So then the user gets a type error in their face instead of the page loading? That doesn't really sound better than the developer getting that error while writing the code, which is what TypeScript currently does.
In my mind, the alternative is running TypeScript while writing and at build time. Type checking at runtime is at best redundant, and at worst an excuse to skip type checking and have it blow up in the users' face.
> In my mind, the alternative is running TypeScript while writing and at build time.
Right now, sure.
But if TS is supported natively in the browser, wouldn't your editor highlight the errors as you type? In which case the chance of deploying a broken TS file to the browser is minimal - you'll have to go out of your way to do so, like writing the TS file in plain notepad.
To go even further, having TS supported in the browser does not mean that you are forced to abandon your build step(s). You are still free to run a build step that either:
1. Does the full compilation to JS, and that's what gets deployed.
or
2. Just lints the file, and has the original TS file deployed.
Nothing in "Native TS in the browser" enforces a no-build-step dev process; it just makes it optional.
There's also the fact that, if JS is no longer the target (either browser-byte-code or native-code will be the target), then type-checking can be improved even further because there will be no requirement to allow things purely due to JS compatibility.
Finally, there's an awfully large number of optimisations that can be done if JS is not the target and native-code is.
Yeah that's kinda the point; you get all that, unless you don't run the build-step/in-editor dev process. And the only reason to make that step optional, is to not run it, which is a worse experience.
So I'm not necessarily saying it's a bad thing, it's just that I don't see the point. And given that there's the major downside of having to go through the standards process, both now and in the future, which will likely involve breaking changes and making it harder to update, I don't see it happening. (Edit: I should add that I do think the "types as comments" proposal makes sense. I do see the advantage of being able to run TS code without a build step. It's just the part where we'd throw an error in the user's face that I don't see providing value to anyone.)
I do think TC39 is progressive enough to be OK with changes to JS if those would allow TS to have more effective type checking (as long as they're backwards compatible, of course, which would also be the case if TS got incorporated into JS), so I don't think it's necessary for that.
Performance improvements enabled by optimisations would be nice, but I believe I heard that no major gains would be expected there, especially compared to something like WASM.
> And the only reason to make that step optional, is to not run it, which is a worse experience.
Yeah, but it's an optional "worse" experience.
IOW, for those people who think it's a better experience (because there is no build step), they can opt into it. For those who think it's a worse experience, they can continue using a build-step.
Nothing changes for those who want a build-step.
(I agree about the standards process thing)
> Performance improvements enabled by optimisations would be nice, but I believe I heard that no major gains would be expected there, especially compared to something like WASM.
If TS is compiled in the browser, then it can be compiled to WASM, not to JS.
It's only when TS is compiled on the server that it has to be compiled to JS.
If the browser support TS natively, the compilation target would not be JS (which is what makes the resulting code slow), but either WASM, or naive code.
> a better experience (because there is no build step)
I feel like the way to make both of us happy is to just strip type annotations, rather than doing full type checking in the browser. Which, luckily, is already making its way through the standards track: https://tc39.es/proposal-type-annotations/
> If the browser support TS natively, the compilation target would not be JS (which is what makes the resulting code slow), but either WASM, or naive code.
I was not talking about compiling it to JS; I seem to recall that native support wouldn't necessarily result in big performance improvements. (Consider that JS is also not being compiled to JS.)
You'll still get all the strong typing without have to wait for it to run.
For example an error in a little used branch would cause an error before the branch even runs.