The messages formed by the plumb(2) library are formatted for
transmission between processes into textual form, using newlines
to separate the fields. Only the data field may contain embedded
newlines. The fields occur in a specified order, and each has
a name, corresponding to the elements of the Plumbmsg
structure, that is used in the plumbing rules. The fields, in
At the moment, only textual data (type=text) is supported.
All fields are optional, but type should usually be set since
it describes the form of the data, and ndata must be an accurate
count (possibly zero) of the number of bytes of data. A missing
field is represented by an empty line.
src application/service generating message|
dst destination `port' for message
wdir working directory (used if data is a file name)
type form of the data, e.g. text
attr attributes of the message, in name=value pairs separated by
white space (the value must follow the usual quoting convention
if it contains white space or quote characters or equal signs;
it cannot contain a newline)
ndata number of bytes of data
data the data itself
The file is a sequence of rule sets, each of which is a set of
one–line rules called patterns and actions. There must be at least
one pattern and one action in each rule set. (The only exception
is that a rule set may contain nothing but plumb to rules; such
a rule set declares the named ports but has no other effect.)
blank line terminates a rule set. Lines beginning with a # character
are commentary and are regarded as blank lines.
A line of the form
The plumber (see plumb(2)) receives messages on its send port
(applications send messages there), interprets and reformats them,
and (typically) emits them from a destination port. Its behavior
is determined by a plumbing rules file, default /usr/$user/lib/plumbing,
which defines a set of pattern/action
rules with which to analyze, rewrite, and dispatch received messages.
substitutes the contents of file for the line, much as in a C
#include statement. Unlike in C, the file name is not quoted.
If file is not an absolute path name, or one beginning ./ or ../,
file is looked for first in the directory in which the plumber
is executing, and then in /sys/lib/plumb.
When a message is received by the plumber, the rule sets are examined
in order. For each rule set, if the message matches all the patterns
in the rule set, the actions associated with the rule set are
triggered to dispose of the message. If a rule set is triggered,
the rest are ignored for this message. If none is
triggered, the message is discarded (giving a write error to the
sender) unless it has a dst field that specifies an existing port,
in which case the message is emitted, unchanged, from there.
Patterns and actions all consist of three components: an object,
a verb, and arguments. These are separated by white space on the
line. The arguments may contain quoted strings and variable substitutions,
described below, and in some cases contain multiple words. The
object and verb are single words from a pre–
The object in a pattern is the name of an element of the message,
such as src or data, or the special case arg, which refers to
the argument component of the current rule. The object in an action
is always the word plumb.
The verbs in the pattern rules describe how the objects and arguments
are to be interpreted. Within a rule set, the patterns are evaluated
in sequence; if one fails, the rule set fails. Some verbs are
predicates that check properties of the message; others rewrite
components of the message and implicitly always succeed.
Such rewritings are permanent, so rules that specify them should
be placed after all pattern–matching rules in the rule set.
The matches verb has special properties that enable the rules
to select which portion of the data is to be sent to the destination.
By default, a data matches rule requires that the entire text
matches the regular expression. If, however, the message has an
attribute named click, that reports that the message was
produced by a mouse click within the text and that the regular
expressions in the rule set should be used to identify what portion
of the data the user intended. Typically, a program such as an
editor will send a white–space delimited block of text containing
the mouse click, using the value of the click attribute (a
number starting from 0) to indicate where in the textual data
the user pointed.
When the message has a click attribute, the data matches rules
extract the longest leftmost match to the regular expression that
contains or abuts the textual location identified by the click.
For a sequence of such rules within a given rule set, each regular
expression, evaluated by this specification, must
match the same subset of the data for the rule set to match the
message. For example, here is a pair of patterns that identify
a message whose data contains the name of an existing file with
a conventional ending for an encoded picture file:
add The object must be attr. Append the argument, which must be
a sequence of name=value pairs, to the list of attributes of the
delete The object must be attr. If the message has an attribute
whose name is the argument, delete it from the list of attributes
of the message. (Even if the message does not, the rule matches
is If the text of the object is identical to the text of the argument,
the rule matches.
isdir If the text of the object is the name of an existing directory,
the rule matches and sets the variable $dir to that directory
isfile If the text of the object is the name of an existing file
(not a directory), the rule matches and sets the variable $file
to that file name.
matchesIf the entire text of the object matches the regular expression
specified in the argument, the rule matches. This verb is described
in more detail below.
set The value of the object is set to the value of the argument.
The first expression extracts the largest subset of the data around
the click that contains file name characters; the second sees
if it ends with, for example, .jpeg. If only the second pattern
were present, a piece of text horse.gift could be misinterpreted
as an image file named horse.gif.
If a click attribute is specified in a message, it will be deleted
by the plumber before sending the message if the data matches
rules expand the selection.
The action rules all have the object plumb. There are only three
verbs for action rules:
data matches '[a–zA–Z0–9_–./]+'|
data matches '([a–zA–Z0–9_–./]+).(jpe?g|gif|bit|ps|pdf)'
The arguments to all rules may contain quoted strings, exactly
as in rc(1). They may also contain simple string variables, identified
by a leading dollar sign $. Variables may be set, between rule
sets, by assignment statements in the style of rc. Only one variable
assignment may appear on a line. The plumber also
maintains some built–in variables:
to The argument is the name of the port to which the message will
be sent. If the message has a destination specified, it must match
the to port of the rule set or the entire rule set will be skipped.
(This is the only rule that is evaluated out of order.)|
client If no application has the port open, the arguments to a
plumb start rule specify a shell program to run in response to
the message. The message will be held, with the supposition that
the program will eventually open the port to retrieve it.
start Like client, but the message is discarded. Only one start
or client rule should be specified in a rule set.
$0 The text that matched the entire regular expression in a previous
data matches rule. $1, $2, etc. refer to text matching the first,
second, etc. parenthesized subexpression.|
$attr The textual representation of the attributes of the message.
$data The contents of the data field of the message.
$dir The directory name resulting from a successful isdir rule.
If no such rule has been applied, it is the string constructed
syntactically by interpreting data as a file name in wdir.
$dst The contents of the dst field of the message.
$file The file name resulting from a successful isfile rule. If
no such rule has been applied, it is the string constructed syntactically
by interpreting data as a file name in wdir.
$type The contents of the type field of the message.
$src The contents of the src field of the message.
$wdir The contents of the wdir field of the message.