Note: I've read Josh's more recent posts and I think he's heading towards the
right answer so I will respond there separately, but wanted to address your
technical questions for the record.
Again, I am not a compiler/language expert. I'm just going with my mental
model based on working on the compiler over the past years.
The source code is parsed into an AST (a tree of nodes). An assignment
statement becomes a BinaryOperatorNode. The "=" is the operator and there is a
leftOperandNode and rightOperandNode. I think the compiler has lots of
expectations that the rightOperandNode can be resolved to a type, and lots of
expectations that types must behave according to the rules of ActionScript. In
this case, that an Object cannot be coerced to a class instance.
So, right now, the rightOperandNode will resolve to be Object. I think you are
suggesting that you want to change resolveType to keep walking up the tree of
nodes across the operator to the leftOperandNode. That seems slow and may
result in big changes to the compiler. I don't think a BinaryOperatorNode is
even an Expression which is a concept for nodes that can resolve to a type.
If you want to "declare" an Object literal as a type, you might write:
BlobPropertyBag({ type: "text/plain"})
That defines the rightOperandNode as an expression of type BlobPropertyBag.
Hence why I think Conversion functions are useful. Now on to Josh's posts.
My 2 cents,
-Alex
On 1/10/19, 1:59 AM, "Harbs" <[email protected]> wrote:
I don’t know what you mean here.
My proposal would not change anything for literals which are typed as
Object.
I’m proposing that literals that are declared as “typdefed” types would be
checked against the typedefs. (I like the use of “typedef” as Haxe does rather
than “interface” to avoid confusion with real interfaces.)
I don’t understand what performance problems you envision and I don’t
understand why this would create language issues.
Harbs
> On Jan 10, 2019, at 10:47 AM, Alex Harui <[email protected]> wrote:
>
> @Harbs: I am not a language/compiler expert, but I do not think that
languages typically have literals that are automatically coerced to types. I
believe the AST node tree of the literal can be reduced/compiled/transpiled
without knowing that it is part of an assignment statement. I believe your
proposal is to change that, and I am concerned about the compiler performance
and language issues that could arise from that. I would strongly recommend we
don't do that and stay with constructs that are more common to other languages.
