esh, the easy shell.
esh was primarily written out of a need for a simple and lightweight shell
for Unix. As such, it deviates completely from all of the traditional
shells, opting instead for a Lisp-like syntax. This allows exceptionally
small size, both in terms of lines of code and memory consumption, while
retaining remarkable flexibility and programmability.
esh
To start the shell, simply type esh. There are no command-line
parameters. However, since all the command-line arguments are available
to the shell programmer, it is a simple matter to write your own argument
processing routine and place it in your .eshrc file.
Both /etc/eshrc and the .eshrc in your home directory will
be run by the shell on startup, if they exist.
If the shell is running interactively, you will get a prompt, $ by
default. Since the shell uses the readline library, you can use line-editing
keystrokes, the history buffer, filename completion, and the rest of the
goodies provided by the readline library.
To run a single executable, simply type it in as you would in any other
shell. Example: /usr/games/fortune -m Unix.
To run a pipeline, type several commands separated by commas or vertical
lines. Example:
cat /var/log/messages, grep pppd, grep "IP address", xmessage -file -.
To run a pipeline with file redirection, place the redirection symbols and the
filenames after the pipeline. Example:
sort, uniq < names.raw > names.sorted. In this case, the contents of
names.raw are used as the standard input of sort, and the output
of the whole pipeline is saved to names.sorted. The order in which
the redirection symbols are given does not matter, but a valid filename
should always follow a redirection symbol.
Note one major pitfall: cd, bg, fg, etc., are all
builtin shell commands! For info on running shell commands, see the
next section. (see section 2.3 Simple Programming)
All shell commands are specified using a parenthetical syntax
very similar to that of Lisp or Scheme. Example: Use (cd /var) to change
the current directory. In this case, cd is the name of the command,
and /var is the argument. Multiple arguments to the same
command are separated by white space. Example: (set HOME /home/ivan).
Builtin commands never accept options in the
style of compiled programs. Also note that quotes around string values are
not required, though they are certainly allowed at any time if you wish
to escape special characters. (In fact, using quotes is the only way to
do this. Backslash escape sequences are not recognized.) Example:
(cd "funny,directory name()"). In this case, the whole string inside
the quotes is used as the argument verbatim. Single quotes and double
quotes are equivalent.
If a command returns anything, the returned data will be printed
by the shell. Example: (get HOME) => /home/ivan.
Note that you can certainly use the return values of one command as input
to another. Example: (cd (get HOME))
If you want to pass a list instead of a string as an argument to a command,
quote the list with a tilde. Example:
(run-simple ~(ls --color=yes)). Here, the run command is given
a list as an argument.
Note that a sublist of a quoted list is also quoted; that is,
~(1.1 (2.1 2.2) 1.2) is a list of three elements, the first and last
are strings and the middle element is a list.
See section 2.6 Basic Usage, for an explanation of what cd, get,
set, and run do.
If the shell is started non-interactively -- for example, when the standard
input to the shell is a file -- the shell will only accept commands.
To run executables and construct pipes, simply use the run
command.
(see section 2.6 Basic Usage)
If the shell is running interactively, you will have access to job control primitives.
To stop a job in the foreground, simply type C-z, as in other shells.
To bring the last known job into foreground or background, issue the fg
or bg command, respectively, without arguments.
(see section 2.3 Simple Programming)
To get a listing of all the currently known jobs, issue the jobs
command, also without arguments.
If the fg or bg command is given a single numeric argument, it
will act on the job number specified by that argument. To find out the job
number for a command, simply issue the jobs command.
Note: the job number is not the PID!
(cd [string]) Change the current directory to the given
argument. If the directory name is a single dash, the previous directory
will be used. (i.e. the value of the environment variable OLDPWD)
If the command is called without any arguments, change to the home directory.
(jobs) List the current jobs.
(fg [number]) If the optional numeric argument is given, bring
the job number given by the argument into the foreground. If this argument
is omitted, bring the first job into the foreground.
(bg [number]) Same as fg, but puts a job into the background.
(run <bool> <file> <file> <list>...) Run a pipeline.
The first argument should be false if you want the command to be
brought into the foreground, or true if otherwise.
The next two arguments are the input and output redirection files,
respectively. You can use (standard) for standard input/output.
The rest of the arguments are the commands you want to run,
given as lists.
If the command was run in the background, then the return value
is the PID of the last executable in the pipeline. Otherwise, the return
value is the exit status of the pipeline.
Examples:
(run (false) (standard) (standard) ~(/usr/games/fortune -m Unix))
is equivalent to
/usr/games/fortune -m Unix.
(run (false) (file-open file "names.raw") (file-open file "names.sorted") ~(sort) ~(uniq)) is equivalent to
sort, uniq < names.raw > names.sorted.
(run-simple <list>...) Equivalent to
(run (false) (standard) (standard) ...).
This command returns the exit status of the pipeline.
(get <string>) Return the environment variable named by the given
argument.
(set <string> <string>) Set the environment variable named by the first
string to be the second string.
(env) List all environment variables.
(print ...) Print the given arguments on the standard output in
a human-readable form.
(+ <number>...) Add all the arguments.
(- <number>...) Subtract the arguments following the first argument
from the first argument.
(* <number>...) Multiply all the arguments.
(/ <number>...) Divide the first argument by the rest of the arguments.
(< <number> <number>) Return true if the first number is
less than the second; otherwise, return false.
(> <number> <number>) Return true if the first number is
less than the second; otherwise, return false.
(if <any> <any> <any>) If the \"eval\" of the
first argument is false, execute eval on the second argument.
Otherwise, execute eval on the third argument. Note that the
arguments must be quoted right! (see section 3.3 Quoting Trickery for
more info on that.)
(list ...) Compose a list made of the given arguments and return it.
(eval ...) If an argument is a string or a hash table, simply return it.
Otherwise, execute the given list as if it was a command.
For example, typing in
(eval ~(print Hello World!)) is equivalent to
(print Hello World!). However, you must make sure that the list given
to eval is quoted properly in order to delay evaluation
until the right moment. (see section 3.3 Quoting Trickery for more info.)
(= <string> <string>) Compare the two strings and return an empty
list if they are not equal. Otherwise, simply return the first string.
A programmer's manual and a tutorial.
Here is a description of the syntax of esh. Note, however, that
this grammar is only a descriptive tool, since the actual parser in the
shell is written by hand.
Also note that using the interactive-command syntax is not allowed when the shell is not started interactively.
openparen-symbol = '('
closeparen-symbol = ')'
delay-symbol = '~' | '$'
pipe-symbol = ',' | '|'
redirection-symbol = '<' | '>'
script = /* Nothing */ |
list script
list = openparen-symbol list-elements closeparen-symbol
list-elements = /* Nothing */ |
string list-elements |
delay-symbol list list-elements |
list list-elements
interactive-command = list |
interactive-pipeline
interactive-pipeline = /* Nothing */ |
interactive-pipe redirection redirection
interactive-pipe = /* Nothing */ |
single-command pipe-symbol interactive-pipe
single-command = /* Nothing */ |
string single-command
redirection = /* Nothing */ |
redirection-symbol string
One major deviation of esh from other programming languages is
that commands are not functions. Any command can return any number of
values, without explicit unpacking syntax. (e.g. Python)
For example, (rest ~(foo bar baz)) => bar baz.
These returned elements are not a list! Therefore, this code
will produce an error: (rest (rest ~(foo bar baz))). Instead, you
should write: (rest (list (rest ~(foo bar baz)))).
The first, most important concept is the define command. This command
allows you to define new commands, which are no different from builtin
commands as far as the programmer is concerned.
The syntax of define is simple: (define <string> ...).
The first argument is simply the name of the command you are defining, and
the rest of the arguments will simply be passed to eval whenever
your new command is called.
It's important to realize that this code snippet
(define foo ~(print bar) bar) (foo) is identical to
this one: (eval ~(print bar) bar). Therefore, anything that
applies to eval also applies to commands defined by define.
(see section 3.3 Quoting Trickery for more on that.)
If you are familiar with Scheme or Lisp, you'll notice the lack of
argument passing information in the syntax of define. The reason for
that is that the argument passing convention is radically different in
esh from other Lisp-like languages.
Every command in esh has a stack all to itself for scratch space
and local variables. When a user-defined command is run, all the arguments
are simply placed on the local-variable stack, the first argument at the
top.
To access the local-variable stack, several commands can be used.
(push <any>) will push a value onto the stack, (pop) will
remove the top element of the stack and return it, (top) will return
a copy of the top element, (stack) will return a copy of the whole
stack, and finally (rot) will switch the top and the next-to-top
elements of the stack and return a copy of the element that just became
the top one.
The top-level of the interpreter also has a stack; that is, typing the stack manipulation commands directly into the interpreter will produce the expected results.
Note that commands called recursively do not inherit the stack of the calling command.
Also note that (stack) does not return a list, it returns an
arbitrary number of elements. If you find typing (list (stack))
frequently, you can instead use the (l-stack) command, since they are
identical.
The shell does not support looping constructs in the traditional sense -- like in Scheme, looping is accomplished by using recursion. Here is a concrete example:
(define print-squish
~(if ~(not-null? (top))
~(begin
(print (squish (pop)))
(print-squish (stack)))
()))
The command begin simply returns the given arguments; its purpose
is to allow the use of several commands where one is asked for.
(see section 3.4 Command List for the syntax of if)
Since esh 0.2, there is an explicit true and false
value; these values are different from all other possible values. Commands
that operate on predicates (if, and, or, =, etc.)
use these values extensively. You can use them in your own scripts with
the true and false commands. Note that if, and,
and or do not explicitly check for true, so to these commands
any value that isn't false is "true", so to say.
(Note: Since esh 0.6, the backslash quote has been removed.
Use the tilde quote instead.)
Note that the meaning of the quote syntax is much different in esh
than in Lisp -- in Lisp, the single quote is a syntactic trick to
allow inputting lists and symbols as literals. In esh, the tilde
quote has a very clear semantic meaning -- whenever a list is marked with
a tilde, it is marked as such throughout it's life. Commands such as
eval can then use this information to see which lists were supposed
to be evaluated, and which ones were meant to be left alone.
In other words, any list marked with a tilde has a numeric flag set on it, which is preserved on the list until the list is deleted.
Also, nested tildes increase the "strength" of the tilde, so to say.
eval will not evaluate lists which have a "strength" greater than
itself, and calling eval recursively will increase it's strength.
This may sound confusing, but it has a simple intuitive meaning:
(eval ~(foo ~(bar baz)))
Here, as you would expect, eval will only evaluate foo, and
pass (bar baz) as a list to foo.
Note that unlike in Lisp, esh has no "special forms"! A command
that works on lists is indistinguishable from a command that works
on code. That's why there is no lambda command, and also why
arguments to if and firends need to be quoted with a tilde.
(alias <string> <string>...)
Create an alias. The first argument is the name of the alias, and the
rest of the arguments is the expansion. For example, (alias rm rm -i)
will create an alias called rm, so that any time rm is run,
rm -i gets executed in actuality.
(alias <string> <list>...)
Mimic an executable with a command. Whenever an executable named by
the first string is run, the lists following get evaluated, and
their output is thrown away. This process will only happen during the
parsing stage, though. That is, this type of alias will only be enabled
when typing commands interactively into te shell.
Example: (alias cd ~(cd (stack))) to mimic the
traditional syntax of cd.
(alias-hash)
Return the alias hash table. Modifying this table also modifies the
actual aliases.
(alive? <PID>)
Return true if the given process is alive, and false otherwise.
(and ...) Return false if any of the arguments evaluate to
false. Warning: the inputs need to be quoted with a tilde, since this
command implements a short-circuited "and"!
For example, (and ~(under-attack) ~(launch-nukes)) is safe to use.
If the "eval" of ~(under-attack) evaluates to false
~(launch-nukes) will never be evaluated.
(begin ...)
Simply copy the inputs to the outputs. This command is useful when you want
to use several commands as if they were a single command. Example:
(if ~(check-me)
~(begin (print-message)
(set-environment-magic))
())
(begin-last ...) Evaluate the given arguments sequentially, and
return the value of last argument.
(if ~(not-null? (top))
~(begin (push foo)
(print (stack))
(pop))
())
(builtin <string> ...) Execute the builtin command named by the
first argument, even if this name has already been defined.
For example, here's how to write a custom replacement for cd:
(define cd
~(push (squish (get HOME) "/.cdlog"))
~(file-write (file-open file (top))
(squish (file-read (file-open file (pop)))
(stack) (nl)))
~(builtin cd (stack)))
(car <list>) Simply return the first element of the list.
(car-l ...) Return the first element.
(cdr <list>) Return the elements of the list after the first one.
(chars <string>) Return the charcters in the given string.
(chars "foo bar") => 'f' 'o' 'o' ' ' 'b' 'a' 'r'
(chop! <string>) Delete the last character in the given string,
and return the given string.
Note: this function modifies the string in-place!
(chop-nl! <string>) Like chop!, except that it only deletes
the last character if it is a newline.
(clone <string> <number>) Return the first argument X number of times,
where X is the numeric value of the second argument.
(copy) Equivalent to begin.
(define <string> ...) Define a new command, named by the
first argument. The rest of the arguments will be passed to eval
when the newly defined command is called. Example:
(define print-nl
~(print (stack) (nl)))
(defined? <string>) Return false if the given string has not
been defined as a command.
(exec <list> ..) Equivalent to eval, except that the stack will
be set to the first argument for the duration of evaluation.
(exit [number]) Exit with the given exit status, or zero if the
command is called with no arguments.
(false ...) Return false.
(file-open <string> <string>) Open a file. The first argument specifies
the type of the file, and the second the name of the file you want to open.
Allowed values for the first argument:
"file" Open a regular file for reading/writing.
"truncate" Open a regular file for reading/writing, but truncate
it first.
"append" Open a regular file for reading/appending.
"string" Simulate a file with a string variable. In this case,
the name is used as the initial contents of the buffer.
(Note: "string" files are implemented as pipes internally.)
(file-read <file>) Read the entire file given by the argument
into a single string, and return the string. Warning: when reading
the output of a pipeline, you may have to call this command several times.
(file-type <string>) Return a string describing what type of file
is named by the given string. If such a file does not exist, return an empty
list. Otherwise, one of these strings is returned:
"link" File is a symbolic link.
"regular" File is a regular file.
"directory" File is a directory.
"character" File is a character device.
"block" File is a block device.
"pipe" File is a FIFO pipe.
"socket" File is a socket.
(file-write <file> <string>) Write the second argument into the
first one.
(filter <string> <list>) Apply the second argument to every character
of the first argument. This command does not change the original string.
(filter "string with spaces"
~(if ~(= (top) ' ')
'_'
~(top)))
returns "string_with_spaces".
(first <list>) Equivalent to car.
(first-l ...) Equivalent to car-l.
(gobble <file> <list>...) Equivalent to run, except that the
output of the pipeline will be returned, as a string.
(hash-get <hash> <string>) Return the element in the given hash table
corresponding to the given key.
(hash-keys <hash>) Return all the keys in the given hash table.
(hash-make) Return a new, empty hash table.
(hash-put <hash> <string> ...) Associate the arguments after the
second one with the given key in the given hash table.
(help) Show version number and all builtin commands.
(interactive?) Return false if the shell has not been
started interactively.
(l-cdr <list>) Equivalent to (list (cdr ...)).
(l-rest <list>) Equivalent to l-cdr.
(l-stack) Equivalent to (list (stack)).
(newline) Return a newline character.
(nl) Equivalent to newline.
(not <bool>) Return false if the argument is true.
(not-null? <any>) Return false if the argument is an empty list.
(null ...) Return an empty list.
(null? <any>) Return true if the argument is an empty list.
(match <string> <string>) Return true if the second argument
matches the first. The first argument is a regular expression.
(or ...) Return false if all the argument evaluate to
false. As with the command and, this command is also
short-circuited. See the description of and for an explanation of why
arguments must be quoted with a tilde.
(parse <string>) Parse the given string as if it was typed into the
shell.
(pop) Excise the top element from the stack, and return it.
(prompt ...) Run the given arguments through (squish (eval ...))
to produce the prompt. This command need only be run once.
(push <any>) Push the argument onto the stack.
(read <string>) Read a line of input from the user, using the given
string as a prompt. Warning: this command only works if the shell is running
interactively.
(repeat <string> ...) Repeat the arguments after the first argument
n number of times, where n is the numeric value of the first
argument.
(rest <list>) Equivalent to cdr.
(reverse ...) Return the given arguments in reverse order.
(rot) Switch the top and next-to-top elements of the stack, and return
the element that just became the top one.
(script <string>) Execute the given filename as a shell script.
(split <string>...) Separate the given string into words. If there are
more than one arguments given, then use the arguments after the first one as
field separators. Otherwise, assume that fields are separated by whitespace.
Note that splitting is more complex than you'd think at first glance --
(split ":.:.foo:.:bar.:baz." ":" ".") => foo bar baz.
In other words, extra field separators are ignored.
(squish ...) Concatenate any string values into a single string. List
structures will be ignored. For example:
(squish ~(foo (bar (baz)))) => foobarbaz.
(stack) Return all the stack elements, top one first.
(standard) Return the standard input/output.
(stderr) Return the standard input/error.
(stderr-handler <file>) Define a standard error handler for
all new subprocesses. This means that from now on, all executables run
from the shell will send their standard error to the given file.
(substring? <string> <string>) Return true if the first argument
is a substring of the second.
(top) Return a copy of the top element of the stack.
(true ...) Return true.
(typecheck <string> ...) Make sure that the given arguments match the
given type specification string. (see section 3.5 Tutorial for a description of this
type specification string.)
(unlist <list>) Return the elements of the given list.
(version) Return the version of the shell, as three numbers.
(void ...) Return nothing at all. Note: This is different from
null! For example:
(if (null (magic-sequence))
~(my-predicate)
true
false)
is an error, since in this case if is given four arguments.
(null returns an empty list.)
On the other hand,
(if (void (magic-sequence))
~(my-predicate)
true
false)
is perfectly fine since void returns nothing at all whatsoever!
(while <list> <list> ...) eval the second argument as long
as the eval of the first argument is not false. The rest of
the arguments are used as the stack for the duration of this command's
execution. Warning: this command is an efficiency hack. It saves the
waste of creating a huge return value which would be inherent in a recursive
command. Use while only if you don't care about the return value of
the given commands.
Example:
(define while-rec
~(if ~(condition)
~(begin (action)
(while-rec))
()))
will return a value equivalent to (action) (action) (action) ....
This value could be a large memory and time waste. Therefore, when you are
looping over large lists and you don't care about return values, use
while.
For starters, let us write a simple command to iterate through a list:
(define for-each
~(if ~(not-null? (rot))
~(begin ((rot) (rot))
(for-each (cdr (list (stack)))))
()))
Using this command is simple:
(define starrify
~(squish '*' (top) '*'))
(for-each starrify foo bar baz)
results in *foo* *bar* *baz*.
Several notes to keep in mind: notice the tricky use of rot to
shuffle stack elements. If you don't remember, rot switches the top
and the next-to-top elements of the stack, and returns a copy of the element
that just became the top one.
Think of it as such: the first call to rot brings the second argument
to the front and checks that the stack is not empty, at the same time.
The second call to rot returns the first argument, and the third
call to rot returns the second element and undoes the previous call
of rot at the same time. At this point, the second argument is still
on top of the stack.
Finally, when for-each is recursively called, it is given only
below the top stack element as arguments. Effectively, the second element
was excised from the stack.
Also note that ((rot) (rot)) is calling a command, even though the
name of the command is not explicit.
Now for another example. Suppose that you want to remind yourself which
directory serves which purpose, and you'd like a descriptive string to
be listed along with a directory name in the prompt.
(i.e. [/home/ivan/src/xcf => Backup of GIMP artwork]$ )
(prompt ~(push (get PWD))
~(push (hash-get (dir-names) (top)))
"["
~(rot)
~(if ~(not-null? (rot))
~(squish " => " (top))
"")
"]$ "
~(null (pop))
~(null (pop)))
The meaning of prompt is simple -- whatever arguments are given to
it are passed through eval and squish to become the string
specifying the prompt. Again, as far as quoting is concerned, prompt
is identical to eval.
Also, the command null simply returns an empty list.
To finish the job, you should insert the following code into your
.eshrc:
(define dir-names (hash-make)) (hash-put (dir-names) "/home/ivan/src/xcf" "Backup of GIMP artwork")
This is a simple script that illustrates simple file I/O:
(define prepend
~(push (file-read (file-open file (pop))))
~(push (squish (pop)
(file-read (file-open file (top)))))
~(file-write (file-open truncate (rot)) (rot)))
A script such as this was used to append a header to every source code file
in the esh distribution. Note the typecheck command; it is
very useful for insuring the consistency of arguments to commands.
The only tricky part about using typecheck is the syntax of the
first argument. Here is an explanation:
s Make sure that the next argument is a single string.
l Make sure that the next argument is a single list.
h Make sure that the next argument is a single hash table.
b Make sure that the next argument is a single boolean.
f Make sure that the next argument is a single file.
p Make sure that the next argument is a PID.
S Match any number of strings.
L Match any number of lists.
H Match any number of hash tables.
B Match any number of booleans.
F Match any number of files.
P Match any number of PID's.
? Match any one element.
* Match any number of any elements.
( Match a list only if the sublist passes typechecking on the
string after the parentheses.
) Match end-of-list.
For example, "H(ss)s" will match any list of arguments that begin
with an arbitrary number of hash tables, followed by a list of two strings,
and ends with a single string.
Note that in esh there is no numeric or "symbol" type -- all scalars are
either strings, booleans, files, or process ID's. Most importantly,
there is no "procedure" type -- hence, no lambda command. From the point
of view of the interpreter, a command and a list are indistinguishable.
Don't be mislead by the naming of define -- it is actually equivalent to
a Lisp-like macro definition.
Also, car and friends are more complicated because commands in
esh
can return any number of arguments. This is why (car (split "foo bar"))
doesn't work -- car expects a list, while split returns
an arbitrary number of elements. In a case like this, it is more convenient
to write (car-l (split "foo bar")) instead, which is equivalent
to (car (list (split "foo bar"))).
The most obvious difference is the syntax -- other shells normally don't
make much of a distinction between commands from a disk executable and
the shell interpreter's builtin commands. Not so in esh.
While other shell normally operate by complicated string substitution
and matching rules, esh mainly uses command definitions and
recursion. In that sense, it is more similar to a "real" programming
language. esh also supports more complicated data structures --
lists and hash tables, for example. However, string operations are
lacking in comparison to other shells.
esh is also more verbose and formal, though this could be beneficial
when you're trying to compose large libraries of useful routines.
(It's possible to write shell "modes", much as in emacs, when using
esh.)
esh has a more generalized support for files. Instead of limiting
file I/O to redirection only, the same commands can be used either on pipes
or on files directly. (e.g. in the future, it may be possible to use a
network socket as the output of a pipeline.)
Also, traditional shells have "exported" and "non-exported" environment
variables, whereas in esh all variables are exported. That is,
there's no setenv command, there's just set.
Jump to: * - + - - - / - = - a - b - c - d - e - f - g - h - i - j - l - m - n - o - p - r - s - t - u - v - w
Jump to: . - a - b - c - d - e - f - g - i - j - l - n - o - p - q - r - s - t - v
eval, quoting for
esh
esh
This document was generated on 15 January 2000 using texi2html 1.56k.