But like a good skeptic I wanted to reserve my opinion until I had a chance to actually use it. This was the same approach which I took with CoffeeScript, you don't really know something until you've made something with it (and for the record I wasn't particularly fussed by CoffeeScript).
Well I decided to do this, I wrote a small library called mathy and I wanted to share some of my thoughts from having made something with it. Keep in mind this is only a small library so it's not exactly extensive, but I feel it's a good start.
I just want to clarify a few things about how I did this project:
- I used Sublime Text 2 as my editor, not Visual Studio
- This is written primarily as a Node.js package but I also want it to work in the browser
- The compiler errors can be nice, caught a few spelling errors and API usage errors through them which I wouldn't have caught until runtime/while the tests were being executed
- Being able to create modules using a keyword is good, save a bunch of boilerplate guff
- I've used a class for part of the API which is very handy and easy to use, but most importantly it’s syntactically simple
- It doesn't try and stop me from writing any of the funky stuff that I actually want to write ;)
=>is really sweet, I’d never really got into CoffeeScript enough to have built anything much but I can see why those guys rave about it
- If you're not using Visual Studio it’s kind of a pain, Sublime Text 2 only has syntax highlighting support so you don't exactly get much benefit, no intellisense or anything
- The fact you can't programmatically use the compiler sucks. Even though TypeScript's compiler is available as a Node.js package you have to execute the compiler yourself and pass in the input. It'd be nicer if you could just
require('typescript')instead of what I have to do in the Makefile
- The way modules are generated can be a real pain if you're trying to target the browser and Node.js. If you want to go with CommonJS the internal modules they generate create a global variable to store your object in, but that won’t work in Node.js as the variable isn't exported. If you make it a public module it assumes there is a public “exports” object to attach to which is fine in Node.js but sucks in the browser! I had to have a shitty implementation to get it working that assumes in the browser there is the
exportsobject. You can use the AMD support but it forces you to use RequireJS (or CurlJS or any other loader). It'd be nicer if there was optional AMD support, like how I have it done in db.js, so you can have something that will export as a AMD if AMDs are available, otherwise just be a global object
- There seems to be no way to plug a definition file into Node.js so all my unit tests are just written in plain JS which really sucks as I changed the API and didn't realise until every single unit test failed
The really shitty
- The generated code isn't in a closure scope unless you use a module, which in turn the closure scope is really ridged (particularly from the CommonJS module point of view) which means
- You can't create interfaces in a function
- You can't create classes in a function
- You can't create modules in a function (ok, I kind of get this one)
So with all that considered what's my thoughts so far? I actually do like it, particularly the way the compiler can catch some stupid mistakes. I'd like to try it on a much bigger project, particularly something in Visual Studio to see how well that goes, especially by having things like intellisense working.
The biggest problem I see is the module system, it's really shit if you don't do exactly as Microsoft does, and here in lies the problem. If you want to load in third party libraries it really doesn't work too nicely, you end up with
any type declarations around which really isn't helpful.