. Generally speaking, the handler is only active when the shell is waiting for input at a command prompt or in the vared builtin. The option has no effect if zle is not active; `[[ -o zle]]' will test for this.

The first session to be opened becomes the current session and subsequent calls to tcp_open do not change it. The current session is stored in the parameter $TCP_SESS; see below for more detail about the parameters used by the system.

The function tcp_on_open, if defined, is called when a session is opened. See the description below.

If the session being closed is the current one, $TCP_SESS is unset, leaving no current session, even if there are other sessions still open.

If the session was opened with tcp_open -f, the file descriptor is closed so long as it is in the range 0 to 9 accessible directly from the command line. If the option -n is given, no attempt will be made to close file descriptors in this case. The -n option is not used for genuine ztcp session; the file descriptors are always closed with the session.

If the option -q is given, no informational messages will be printed.

The function checks for new data available on all the sessions listed. Unless the -b option is given, it will not block waiting for new data. Any one line of data from any of the available sessions will be read, stored in the parameter $TCP_LINE, and displayed to standard output unless $TCP_SILENT contains a non-empty string. When printed to standard output the string $TCP_PROMPT will be shown at the start of the line; the default form for this includes the name of the session being read. See below for more information on these parameters. In this mode, tcp_read can be called repeatedly until it returns status 2 which indicates all pending input from all specified sessions has been handled.

With the option -b, equivalent to an infinite timeout, the function will block until a line is available to read from one of the specified sessions. However, only a single line is returned.

The option -d indicates that all pending input should be drained. In this case tcp_read may process multiple lines in the manner given above; only the last is stored in $TCP_LINE, but the complete set is stored in the array $tcp_lines. This is cleared at the start of each call to tcp_read.

The options -t and -T specify a timeout in seconds, which may be a floating point number for increased accuracy. With -t the timeout is applied before each line read. With -T, the timeout applies to the overall operation, possibly including multiple read operations if the option -d is present; without this option, there is no distinction between -t and -T.

The function does not print informational messages, but if the option -q is given, no error message is printed for a non-existent session.

A return status of 2 indicates a timeout or no data to read. Any other non-zero return status indicates some error condition.

See tcp_log for how to control where data is sent by tcp_read.

The option -c causes tcp_send to behave like cat. It reads lines from standard input until end of input and sends them in turn to the specified session(s) exactly as if they were given as data arguments to individual tcp_send commands.

The option -n prevents tcp_send from putting a newline at the end of the data strings.

The remaining options all behave as for tcp_read.

The data arguments are not further processed once they have been passed to tcp_send; they are simply passed down to print -r.

If the parameter $TCP_OUTPUT is a non-empty string and logging is enabled then the data sent to each session will be echoed to the log file(s) with $TCP_OUTPUT in front where appropriate, much in the manner of $TCP_PROMPT.

The first form creates an alias for a session name; alias can then be used to refer to the existing session sess. As many aliases may be listed as required.

The second form lists any aliases specified, or all aliases if none.

The third form deletes all the aliases listed. The underlying sessions are not affected.

The option -q suppresses an inconsistently chosen subset of error messages.

With the option -s, per-session logging is enabled. Input from tcp_read is output to the file logfile.sess. As the session is automatically discriminated by the filename, the contents are raw (no $TCP_PROMPT). The option -a applies as above. Per-session logging and logging of all data in one file are not mutually exclusive.

The option -c closes all logging, both complete and per-session logs.

The options -n and -N respectively turn off or restore output of data read by tcp_read to standard output; hence `tcp_log -cn' turns off all output by tcp_read.

The function is purely a convenient front end to setting the parameters $TCP_LOG, $TCP_LOG_SESS, $TCP_SILENT, which are described below.

With a sess argument, set the current session to sess. This is equivalent to changing $TCP_SESS directly.

With additional arguments, temporarily set the current session while executing `command arg ...'. command is re-evaluated so as to expand aliases etc., but the remaining args are passed through as that appear to tcp_sess. The original session is restored when tcp_sess exits.

This is generally only useful for interactive use, to prevent the display becoming fragmented by output returned from the connection. Within a programme or function it is generally better to handle reading data by a more explicit method.

Sessions are specified in the same way as tcp_read: the default is to use the current session, otherwise the sessions specified by -a, -s, or -l are used.

Each pattern is a standard zsh extended-globbing pattern; note that it needs to be quoted to avoid it being expanded immediately by filename generation. It must match the full line, so to match a substring there must be a `*' at the start and end. The line matched against includes the $TCP_PROMPT added by tcp_read. It is possible to include the globbing flags `#b' or `#m' in the patterns to make backreferences available in the parameters $MATCH, $match, etc., as described in the base zsh documentation on pattern matching.

Unlike tcp_read, the default behaviour of tcp_expect is to block indefinitely until the required input is found. This can be modified by specifying a timeout with -t or -T; these function as in tcp_read, specifying a per-read or overall timeout, respectively, in seconds, as an integer or floating-point number. As tcp_read, the function returns status 2 if a timeout occurs.

The function returns as soon as any one of the patterns given match. If the caller needs to know which of the patterns matched, the option -p var can be used; on return, $var is set to the number of the pattern using ordinary zsh indexing, i.e. the first is 1, and so on. Note the absence of a `$' in front of var. To avoid clashes, the parameter cannot begin with `_expect'. The index -1 is used if there is a timeout and 0 if there is no match.

The option -P var works similarly to -p, but instead of numerical indexes the regular arguments must begin with a prefix followed by a colon: that prefix is then used as a tag to which var is set when the argument matches. The tag timeout is used if there is a timeout and the empty string if there is no match. Note it is acceptable for different arguments to start with the same prefix if the matches do not need to be distinguished.

The option -q is passed directly down to tcp_read.

As all input is done via tcp_read, all the usual rules about output of lines read apply. One exception is that the parameter $tcp_lines will only reflect the line actually matched by tcp_expect; use $tcp_expect_lines for the full set of lines read during the function call.

The first argument is a TCP port on which the function will listen.

The remaining arguments give a command and its arguments to execute with standard input, standard output and standard error redirected to the file descriptor on which the TCP session has been accepted. If no command is given, a new zsh is started. This gives everyone on your network direct access to your account, which in many cases will be a bad thing.

The command is run in the background, so tcp_proxy can then accept new connections. It continues to accept new connections until interrupted.

The sessions may be selected explicitly with the standard -a, -s or -l options, or may be chosen implicitly. If none of the three options is given the rules are: first, if the array $tcp_spam_list is set, this is taken as the list of sessions, otherwise all sessions are taken. Second, any sessions given in the array $tcp_no_spam_list are removed from the list of sessions.

Normally, any sessions added by the `-a' flag or when all sessions are chosen implicitly are spammed in alphabetic order; sessions given by the $tcp_spam_list array or on the command line are spammed in the order given. The -r flag reverses the order however it was arrived it.

The -v flag specifies that a $TCP_PROMPT will be output before each session. This is output after any modification to TCP_SESS by the user-defined tcp_on_spam function described below. (Obviously that function is able to generate its own output.)

If the option -e is present, the line given as `cmd [ arg ... ]' is executed using eval, otherwise it is executed without any further processing.

An escape string, $TCP_TALK_ESCAPE, default `:', is used to allow access to normal shell operation. If it is on its own at the start of the line, or followed only by whitespace, the line editor returns to normal operation. Otherwise, the string and any following whitespace are skipped and the remainder of the line executed as shell input without any change of the line editor's operating mode.

The current implementation is somewhat deficient in terms of use of the command history. For this reason, many users will prefer to use some form of alternative approach for sending data easily to the current session. One simple approach is to alias some special character (such as `%') to `tcp_command --'.

To transfer a file from host woodcock to host springes, on springes:

tcp_point 8091 >output_file

and on woodcock:

tcp_shoot springes 8091 <input_file

As these two functions do not require tcp_open to set up a TCP connection first, they may need to be autoloaded separately.