8 Expressions

In this chapter, all valid Erlang expressions are listed. When writing Erlang programs, it is also allowed to use macro- and record expressions. However, these expressions are expanded during compilation and are in that sense not true Erlang expressions. Macro- and record expressions are covered in separate chapters: Macros and Records.

8.1 Expression Evaluation

All subexpressions are evaluated before an expression itself is evaluated, unless explicitly stated otherwise. For example, consider the expression:

Expr1 + Expr2

Expr1 and Expr2, which are also expressions, are evaluated first - in any order - before the addition is performed.

Many of the operators can only be applied to arguments of a certain type. For example, arithmetic operators can only be applied to numbers. An argument of the wrong type will cause a badarg run-time error.

8.2 Terms

The simplest form of expression is a term, that is an integer, float, atom, string, list, map or tuple. The return value is the term itself.

8.3 Variables

A variable is an expression. If a variable is bound to a value, the return value is this value. Unbound variables are only allowed in patterns.

Variables start with an uppercase letter or underscore (_) and may contain alphanumeric characters, underscore and @. Examples:

X
Name1
PhoneNumber
Phone_number
_
_Height

Variables are bound to values using pattern matching. Erlang uses single assignment, a variable can only be bound once.

The anonymous variable is denoted by underscore (_) and can be used when a variable is required but its value can be ignored. Example:

[H|_] = [1,2,3]

Variables starting with underscore (_), for example _Height, are normal variables, not anonymous. They are however ignored by the compiler in the sense that they will not generate any warnings for unused variables. Example: The following code

member(_, []) ->
    [].

can be rewritten to be more readable:

member(Elem, []) ->
    [].

This will however cause a warning for an unused variable Elem, if the code is compiled with the flag warn_unused_vars set. Instead, the code can be rewritten to:

member(_Elem, []) ->
    [].

Note that since variables starting with an underscore are not anonymous, this will match:

{_,_} = {1,2}

But this will fail:

{_N,_N} = {1,2}

The scope for a variable is its function clause. Variables bound in a branch of an if, case, or receive expression must be bound in all branches to have a value outside the expression, otherwise they will be regarded as 'unsafe' outside the expression.

For the try expression introduced in Erlang 5.4/OTP-R10B, variable scoping is limited so that variables bound in the expression are always 'unsafe' outside the expression. This will be improved.

8.4 Patterns

A pattern has the same structure as a term but may contain unbound variables. Example:

Name1
[H|T]
{error,Reason}

Patterns are allowed in clause heads, case and receive expressions, and match expressions.

Match Operator = in Patterns

If Pattern1 and Pattern2 are valid patterns, then the following is also a valid pattern:

Pattern1 = Pattern2

When matched against a term, both Pattern1 and Pattern2 will be matched against the term. The idea behind this feature is to avoid reconstruction of terms. Example:

f({connect,From,To,Number,Options}, To) ->
    Signal = {connect,From,To,Number,Options},
    ...;
f(Signal, To) ->
    ignore.

can instead be written as

f({connect,_,To,_,_} = Signal, To) ->
    ...;
f(Signal, To) ->
    ignore.

String Prefix in Patterns

When matching strings, the following is a valid pattern:

f("prefix" ++ Str) -> ...

This is syntactic sugar for the equivalent, but harder to read

f([$p,$r,$e,$f,$i,$x | Str]) -> ...

Expressions in Patterns

An arithmetic expression can be used within a pattern, if it uses only numeric or bitwise operators, and if its value can be evaluated to a constant at compile-time. Example:

case {Value, Result} of
    {?THRESHOLD+1, ok} -> ...

This feature was added in Erlang 5.0/OTP R7.

8.5 Match

Expr1 = Expr2

Matches Expr1, a pattern, against Expr2. If the matching succeeds, any unbound variable in the pattern becomes bound and the value of Expr2 is returned.

If the matching fails, a badmatch run-time error will occur.

Examples:

1> {A, B} = {answer, 42}.
{answer,42}
2> A.
answer
3> {C, D} = [1, 2].
** exception error: no match of right hand side value [1,2]

8.6 Function Calls

ExprF(Expr1,...,ExprN)
ExprM:ExprF(Expr1,...,ExprN)

In the first form of function calls, ExprM:ExprF(Expr1,...,ExprN), each of ExprM and ExprF must be an atom or an expression that evaluates to an atom. The function is said to be called by using the fully qualified function name. This is often referred to as a remote or external function call. Example:

lists:keysearch(Name, 1, List)

In the second form of function calls, ExprF(Expr1,...,ExprN), ExprF must be an atom or evaluate to a fun.

If ExprF is an atom the function is said to be called by using the implicitly qualified function name. If the function ExprF is locally defined, it is called. Alternatively if ExprF is explicitly imported from module M, M:ExprF(Expr1,...,ExprN) is called. If ExprF is neither declared locally nor explicitly imported, ExprF must be the name of an automatically imported BIF. Examples:

handle(Msg, State)
spawn(m, init, [])

Examples where ExprF is a fun:

Fun1 = fun(X) -> X+1 end
Fun1(3)
=> 4

fun lists:append/2([1,2], [3,4])
=> [1,2,3,4]

Note that when calling a local function, there is a difference between using the implicitly or fully qualified function name, as the latter always refers to the latest version of the module. See Compilation and Code Loading.

See also the chapter about Function Evaluation.

Local Function Names Clashing With Auto-imported BIFs

If a local function has the same name as an auto-imported BIF, the semantics is that implicitly qualified function calls are directed to the locally defined function, not to the BIF. To avoid confusion, there is a compiler directive available, -compile({no_auto_import,[F/A]}), that makes a BIF not being auto-imported. In certain situations, such a compile-directive is mandatory.

Warning

Before OTP R14A (ERTS version 5.8), an implicitly qualified function call to a function having the same name as an auto-imported BIF always resulted in the BIF being called. In newer versions of the compiler the local function is instead called. The change is there to avoid that future additions to the set of auto-imported BIFs does not silently change the behavior of old code.

However, to avoid that old (pre R14) code changed its behavior when compiled with OTP version R14A or later, the following restriction applies: If you override the name of a BIF that was auto-imported in OTP versions prior to R14A (ERTS version 5.8) and have an implicitly qualified call to that function in your code, you either need to explicitly remove the auto-import using a compiler directive, or replace the call with a fully qualified function call, otherwise you will get a compilation error. See example below:

-export([length/1,f/1]).

-compile({no_auto_import,[length/1]}). % erlang:length/1 no longer autoimported

length([]) ->
    0;
length([H|T]) ->
    1 + length(T). %% Calls the local function length/1

f(X) when erlang:length(X) > 3 -> %% Calls erlang:length/1,
                                  %% which is allowed in guards
    long.

The same logic applies to explicitly imported functions from other modules as to locally defined functions. To both import a function from another module and have the function declared in the module at the same time is not allowed.

-export([f/1]).

-compile({no_auto_import,[length/1]}). % erlang:length/1 no longer autoimported

-import(mod,[length/1]).

f(X) when erlang:length(X) > 33 -> %% Calls erlang:length/1,
                                   %% which is allowed in guards

    erlang:length(X);              %% Explicit call to erlang:length in body

f(X) ->
    length(X).                     %% mod:length/1 is called

For auto-imported BIFs added to Erlang in release R14A and thereafter, overriding the name with a local function or explicit import is always allowed. However, if the -compile({no_auto_import,[F/A]) directive is not used, the compiler will issue a warning whenever the function is called in the module using the implicitly qualified function name.

8.7 If

if
    GuardSeq1 ->
        Body1;
    ...;
    GuardSeqN ->
        BodyN
end

The branches of an if-expression are scanned sequentially until a guard sequence GuardSeq which evaluates to true is found. Then the corresponding Body (sequence of expressions separated by ',') is evaluated.

The return value of Body is the return value of the if expression.

If no guard sequence is true, an if_clause run-time error will occur. If necessary, the guard expression true can be used in the last branch, as that guard sequence is always true.

Example:

is_greater_than(X, Y) ->
    if
        X>Y ->
            true;
        true -> % works as an 'else' branch
            false
    end

8.8 Case

case Expr of
    Pattern1 [when GuardSeq1] ->
        Body1;
    ...;
    PatternN [when GuardSeqN] ->
        BodyN
end

The expression Expr is evaluated and the patterns Pattern are sequentially matched against the result. If a match succeeds and the optional guard sequence GuardSeq is true, the corresponding Body is evaluated.

The return value of Body is the return value of the case expression.

If there is no matching pattern with a true guard sequence, a case_clause run-time error will occur.

Example:

is_valid_signal(Signal) ->
    case Signal of
        {signal, _What, _From, _To} ->
            true;
        {signal, _What, _To} ->
            true;
        _Else ->
            false
    end.

8.9 Send

Expr1 ! Expr2

Sends the value of Expr2 as a message to the process specified by Expr1. The value of Expr2 is also the return value of the expression.

Expr1 must evaluate to a pid, a registered name (atom) or a tuple {Name,Node}, where Name is an atom and Node a node name, also an atom.

If Expr1 evaluates to a name, but this name is not registered, a badarg run-time error will occur.
Sending a message to a pid never fails, even if the pid identifies a non-existing process.
Distributed message sending, that is if Expr1 evaluates to a tuple {Name,Node} (or a pid located at another node), also never fails.

8.10 Receive

receive
    Pattern1 [when GuardSeq1] ->
        Body1;
    ...;
    PatternN [when GuardSeqN] ->
        BodyN
end

Receives messages sent to the process using the send operator (!). The patterns Pattern are sequentially matched against the first message in time order in the mailbox, then the second, and so on. If a match succeeds and the optional guard sequence GuardSeq is true, the corresponding Body is evaluated. The matching message is consumed, that is removed from the mailbox, while any other messages in the mailbox remain unchanged.

The return value of Body is the return value of the receive expression.

receive never fails. Execution is suspended, possibly indefinitely, until a message arrives that does match one of the patterns and with a true guard sequence.

Example:

wait_for_onhook() ->
    receive
        onhook ->
            disconnect(),
            idle();
        {connect, B} ->
            B ! {busy, self()},
            wait_for_onhook()
    end.

It is possible to augment the receive expression with a timeout:

receive
    Pattern1 [when GuardSeq1] ->
        Body1;
    ...;
    PatternN [when GuardSeqN] ->
        BodyN
after
    ExprT ->
        BodyT
end

ExprT should evaluate to an integer. The highest allowed value is 16#ffffffff, that is, the value must fit in 32 bits. receive..after works exactly as receive, except that if no matching message has arrived within ExprT milliseconds, then BodyT is evaluated instead and its return value becomes the return value of the receive..after expression.

Example:

wait_for_onhook() ->
    receive
        onhook ->
            disconnect(),
            idle();
        {connect, B} ->
            B ! {busy, self()},
            wait_for_onhook()
    after
        60000 ->
            disconnect(),
            error()
    end.

It is legal to use a receive..after expression with no branches:

receive
after
    ExprT ->
        BodyT
end

This construction will not consume any messages, only suspend execution in the process for ExprT milliseconds and can be used to implement simple timers.

Example:

timer() ->
    spawn(m, timer, [self()]).

timer(Pid) ->
    receive
    after
        5000 ->
            Pid ! timeout
    end.

There are two special cases for the timeout value ExprT:

infinity: The process should wait indefinitely for a matching message -- this is the same as not using a timeout. Can be useful for timeout values that are calculated at run-time.
0: If there is no matching message in the mailbox, the timeout will occur immediately.

8.11 Term Comparisons

Expr1 op Expr2

op	Description
==	equal to
/=	not equal to
=<	less than or equal to
<	less than
>=	greater than or equal to
>	greater than
=:=	exactly equal to
=/=	exactly not equal to

Table 8.1: Term Comparison Operators.

The arguments may be of different data types. The following order is defined:

number < atom < reference < fun < port < pid < tuple < list < bit string

Lists are compared element by element. Tuples are ordered by size, two tuples with the same size are compared element by element.

When comparing an integer to a float, the term with the lesser precision will be converted into the other term's type, unless the operator is one of =:= or =/=. A float is more precise than an integer until all significant figures of the float are to the left of the decimal point. This happens when the float is larger/smaller than +/-9007199254740992.0. The conversion strategy is changed depending on the size of the float because otherwise comparison of large floats and integers would lose their transitivity.

Returns the Boolean value of the expression, true or false.

Examples:

1> 1==1.0.
true
2> 1=:=1.0.
false
3> 1 > a.
false

8.12 Arithmetic Expressions

op Expr
Expr1 op Expr2

op	Description	Argument type
+	unary +	number
-	unary -	number
+		number
-		number
*		number
/	floating point division	number
bnot	unary bitwise not	integer
div	integer division	integer
rem	integer remainder of X/Y	integer
band	bitwise and	integer
bor	bitwise or	integer
bxor	arithmetic bitwise xor	integer
bsl	arithmetic bitshift left	integer
bsr	bitshift right	integer

Table 8.2: Arithmetic Operators.

Examples:

1> +1.
1
2> -1.
-1
3> 1+1.
2
4> 4/2.
2.0
5> 5 div 2.
2
6> 5 rem 2.
1
7> 2#10 band 2#01.
0
8> 2#10 bor 2#01.
3
9> a + 10.
** exception error: an error occurred when evaluating an arithmetic expression
     in operator  +/2
        called as a + 10
10> 1 bsl (1 bsl 64).
** exception error: a system limit has been reached
     in operator  bsl/2
        called as 1 bsl 18446744073709551616

8.13 Boolean Expressions

op Expr
Expr1 op Expr2

op	Description
not	unary logical not
and	logical and
or	logical or
xor	logical xor

Table 8.3: Logical Operators.

Examples:

1> not true.
false
2> true and false.
false
3> true xor false.
true
4> true or garbage.
** exception error: bad argument
     in operator  or/2
        called as true or garbage

8.14 Short-Circuit Expressions

Expr1 orelse Expr2
Expr1 andalso Expr2

Expressions where Expr2 is evaluated only if necessary. That is, Expr2 is evaluated only if Expr1 evaluates to false in an orelse expression, or only if Expr1 evaluates to true in an andalso expression. Returns either the value of Expr1 (that is, true or false) or the value of Expr2 (if Expr2 was evaluated).

Example 1:

case A >= -1.0 andalso math:sqrt(A+1) > B of

This will work even if A is less than -1.0, since in that case, math:sqrt/1 is never evaluated.

Example 2:

OnlyOne = is_atom(L) orelse
         (is_list(L) andalso length(L) == 1),

From R13A, Expr2 is no longer required to evaluate to a boolean value. As a consequence, andalso and orelse are now tail-recursive. For instance, the following function is tail-recursive in R13A and later:

all(Pred, [Hd|Tail]) ->
    Pred(Hd) andalso all(Pred, Tail);
all(_, []) ->
    true.

8.15 List Operations

Expr1 ++ Expr2
Expr1 -- Expr2

The list concatenation operator ++ appends its second argument to its first and returns the resulting list.

The list subtraction operator -- produces a list which is a copy of the first argument, subjected to the following procedure: for each element in the second argument, the first occurrence of this element (if any) is removed.

Example:

1> [1,2,3]++[4,5].
[1,2,3,4,5]
2> [1,2,3,2,1,2]--[2,1,2].
[3,1,2]

Warning

The complexity of A -- B is proportional to length(A)*length(B), meaning that it will be very slow if both A and B are long lists.

8.16 Map Expressions

Creating Maps

Constructing a new map is done by letting an expression K be associated with another expression V:

#{ K => V }

New maps may include multiple associations at construction by listing every association:

#{ K1 => V1, .., Kn => Vn }

An empty map is constructed by not associating any terms with each other:

#{}

All keys and values in the map are terms. Any expression is first evaluated and then the resulting terms are used as key and value respectively.

Keys and values are separated by the => arrow and associations are separated by ,.

Examples:

M0 = #{},                 % empty map
M1 = #{a => <<"hello">>}, % single association with literals
M2 = #{1 => 2, b => b},   % multiple associations with literals
M3 = #{k => {A,B}},       % single association with variables
M4 = #{{"w", 1} => f()}.  % compound key associated with an evaluated expression

where, A and B are any expressions and M0 through M4 are the resulting map terms.

If two matching keys are declared, the latter key will take precedence.

Example:

1> #{1 => a, 1 => b}.
#{1 => b }
2> #{1.0 => a, 1 => b}.
#{1 => b, 1.0 => a}

The order in which the expressions constructing the keys and their associated values are evaluated is not defined. The syntactic order of the key-value pairs in the construction is of no relevance, except in the above mentioned case of two matching keys.

Updating Maps

Updating a map has similar syntax as constructing it.

An expression defining the map to be updated is put in front of the expression defining the keys to be updated and their respective values.

M#{ K => V }

where M is a term of type map and K and V are any expression.

If key K does not match any existing key in the map, a new association will be created from key K to value V. If key K matches an existing key in map M its associated value will be replaced by the new value V. In both cases the evaluated map expression will return a new map.

If M is not of type map an exception of type badmap is thrown.

To only update an existing value, the following syntax is used,

M#{ K := V }

where M is an term of type map, V is an expression and K is an expression which evaluates to an existing key in M.

If key K does not match any existing keys in map M an exception of type badarg will be triggered at runtime. If a matching key K is present in map M its associated value will be replaced by the new value V and the evaluated map expression returns a new map.

If M is not of type map an exception of type badmap is thrown.

Examples:

M0 = #{},
M1 = M0#{a => 0},
M2 = M1#{a => 1, b => 2},
M3 = M2#{"function" => fun() -> f() end},
M4 = M3#{a := 2, b := 3}.  % 'a' and 'b' was added in `M1` and `M2`.

where M0 is any map. It follows that M1 .. M4 are maps as well.

More Examples:

1> M = #{1 => a}.
#{1 => a }
2> M#{1.0 => b}.
#{1 => a, 1.0 => b}.
3> M#{1 := b}.
#{1 => b}
4> M#{1.0 := b}.
** exception error: bad argument

As in construction, the order in which the key and value expressions are evaluated is not defined. The syntactic order of the key-value pairs in the update is of no relevance, except in the case where two keys match, in which case the latter value is used.

Maps in Patterns

Matching of key-value associations from maps is done in the following way:

#{ K := V } = M

where M is any map. The key K has to be an expression with bound variables or a literals, and V can be any pattern with either bound or unbound variables.

If the variable V is unbound, it will be bound to the value associated with the key K, which has to exist in the map M. If the variable V is bound, it has to match the value associated with K in M.

Example:

1> M = #{"tuple" => {1,2}}.
#{"tuple" => {1,2}}
2> #{"tuple" := {1,B}} = M.
#{"tuple" => {1,2}}
3> B.
2.

This will bind variable B to integer 2.

Similarly, multiple values from the map may be matched:

#{ K1 := V1, .., Kn := Vn } = M

where keys K1 .. Kn are any expressions with literals or bound variables. If all keys exist in map M all variables in V1 .. Vn will be matched to the associated values of their respective keys.

If the matching conditions are not met, the match will fail, either with

a badmatch exception, if used in the context of the matching operator as in the example,
or resulting in the next clause being tested in function heads and case expressions.

Matching in maps only allows for := as delimiters of associations. The order in which keys are declared in matching has no relevance.

Duplicate keys are allowed in matching and will match each pattern associated to the keys.

#{ K := V1, K := V2 } = M

Matching an expression against an empty map literal will match its type but no variables will be bound:

#{} = Expr

This expression will match if the expression Expr is of type map, otherwise it will fail with an exception badmatch.

Matching syntax: Example with literals in function heads

Matching of literals as keys are allowed in function heads.

%% only start if not_started
handle_call(start, From, #{ state := not_started } = S) ->
...
    {reply, ok, S#{ state := start }};

%% only change if started
handle_call(change, From, #{ state := start } = S) ->
...
    {reply, ok, S#{ state := changed }};

Maps in Guards

Maps are allowed in guards as long as all sub-expressions are valid guard expressions.

Two guard BIFs handles maps:

8.17 Bit Syntax Expressions

<<>>
<<E1,...,En>>

Each element Ei specifies a segment of the bit string. Each element Ei is a value, followed by an optional size expression and an optional type specifier list.

Ei = Value |
     Value:Size |
     Value/TypeSpecifierList |
     Value:Size/TypeSpecifierList

Used in a bit string construction, Value is an expression which should evaluate to an integer, float or bit string. If the expression is something else than a single literal or variable, it should be enclosed in parenthesis.

Used in a bit string matching, Value must be a variable, or an integer, float or string.

Note that, for example, using a string literal as in <<"abc">> is syntactic sugar for <<$a,$b,$c>>.

Used in a bit string construction, Size is an expression which should evaluate to an integer.

Used in a bit string matching, Size must be an integer or a variable bound to an integer.

The value of Size specifies the size of the segment in units (see below). The default value depends on the type (see below). For integer it is 8, for float it is 64, for binary and bitstring it is the whole binary or bit string. In matching, this default value is only valid for the very last element. All other bit string or binary elements in the matching must have a size specification.

For the utf8, utf16, and utf32 types, Size must not be given. The size of the segment is implicitly determined by the type and value itself.

TypeSpecifierList is a list of type specifiers, in any order, separated by hyphens (-). Default values are used for any omitted type specifiers.

Type= integer | float | binary | bytes | bitstring | bits | utf8 | utf16 | utf32: The default is integer. bytes is a shorthand for binary and bits is a shorthand for bitstring. See below for more information about the utf types.
Signedness= signed | unsigned: Only matters for matching and when the type is integer. The default is unsigned.
Endianness= big | little | native: Native-endian means that the endianness will be resolved at load time to be either big-endian or little-endian, depending on what is native for the CPU that the Erlang machine is run on. Endianness only matters when the Type is either integer, utf16, utf32, or float. The default is big.
Unit= unit:IntegerLiteral: The allowed range is 1..256. Defaults to 1 for integer, float and bitstring, and to 8 for binary. No unit specifier must be given for the types utf8, utf16, and utf32.

The value of Size multiplied with the unit gives the number of bits. A segment of type binary must have a size that is evenly divisible by 8.

Note

When constructing binaries, if the size N of an integer segment is too small to contain the given integer, the most significant bits of the integer will be silently discarded and only the N least significant bits will be put into the binary.

The types utf8, utf16, and utf32 specifies encoding/decoding of the Unicode Transformation Formats UTF-8, UTF-16, and UTF-32, respectively.

When constructing a segment of a utf type, Value must be an integer in the range 0..16#D7FF or 16#E000....16#10FFFF. Construction will fail with a badarg exception if Value is outside the allowed ranges. The size of the resulting binary segment depends on the type and/or Value. For utf8, Value will be encoded in 1 through 4 bytes. For utf16, Value will be encoded in 2 or 4 bytes. Finally, for utf32, Value will always be encoded in 4 bytes.

When constructing, a literal string may be given followed by one of the UTF types, for example: <<"abc"/utf8>> which is syntatic sugar for <<$a/utf8,$b/utf8,$c/utf8>>.

A successful match of a segment of a utf type results in an integer in the range 0..16#D7FF or 16#E000..16#10FFFF. The match will fail if returned value would fall outside those ranges.

A segment of type utf8 will match 1 to 4 bytes in the binary, if the binary at the match position contains a valid UTF-8 sequence. (See RFC-3629 or the Unicode standard.)

A segment of type utf16 may match 2 or 4 bytes in the binary. The match will fail if the binary at the match position does not contain a legal UTF-16 encoding of a Unicode code point. (See RFC-2781 or the Unicode standard.)

A segment of type utf32 may match 4 bytes in the binary in the same way as an integer segment matching 32 bits. The match will fail if the resulting integer is outside the legal ranges mentioned above.

Examples:

1> Bin1 = <<1,17,42>>.
<<1,17,42>>
2> Bin2 = <<"abc">>.
<<97,98,99>>
3> Bin3 = <<1,17,42:16>>.
<<1,17,0,42>>
4> <<A,B,C:16>> = <<1,17,42:16>>.
<<1,17,0,42>>
5> C.
42
6> <<D:16,E,F>> = <<1,17,42:16>>.
<<1,17,0,42>>
7> D.
273
8> F.
42
9> <<G,H/binary>> = <<1,17,42:16>>.
<<1,17,0,42>>
10> H.
<<17,0,42>>
11> <<G,H/bitstring>> = <<1,17,42:12>>.
<<1,17,1,10:4>>
12> H.
<<17,1,10:4>>
13> <<1024/utf8>>.
<<208,128>>

Note that bit string patterns cannot be nested.

Note also that "B=<<1>>" is interpreted as "B =<<1>>" which is a syntax error. The correct way is to write a space after '=': "B= <<1>>.

More examples can be found in Programming Examples.

8.18 Fun Expressions

fun
    [Name](Pattern11,...,Pattern1N) [when GuardSeq1] ->
              Body1;
    ...;
    [Name](PatternK1,...,PatternKN) [when GuardSeqK] ->
              BodyK
end

A fun expression begins with the keyword fun and ends with the keyword end. Between them should be a function declaration, similar to a regular function declaration, except that the function name is optional and should be a variable if any.

Variables in a fun head shadow the function name and both shadow variables in the function clause surrounding the fun expression, and variables bound in a fun body are local to the fun body.

The return value of the expression is the resulting fun.

Examples:

1> Fun1 = fun (X) -> X+1 end.
#Fun<erl_eval.6.39074546>
2> Fun1(2).
3
3> Fun2 = fun (X) when X>=5 -> gt; (X) -> lt end.
#Fun<erl_eval.6.39074546>
4> Fun2(7).
gt
5> Fun3 = fun Fact(1) -> 1; Fact(X) when X > 1 -> X * Fact(X - 1) end.
#Fun<erl_eval.6.39074546>
6> Fun3(4).
24

The following fun expressions are also allowed:

fun Name/Arity
fun Module:Name/Arity

In Name/Arity, Name is an atom and Arity is an integer. Name/Arity must specify an existing local function. The expression is syntactic sugar for:

fun (Arg1,...,ArgN) -> Name(Arg1,...,ArgN) end

In Module:Name/Arity, Module and Name are atoms and Arity is an integer. Starting from the R15 release, Module, Name, and Arity may also be variables. A fun defined in this way will refer to the function Name with arity Arity in the latest version of module Module. A fun defined in this way will not be dependent on the code for module in which it is defined.

More examples can be found in Programming Examples.

8.19 Catch and Throw

catch Expr

Returns the value of Expr unless an exception occurs during the evaluation. In that case, the exception is caught. For exceptions of class error, that is run-time errors: {'EXIT',{Reason,Stack}} is returned. For exceptions of class exit, that is the code called exit(Term): {'EXIT',Term} is returned. For exceptions of class throw, that is the code called throw(Term): Term is returned.

Reason depends on the type of error that occurred, and Stack is the stack of recent function calls, see Errors and Error Handling.

Examples:

1> catch 1+2.
3
2> catch 1+a.
{'EXIT',{badarith,[...]}}

Note that catch has low precedence and catch subexpressions often needs to be enclosed in a block expression or in parenthesis:

3> A = catch 1+2.
** 1: syntax error before: 'catch' **
4> A = (catch 1+2).
3

The BIF throw(Any) can be used for non-local return from a function. It must be evaluated within a catch, which will return the value Any. Example:

5> catch throw(hello).
hello

If throw/1 is not evaluated within a catch, a nocatch run-time error will occur.

8.20 Try

try Exprs
catch
    [Class1:]ExceptionPattern1 [when ExceptionGuardSeq1] ->
        ExceptionBody1;
    [ClassN:]ExceptionPatternN [when ExceptionGuardSeqN] ->
        ExceptionBodyN
end

This is an enhancement of catch that appeared in Erlang 5.4/OTP-R10B. It gives the possibility do distinguish between different exception classes, and to choose to handle only the desired ones, passing the others on to an enclosing try or catch or to default error handling.

Note that although the keyword catch is used in the try expression, there is not a catch expression within the try expression.

Returns the value of Exprs (a sequence of expressions Expr1, ..., ExprN) unless an exception occurs during the evaluation. In that case the exception is caught and the patterns ExceptionPattern with the right exception class Class are sequentially matched against the caught exception. An omitted Class is shorthand for throw. If a match succeeds and the optional guard sequence ExceptionGuardSeq is true, the corresponding ExceptionBody is evaluated to become the return value.

If an exception occurs during evaluation of Exprs but there is no matching ExceptionPattern of the right Class with a true guard sequence, the exception is passed on as if Exprs had not been enclosed in a try expression.

If an exception occurs during evaluation of ExceptionBody it is not caught.

The try expression can have an of section:

try Exprs of
    Pattern1 [when GuardSeq1] ->
        Body1;
    ...;
    PatternN [when GuardSeqN] ->
        BodyN
catch
    [Class1:]ExceptionPattern1 [when ExceptionGuardSeq1] ->
        ExceptionBody1;
    ...;
    [ClassN:]ExceptionPatternN [when ExceptionGuardSeqN] ->
        ExceptionBodyN
end

If the evaluation of Exprs succeeds without an exception, the patterns Pattern are sequentially matched against the result in the same way as for a case expression, except that if the matching fails, a try_clause run-time error will occur.

An exception occurring during the evaluation of Body is not caught.

The try expression can also be augmented with an after section, intended to be used for cleanup with side effects:

try Exprs of
    Pattern1 [when GuardSeq1] ->
        Body1;
    ...;
    PatternN [when GuardSeqN] ->
        BodyN
catch
    [Class1:]ExceptionPattern1 [when ExceptionGuardSeq1] ->
        ExceptionBody1;
    ...;
    [ClassN:]ExceptionPatternN [when ExceptionGuardSeqN] ->
        ExceptionBodyN
after
    AfterBody
end

AfterBody is evaluated after either Body or ExceptionBody no matter which one. The evaluated value of AfterBody is lost; the return value of the try expression is the same with an after section as without.

Even if an exception occurs during evaluation of Body or ExceptionBody, AfterBody is evaluated. In this case the exception is passed on after AfterBody has been evaluated, so the exception from the try expression is the same with an after section as without.

If an exception occurs during evaluation of AfterBody itself it is not caught, so if AfterBody is evaluated after an exception in Exprs, Body or ExceptionBody, that exception is lost and masked by the exception in AfterBody.

The of, catch and after sections are all optional, as long as there is at least a catch or an after section, so the following are valid try expressions:

try Exprs of 
    Pattern when GuardSeq -> 
        Body 
after 
    AfterBody 
end

try Exprs
catch 
    ExpressionPattern -> 
        ExpressionBody
after
    AfterBody
end

try Exprs after AfterBody end

Example of using after, this code will close the file even in the event of exceptions in file:read/2 or in binary_to_term/1, and exceptions will be the same as without the try...after...end expression:

termize_file(Name) ->
    {ok,F} = file:open(Name, [read,binary]),
    try
        {ok,Bin} = file:read(F, 1024*1024),
        binary_to_term(Bin)
    after
        file:close(F)
    end.

Example: Using try to emulate catch Expr.

try Expr
catch
    throw:Term -> Term;
    exit:Reason -> {'EXIT',Reason}
    error:Reason -> {'EXIT',{Reason,erlang:get_stacktrace()}}
end

8.21 Parenthesized Expressions

(Expr)

Parenthesized expressions are useful to override operator precedences, for example in arithmetic expressions:

1> 1 + 2 * 3.
7
2> (1 + 2) * 3.
9

8.22 Block Expressions

begin
   Expr1,
   ...,
   ExprN
end

Block expressions provide a way to group a sequence of expressions, similar to a clause body. The return value is the value of the last expression ExprN.

8.23 List Comprehensions

List comprehensions are a feature of many modern functional programming languages. Subject to certain rules, they provide a succinct notation for generating elements in a list.

List comprehensions are analogous to set comprehensions in Zermelo-Frankel set theory and are called ZF expressions in Miranda. They are analogous to the setof and findall predicates in Prolog.

List comprehensions are written with the following syntax:

[Expr || Qualifier1,...,QualifierN]

Expr is an arbitrary expression, and each Qualifier is either a generator or a filter.

A generator is written as:
Pattern <- ListExpr.
ListExpr must be an expression which evaluates to a list of terms.
A bit string generator is written as:
BitstringPattern <= BitStringExpr.
BitStringExpr must be an expression which evaluates to a bitstring.
A filter is an expression which evaluates to true or false.

The variables in the generator patterns shadow variables in the function clause surrounding the list comprehensions.

A list comprehension returns a list, where the elements are the result of evaluating Expr for each combination of generator list elements and bit string generator elements for which all filters are true.

Example:

1> [X*2 || X <- [1,2,3]].
[2,4,6]

More examples can be found in Programming Examples.

8.24 Bit String Comprehensions

Bit string comprehensions are analogous to List Comprehensions. They are used to generate bit strings efficiently and succinctly.

Bit string comprehensions are written with the following syntax:

<< BitString || Qualifier1,...,QualifierN >>

BitString is a bit string expression, and each Qualifier is either a generator, a bit string generator or a filter.

A generator is written as:
Pattern <- ListExpr.
ListExpr must be an expression which evaluates to a list of terms.
A bit string generator is written as:
BitstringPattern <= BitStringExpr.
BitStringExpr must be an expression which evaluates to a bitstring.
A filter is an expression which evaluates to true or false.

The variables in the generator patterns shadow variables in the function clause surrounding the bit string comprehensions.

A bit string comprehension returns a bit string, which is created by concatenating the results of evaluating BitString for each combination of bit string generator elements for which all filters are true.

Example:

1> << << (X*2) >> || 
<<X>> <= << 1,2,3 >> >>.
<<2,4,6>>

More examples can be found in Programming Examples.

8.25 Guard Sequences

A guard sequence is a sequence of guards, separated by semicolon (;). The guard sequence is true if at least one of the guards is true. (The remaining guards, if any, will not be evaluated.)
Guard1;...;GuardK

A guard is a sequence of guard expressions, separated by comma (,). The guard is true if all guard expressions evaluate to true.
GuardExpr1,...,GuardExprN

The set of valid guard expressions (sometimes called guard tests) is a subset of the set of valid Erlang expressions. The reason for restricting the set of valid expressions is that evaluation of a guard expression must be guaranteed to be free of side effects. Valid guard expressions are:

the atom true,
other constants (terms and bound variables), all regarded as false,
calls to the BIFs specified below,
term comparisons,
arithmetic expressions,
boolean expressions, and
short-circuit expressions (andalso/orelse).

is_atom/1

is_binary/1

is_bitstring/1

is_boolean/1

is_float/1

is_function/1

is_function/2

is_integer/1

is_list/1

is_map/1

is_number/1

is_pid/1

is_port/1

is_record/2

is_record/3

is_reference/1

is_tuple/1

Table 8.4: Type Test BIFs.

Note that most type test BIFs have older equivalents, without the is_ prefix. These old BIFs are retained for backwards compatibility only and should not be used in new code. They are also only allowed at top level. For example, they are not allowed in boolean expressions in guards.

abs(Number)

bit_size(Bitstring)

byte_size(Bitstring)

element(N, Tuple)

float(Term)

hd(List)

length(List)

map_size(Map)

node()

node(Pid|Ref|Port)

round(Number)

self()

size(Tuple|Bitstring)

tl(List)

trunc(Number)

tuple_size(Tuple)

Table 8.5: Other BIFs Allowed in Guard Expressions.

If an arithmetic expression, a boolean expression, a short-circuit expression, or a call to a guard BIF fails (because of invalid arguments), the entire guard fails. If the guard was part of a guard sequence, the next guard in the sequence (that is, the guard following the next semicolon) will be evaluated.

8.26 Operator Precedence

Operator precedence in falling priority:

:
#
Unary + - bnot not
/ * div rem band and	Left associative
+ - bor bxor bsl bsr or xor	Left associative
++ --	Right associative
== /= =< < >= > =:= =/=
andalso
orelse
= !	Right associative
catch

Table 8.6: Operator Precedence.

When evaluating an expression, the operator with the highest priority is evaluated first. Operators with the same priority are evaluated according to their associativity. Example: The left associative arithmetic operators are evaluated left to right:

6 + 5 * 4 - 3 / 2 evaluates to
6 + 20 - 1.5 evaluates to
26 - 1.5 evaluates to
24.5