Enums
Tomo supports tagged enumerations, also known as “sum types.” Users can
define their own using the enum keyword:
enum VariousThings(AnInteger(i:Int), TwoWords(word1, word2:Text), Nothing)
...
a := VariousThings.AnInteger(5)
b := VariousThings.TwoWords("one", "two")
c := VariousThings.NothingPattern Matching
The values inside an enum can be accessed with pattern matching
when x is AnInteger(i)
say("It was $i")
is TwoWords(x, y)
say("It was $x and $y")
is Nothing
say("It was nothing")Pattern matching blocks are always checked for exhaustiveness, but you can
add an else block to handle all unmatched patterns.
Tag Checking
Tags can also be quickly checked using the .TagName field:
assert a.AnInteger != none
assert a.TwoWords == noneReducing Boilerplate
There are three main areas where we can easily reduce the amount of
boilerplate around enums. We don’t need to type VariousThings. in
front of enum values when we already know what type of enum we’re dealing
with. This means that we don’t need the name of the type for pattern matching
(because we can infer the type of the expression being matched). We also don’t
need the name of the type when calling a function with an enum argument, nor
when returning an enum value from a function with an explicit return type:
enum ArgumentType(AnInt(x:Int), SomeText(text:Text))
enum ReturnType(Nothing, AnInt(x:Int))
func increment(arg:ArgumentType -> ReturnType)
when arg is AnInt(x)
return AnInt(x + 1)
is SomeText
return Nothing
...
assert increment(AnInt(5)) == AnInt(6)
assert increment(SomeText("HI")) == NothiingThis lets us have overlapping tag names for different types, but smartly infer which enum’s value is being created when we know what we’re expecting to get. This also works for variable assignment to a variable whose type is already known.
Namespacing
Enums can also define their own methods and variables inside their namespace:
enum VariousThings(AnInteger(i:Int), TwoWords(word1, word2:Text), Nothing)
meaningful_thing := AnInteger(42)
func doop(v:VariousThings)
say("$v")Functions defined in an enum’s namespace can be invoked as methods with
: if the first argument is the enum’s type or a pointer to one
(vt.doop()).
Anonymous Enums
In some cases, you may want to use anonymous inline-defined enums. This lets you define a lightweight type without a name for cases where that’s more convenient. For example, a function that has a simple variant for an argument:
func pad_text(text:Text, width:Int, align:enum(Left,Right,Center) = Left -> Text)
...
...
padded := pad_text(text, 10, Right)This could be defined explicitly as
enum TextAlignment(Left,Right,Center) with pad_text
defining align:TextAlignment, but this adds a new symbol to the
top-level scope and forces the user to think about which name is being used.
In some applications, that overhead is not necessary or desirable.
Anonymous enums can be used in any place where a type is specified:
- Declarations:
my_variable : enum(A, B, C) = A - Function arguments:
func foo(arg:enum(A, B, C)) - Function return values:
func foo(x:Int -> enum(Valid(result:Int), Invalid(reason:Text))) - Struct members:
struct Foo(x:enum(A,B,C)),enum Baz(Thing(type:enum(A,B,C)))
In general, anonymous enums should be used sparingly in cases where there
are only a small number of options and the enum code is short. If you expect
users to refer to the enum type, it ought to be defined with a proper name. In
the pad_text example, the anonymous enum would cause problems if
you wanted to make a wrapper around it, because you would not be able to refer
to the pad_text align argument’s type:
func pad_text_wrapper(text:Text, width:Int, align:???)
...pad_text(text, width, align)...Note: Each enum type is distinct, regardless of whether the enum shares the same values with another enum, so you can’t define another enum with the same values and use that in places where a different anonymous enum is expected.
