[[constraints]] == Constraints Constraints are validation and conversion functions applied to user input. They are used in various places in Cowboy, including the router and the `cowboy_req` match functions. === Syntax Constraints are provided as a list of fields. For each field in the list, specific constraints can be applied, as well as a default value if the field is missing. A field can take the form of an atom `field`, a tuple with constraints `{field, Constraints}` or a tuple with constraints and a default value `{field, Constraints, Default}`. The `field` form indicates the field is mandatory. Note that when used with the router, only the second form makes sense, as it does not use the default and the field is always defined. Constraints for each field are provided as an ordered list of atoms or funs to apply. Built-in constraints are provided as atoms, while custom constraints are provided as funs. When multiple constraints are provided, they are applied in the order given. If the value has been modified by a constraint then the next one receives the new value. For example, the following constraints will first validate and convert the field `my_value` to an integer, and then check that the integer is positive: [source,erlang] ---- PositiveFun = fun (_, V) when V > 0 -> {ok, V}; (_, _) -> {error, not_positive} end, {my_value, [int, PositiveFun]}. ---- We ignore the first fun argument in this snippet. We shouldn't. We will simply learn what it is later in this chapter. When there's only one constraint, it can be provided directly without wrapping it into a list: [source,erlang] ---- {my_value, int} ---- === Built-in constraints Built-in constraints are specified as an atom: [cols="<,<",options="header"] |=== | Constraint | Description | int | Converts binary value to integer. | nonempty | Ensures the binary value is non-empty. |=== === Custom constraints Custom constraints are specified as a fun. This fun takes two arguments. The first argument indicates the operation to be performed, and the second is the value. What the value is and what must be returned depends on the operation. Cowboy currently defines three operations. The operation used for validating and converting user input is the `forward` operation. [source,erlang] ---- int(forward, Value) -> try {ok, binary_to_integer(Value)} catch _:_ -> {error, not_an_integer} end; ---- The value must be returned even if it is not converted by the constraint. The `reverse` operation does the opposite: it takes a converted value and changes it back to what the user input would have been. [source,erlang] ---- int(reverse, Value) -> try {ok, integer_to_binary(Value)} catch _:_ -> {error, not_an_integer} end; ---- Finally, the `format_error` operation takes an error returned by any other operation and returns a formatted human-readable error message. [source,erlang] ---- int(format_error, {not_an_integer, Value}) -> io_lib:format("The value ~p is not an integer.", [Value]). ---- Notice that for this case you get both the error and the value that was given to the constraint that produced this error. Cowboy will not catch exceptions coming from constraint functions. They should be written to not emit any exceptions.