The following example is a long dialog with the shell. Commands starting with > are inputs to the shell. All other lines are output from the shell.

strider 1> erl
Erlang (BEAM) emulator version 5.3 [hipe] [threads:0]

Eshell V5.3  (abort with ^G)
1> Str = "abcd".
"abcd"

Command 1 sets variable Str to string "abcd".

2> L = length(Str).
4

Command 2 sets L to the length of string Str.

3> Descriptor = {L, list_to_atom(Str)}.
{4,abcd}

Command 3 builds the tuple Descriptor, evaluating the BIF list_to_atom/1 .

4> L. 
4

Command 4 prints the value of variable L.

5> b().
Descriptor = {4,abcd}
L = 4
Str = "abcd"
ok

Command 5 evaluates the internal shell command b(), which is an abbreviation of "bindings". This prints the current shell variables and their bindings. ok at the end is the return value of function b().

6> f(L). 
ok

Command 6 evaluates the internal shell command f(L) (abbreviation of "forget"). The value of variable L is removed.

7> b().
Descriptor = {4,abcd}
Str = "abcd"
ok

Command 7 prints the new bindings.

8> f(L).
ok

Command 8 has no effect, as L has no value.

9> {L, _} = Descriptor.
{4,abcd}

Command 9 performs a pattern matching operation on Descriptor, binding a new value to L.

10> L.
4

Command 10 prints the current value of L.

11> {P, Q, R} = Descriptor.
** exception error: no match of right hand side value {4,abcd}

Command 11 tries to match {P, Q, R} against Descriptor, which is {4, abc}. The match fails and none of the new variables become bound. The printout starting with "** exception error:" is not the value of the expression (the expression had no value because its evaluation failed), but a warning printed by the system to inform the user that an error has occurred. The values of the other variables (L, Str, and so on) are unchanged.

12> P.
* 1: variable 'P' is unbound
13> Descriptor.
{4,abcd}

Commands 12 and 13 show that P is unbound because the previous command failed, and that Descriptor has not changed.

14>{P, Q} = Descriptor.
{4,abcd}
15> P.
4

Commands 14 and 15 show a correct match where P and Q are bound.

16> f().
ok

Command 16 clears all bindings.

The next few commands assume that test1:demo(X) is defined as follows:

demo(X) ->

    put(aa, worked),

    X = 1,

    X + 10.

17> put(aa, hello).
undefined
18> get(aa).
hello

Commands 17 and 18 set and inspect the value of item aa in the process dictionary.

19> Y = test1:demo(1).
11

Command 19 evaluates test1:demo(1). The evaluation succeeds and the changes made in the process dictionary become visible to the shell. The new value of dictionary item aa can be seen in command 20.

20> get().
[{aa,worked}]
21> put(aa, hello).
worked
22> Z = test1:demo(2).
** exception error: no match of right hand side value 1
     in function  test1:demo/1

Commands 21 and 22 change the value of dictionary item aa to hello and call test1:demo(2). Evaluation fails and the changes made to the dictionary in test1:demo(2), before the error occurred, are discarded.

23> Z.
* 1: variable 'Z' is unbound
24> get(aa).
hello

Commands 23 and 24 show that Z was not bound and that dictionary item aa has retained its original value.

25> erase(), put(aa, hello).
undefined
26> spawn(test1, demo, [1]).
<0.57.0>
27> get(aa).
hello

Commands 25, 26, and 27 show the effect of evaluating test1:demo(1) in the background. In this case, the expression is evaluated in a newly spawned process. Any changes made in the process dictionary are local to the newly spawned process and therefore not visible to the shell.

28> io:format("hello hello\n").
hello hello
ok
29> e(28).
hello hello
ok
30> v(28).
ok

Commands 28, 29 and 30 use the history facilities of the shell. Command 29 re-evaluates command 28. Command 30 uses the value (result) of command 28. In the cases of a pure function (a function with no side effects), the result is the same. For a function with side effects, the result can be different.

The next few commands show some record manipulation. It is assumed that ex.erl defines a record as follows:

-record(rec, {a, b = val()}).

val() ->

    3.

31> c(ex).
{ok,ex}
32> rr(ex).
[rec]

Commands 31 and 32 compile file ex.erl and read the record definitions in ex.beam. If the compiler did not output any record definitions on the BEAM file, rr(ex) tries to read record definitions from the source file instead.

33> rl(rec).
-record(rec,{a,b = val()}).
ok

Command 33 prints the definition of the record named rec.

34> #rec{}.
** exception error: undefined shell command val/0

Command 34 tries to create a rec record, but fails as function val/0 is undefined.

35> #rec{b = 3}.
#rec{a = undefined,b = 3}

Command 35 shows the workaround: explicitly assign values to record fields that cannot otherwise be initialized.

36> rp(v(-1)).
#rec{a = undefined,b = 3}
ok

Command 36 prints the newly created record using record definitions maintained by the shell.

37> rd(rec, {f = orddict:new()}).
rec

Command 37 defines a record directly in the shell. The definition replaces the one read from file ex.beam.

38> #rec{}.
#rec{f = []}
ok

Command 38 creates a record using the new definition, and prints the result.

39> rd(rec, {c}), A.
* 1: variable 'A' is unbound
40> #rec{}.
#rec{c = undefined}
ok

Command 39 and 40 show that record definitions are updated as side effects. The evaluation of the command fails, but the definition of rec has been carried out.

For the next command, it is assumed that test1:loop(N) is defined as follows:

loop(N) ->

    io:format("Hello Number: ~w~n", [N]),

    loop(N+1).

41> test1:loop(0).
Hello Number: 0
Hello Number: 1
Hello Number: 2
Hello Number: 3

User switch command
 --> i
 --> c
.
.
.
Hello Number: 3374
Hello Number: 3375
Hello Number: 3376
Hello Number: 3377
Hello Number: 3378
** exception exit: killed

Command 41 evaluates test1:loop(0), which puts the system into an infinite loop. At this point the user types ^G (Control G), which suspends output from the current process, which is stuck in a loop, and activates JCL mode. In JCL mode the user can start and stop jobs.

In this particular case, command i ("interrupt") terminates the looping program, and command c connects to the shell again. As the process was running in the background before we killed it, more printouts occur before message "** exception exit: killed" is shown.

42> E = ets:new(t, []).
17

Command 42 creates an ETS table.

43> ets:insert({d,1,2}).
** exception error: undefined function ets:insert/1

Command 43 tries to insert a tuple into the ETS table, but the first argument (the table) is missing. The exception kills the evaluator process.

44> ets:insert(E, {d,1,2}).
** exception error: argument is of wrong type
     in function  ets:insert/2
        called as ets:insert(16,{d,1,2})

Command 44 corrects the mistake, but the ETS table has been destroyed as it was owned by the killed evaluator process.

45> f(E).
ok
46> catch_exception(true).
false

Command 46 sets the exception handling of the evaluator process to true. The exception handling can also be set when starting Erlang by erl -stdlib shell_catch_exception true.

47> E = ets:new(t, []).
18
48> ets:insert({d,1,2}).
* exception error: undefined function ets:insert/1

Command 48 makes the same mistake as in command 43, but this time the evaluator process lives on. The single star at the beginning of the printout signals that the exception has been caught.

49> ets:insert(E, {d,1,2}).
true

Command 49 successfully inserts the tuple into the ETS table.

50> halt().
strider 2>

Command 50 exits the Erlang runtime system.

When the shell starts, it starts a single evaluator process. This process, together with any local processes that it spawns, is referred to as a job. Only the current job, which is said to be connected, can perform operations with standard I/O. All other jobs, which are said to be detached, are blocked if they attempt to use standard I/O.

All jobs that do not use standard I/O run in the normal way.

The shell escape key ^G (Control G) detaches the current job and activates JCL mode. The JCL mode prompt is "-->". If "?" is entered at the prompt, the following help message is displayed:

--> ?
c [nn]            - connect to job
i [nn]            - interrupt job
k [nn]            - kill job
j                 - list all jobs
s [shell]         - start local shell
r [node [shell]]  - start remote shell
q                 - quit erlang
? | h             - this message

The JCL commands have the following meaning:

The behavior of shell escape can be changed by the STDLIB application variable shell_esc. The value of the variable can be either jcl (erl -stdlib shell_esc jcl) or abort (erl -stdlib shell_esc abort). The first option sets ^G to activate JCL mode (which is also default behavior). The latter sets ^G to terminate the current shell and start a new one. JCL mode cannot be invoked when shell_esc is set to abort.

If you want an Erlang node to have a remote job active from the start (rather than the default local job), start Erlang with flag -remsh, for example, erl -sname this_node -remsh other_node@other_host

The shell can be started in a restricted mode. In this mode, the shell evaluates a function call only if allowed. This feature makes it possible to, for example, prevent a user from accidentally calling a function from the prompt that could harm a running system (useful in combination with system flag +Bi).

When the restricted shell evaluates an expression and encounters a function call or an operator application, it calls a callback function (with information about the function call in question). This callback function returns true to let the shell go ahead with the evaluation, or false to abort it. There are two possible callback functions for the user to implement:

These callback functions are called from local and non-local evaluation function handlers, described in the erl_eval manual page. (Arguments in ArgList are evaluated before the callback functions are called.)

Argument State is a tuple {ShellState,ExprState}. The return value NewState has the same form. This can be used to carry a state between calls to the callback functions. Data saved in ShellState lives through an entire shell session. Data saved in ExprState lives only through the evaluation of the current expression.

There are two ways to start a restricted shell session:

Notes:

Errors when loading the callback module is handled in different ways depending on how the restricted shell is activated:

The default shell prompt function displays the name of the node (if the node can be part of a distributed system) and the current command number. The user can customize the prompt function by calling prompt_func/1 or by setting application configuration parameter shell_prompt_func for the STDLIB application.

A customized prompt function is stated as a tuple {Mod, Func}. The function is called as Mod:Func(L), where L is a list of key-value pairs created by the shell. Currently there is only one pair: {history, N}, where N is the current command number. The function is to return a list of characters or an atom. This constraint is because of the Erlang I/O protocol. Unicode characters beyond code point 255 are allowed in the list. Notice that in restricted mode the call Mod:Func(L) must be allowed or the default shell prompt function is called.