Interpolated String Singleton Types

docs/syntax-interpolated-string-singleton-types.md

@@ -0,0 +1,145 @@
+# Interpolated String Singleton Types
+
+## Summary
+
+Introduce string interpolation syntax for string singleton types (stringletons hereon), and bindings which currently
+generalize to stringletons will do so for interpolation string expressions if all sub-expressions are singleton types.
+
+## Motivation
+
+Luau has four types of string expressions:
+
+```luau
+-- single quoted
+const nyanya = 'mew'
+-- double quoted
+const mrrp = "chirp"
+-- long
+const meow = [[
+purr
+hiss
+]]
+-- interpolated
+const kya = `claw {mrrp} {nyanya}`
+```
+
+Currently, stringleton type annotations are supported in the form of single-quoted, double-quoted, and
+long strings - but not interp-strings. This makes type syntax inconsistent with runtime expressions, and makes it very
+difficult to combine string types. An existing solution exists by creating a user-defined type function to concatenate
+string singletons. However, access to basic formatting in the type system would provide:
+
+
+- syntactic consistency; the benefit here is obvious. Users of the language may be surprised or confused that they
+cannot use interpolation strings inside type annotations.
+
+- ease-of-use; Some style guides or formatters may prefer interpolated strings by default for cosmetic purposes or
+because it is faster to add a sub-expression than to change the quote style of a string, deal with escaping, etc..., and
+the same applies to type syntax as well.
+
+- higher quality inference; This is an implementation detail, but user-defined type functions also currently block
+bi-directional type inference. Although they may improve in the future, it is fundamentally impossible to reason about
+the result of a user-defined type function as empirically as you can with built-in types. Although admittedly uncertain
+& abstract, it's possible future improvements to polymorphic type signatures will add usefulness to interpolated string
+types.
+
+## Design
+
+### Syntax
+
+Interpolation quotes will become valid, usable stringleton syntax:
+
+```luau
+type Yowl = `growl :3`
+```
+
+It will also support sub-expression types:
+
+```luau
+type Headbutt = "Hello"
+-- 'Hello World'
+type PupilsDilated = `{Headbutt} World`
+```
+
+### General Purpose Stringification & Inference
+
+Exposing any kind of literal stringification to the type system would be fragile in concept because any interface we use
+would need to be completely stable. We probably don't want `__tostring` because runtimes can (and do) modify that. We
+probably do not want to expose the type system's `ToString.h` because it is disjoint from the runtime type. The `print`
+global is even less standardized. Returning 'string' would be arduous to debug - so non-singleton sub-expressions will
+instead report a type error for now:
+
+```luau
+type Headbutt = "Hello"
+type Name = string
+-- TypeError: Interpolated string sub-expressions must be literal strings, but this expression is of type 'string'.
+--    Type error span is here -> |    |
+type PupilsDilated = `{Headbutt} {Name}`
+
+-- We will allow boolean and nil singletons too. If number singleton types are ever introduced, supporting them should
+-- be discussed seprately.
+type PupilsConstricted = `{true} {nil}`
+```
+
+We also need to represent string singletons which depend on a generic type or unsolved type function - consider the
+following code:
+
+```luau
+const function cat<Name>(person: Human<Name>): `Brush {Name}`
+	return `Brush {person.name}`
+end
+```
+
+We will create an unresolved type function to represent interpolated stringletons which cannot be resolved yet. This
+type function won't be accessible in the user's environment. Analysis outputs containing these signatures will stringify
+it syntactically to interpolation quotes and sub-expressions.
+
+Currently, there are situations where a string expression will be inferred as a stringleton at a callsite. In any of
+these, the callsite will contribute to inference of sub-expressions:
+
+```luau
+const function leopard<loaf>(value: loaf | keyof<some_tbl>): loaf
+end
+
+local trill = {
+	chitter = ":3"
+}
+-- currently, this causes `trill.chitter` to infer to '":3"'
+leopard(trill.chitter)
+-- this currently instantiates leopard with 'loaf' as 'string'
+-- instead, it could coerce trill.chitter.
+leopard(`hello {trill.chitter}`)
+
+const spit: unknown = ...
+-- we will need to infer the lower bound of 'loaf' as 'string' here, because a sub-expression is not a singleton.
+leopard(`hello {spit}`)
+```
+
+### Stringleton Unions as Sub-Expressions
+
+Unions are particularly difficult to deal with. For now, we will define the behavior as shown in this code:
+
+```luau
+type Huff =
+	| "Squeak"
+	| "Click"
+type NyaNya = "Purr"
+-- This will materialize as "Hello Squeak Purr" | "Hello Click Purr"
+type Kya = `Hello {Huff} {NyaNya}`
+-- To avoid materializing exponentially many unions, this will emit a type error complaining about only up to one string
+-- singleton union sub-expression being allowed.
+type Mrrp = `{Huff} {Huff}`
+```
+
+## Drawbacks
+
+- Backtick quotes will become unavailable in type syntax. It is likely any other usage for them would be confusing
+anyways though.
+- Increases the complexity of type inference around string singletons.
+
+## Alternatives
+
+- Do nothing. People can use existing workarounds for the forseeable future, and the syntactic consistency is not
+critically important when weighed against potential drawbacks.
+- Handle unions via something more specialized. this isn't exlusive with this rfc.
+- Introduce utility type function(s) based on string formatting
+- Introduce concatenation syntax operator for types with `..`