Guards
Guards are a way to augment pattern matching with more complex checks; they are allowed in a predefined set of constructs where pattern matching is allowed.
List of allowed expressions
For reference, the following is a comprehensive list of all expressions allowed in guards:
- comparison operators (
==
,!=
,===
,!==
,>
,>=
,<
,<=
) - strictly boolean operators (
and
,or
,not
) (the&&
,||
, and!
sibling operators are not allowed as they’re not strictly boolean - meaning they don’t require both sides to be booleans) - arithmetic binary operators (
+
,-
,*
,/
) - arithmetic unary operators (
+
,-
) - binary concatenation operator (
<>
) in
andnot in
operators (as long as the right-hand side is a list or a range)the following “type-check” functions (all documented in the
Kernel
module):the following guard-friendly functions (all documented in the
Kernel
module):the following handful of Erlang bitwise operations, if imported from the
Bitwise
module:band/2
or the&&&
operatorbor/2
or the|||
operatorbnot/1
or the~~~
operatorbsl/1
or the<<<
operatorbsr/1
or the>>>
operatorbxor/2
or the^^^
operator
Macros constructed out of any combination of the above guards are also valid guards - for example, Integer.is_even/1
. See the section “Defining custom guard expressions” below.
Why guards
Let’s see an example of a guard used in a function clause:
def empty_map?(map) when map_size(map) == 0, do: true
def empty_map?(map) when is_map(map), do: false
Guards start with the when
keyword, which is followed by a boolean expression (we will define the grammar of guards more formally later on).
Writing the empty_map?/1
function by only using pattern matching would not be possible (as pattern matching on %{}
would match every map, not empty maps).
Where guards can be used
In the example above, we show how guards can be used in function clauses. There are several constructs that allow guards; for example:
- function clauses:
def foo(term) when is_integer(term), do: term
def foo(term) when is_float(term), do: round(term)
case
expressions:
case x do
1 -> :one
2 -> :two
n when is_integer(n) and n > 2 -> :larger_than_two
end
- anonymous functions (
fn
s):
larger_than_two? = fn
n when is_integer(n) and n > 2 -> true
n when is_integer(n) -> false
end
Other constructs are for
, with
, try
/rescue
/catch
/else
/, and the match?/2
macro in the Kernel
module.
Failing guards
Errors in guards do not result in runtime errors, but in guards failing. For example, the length/1
function only works with lists. If we use it with anything else, a runtime error is raised:
iex> length("hello")
** (ArgumentError) argument error
However, when used in guards, the corresponding clause simply fails to match:
iex> case "hello" do
...> something when length(something) > 0 ->
...> :length_worked
...> _anything_else ->
...> :length_failed
...> end
:length_failed
In many cases, we can take advantage of this. In the code above, we used length/1
to both check that the given thing is a list and check some properties of its length (instead of using is_list(something) and length(something) > 0
).
Expressions in guard clauses
Not all expressions are allowed in guard clauses, but only a handful of them. This is a deliberate choice: only a predefined set of side-effect-free functions are allowed. This way, Elixir (and Erlang) can make sure that nothing bad happens while executing guards and no mutations happen anywhere. This behaviour is also coherent with pattern match, which is a naturally a side-effect-free operation. Finally, keeping expressions allowed in clauses to a close set of predefined ones allows the compiler to optimize the code related to choosing the right clause.
Defining custom guard expressions
As mentioned before, only the expressions listed in this page are allowed in guards. However, we can take advantage of macros to write custom guards that can simplify our programs or make them more domain-specific. At the end of the day, what matters is that the output of the macros (which is what will be compiled) boils down to a combinations of the allowed expressions.
Let’s look at a quick case study: we want to check that a function argument is an even or odd integer. With pattern matching, this is impossible to do since there are infinite integers, and thus we can’t pattern match on the single even/odd numbers. Let’s focus on checking for even numbers since checking for odd ones is almost identical.
Such a guard would look like this:
def my_function(number) when is_integer(number) and rem(number, 2) == 0 do
# do stuff
end
This would be repetitive to write every time we need this check, so, as mentioned at the beginning of this section, we can abstract this away using a macro. Remember that defining a function that performs this check wouldn’t work because we can’t use custom functions in guards. Our macro would look like this:
defmodule MyInteger do
defmacro is_even(number) do
quote do
is_integer(unquote(number)) and rem(unquote(number), 2) == 0
end
end
end
and then:
import MyInteger, only: [is_even: 1]
def my_function(number) when is_even(number) do
# do stuff
end
Multiple guards in the same clause
There exists an additional way to simplify a chain of or
s in guards: Elixir supports writing “multiple guards” in the same clause. This:
def foo(term) when is_integer(term) or is_float(term) or is_nil(term),
do: :maybe_number
def foo(_other),
do: :something_else
can be alternatively written as:
def foo(term)
when is_integer(term)
when is_float(term)
when is_nil(term) do
:maybe_number
end
def foo(_other) do
:something_else
end
For most cases, the two forms are exactly the same. However, there exists a subtle difference in the case of failing guards, as discussed in the section above. In case of a boolean expression guard, a failed element means the whole guard fails. In case of multiple guards it means the next one will be evaluated. The difference can be highlighted with an example:
def multiguard(value)
when map_size(value) < 1
when tuple_size(value) < 1 do
:guard_passed
end
def multiguard(_value) do
:guard_failed
end
def boolean(value) when map_size(value) < 1 or tuple_size(value) < 1 do
:guard_passed
end
def boolean(value) do
:guard_failed
end
multiguard(%{}) #=> :guard_passed
multiguard({}) #=> :guard_passed
boolean(%{}) #=> :guard_passed
boolean({}) #=> :guard_failed
For cases where guards do not rely on the failing guard behavior the two forms are exactly the same semantically but there are cases where multiple guard clauses may be more aesthetically pleasing.