On 12/2/15, 2:37 PM, "Josh Tynjala" <[email protected]> wrote:
>Oh, I see. I didn't realize that you were thinking that users would import
>something and also manually call require(). In that case, yes, you could
>still potentially have proper type checking.
>
>I like the idea of using some kind of wrapper class to hide the require()
>call. It's kind of messy to both import and call require(), so I really
>would prefer to hide one or the other, if possible.
It's up to you or whoever writes this code. A wrapper adds a layer of
function calls but does let you hide some ugliness.
>
>There's one thing that seems like an issue to me, though. Using an as cast
>with require() will probably result in a null result because what is
>returned by require("fs") doesn't actually extend an FS class (or
>implement
>an FS interface).
>
>var fs:FS = require("fs") as FS;
We have this problem already. There is a special ASDoc tag called
@flexjsignorecoercion that suppresses "as" code in the cross-compiled
output. I've considered trying to get the compiler to automatically
suppress the "as" code for any "as" usage that references a class in an
external-library-path SWC. That seems like it could fail you in some
distributed development cases where you have other code that does support
AS in an external-library-path SWC. We could add file-level suppression
as well, or have a config option that lists certain coercions to always
suppress. Definitely open to ideas on this, but the fact is, the compiler
will not let you have type-checking without using "as" so you sort of have
to have a scheme to deal with it.
-Alex