Here's the description of the "strict" compiler option that goes a long way to make your program more reliable :
"Enable all strict type checking options. Enabling --strict enables --noImplicitAny, --noImplicitThis, --alwaysStrict and --strictNullChecks."
Some of the soundness issues are by design (making directly overlayed JS code reasonable to type/interface with), some are due to 'taking too much time/design complexity if implemented'.
I won't go into them one by one, but you can jump into most of the issues here to get a sense:
https://github.com/Microsoft/TypeScript/issues?utf8=&q=Sound...
https://github.com/Microsoft/TypeScript/issues?utf8=&q=unsou...
I won't take a position on the matter other than to say that it's a tricky balance and the TS team are at the very least extremely clear in saying that you should not be looking for a high degree of soundness here (JS interop is a higher priority).
If those things are important, I would recommend Scala.js or PureScript or... anything more self contained (and remember that they're still ultimately dynamically typed on the JS boundary, just not inside the language).
Here's the TS team explaining their position themselves:
https://github.com/Microsoft/TypeScript/issues/9825#issuecom...
https://github.com/Microsoft/TypeScript/issues/13003 (not the most cordial commenter unfortunately)
Overall though, getting bitten by the unsoundness is fairly rare , imo.
I think I wrote a lot of code assuming that the type checker was sound and could check my abstractions, only to find that it wasn't and it couldn't.
If it were more sound, I would rely on the language to do more checking.
I don't get bitten as much any more, but I'm pretty sure that's mostly because I know what tradeoffs the language is making and I use other mechanisms to ensure safety.
And no, Flow is MUCH more sound, I have worked with both. Flow's limitations tend to be false negatives (code that should ideally work does not compile as is) whereas Typescript's are false positives (fails to detect type errors).
The killer app for type systems has always been code completion.
has always been code completion.
- Enabling ahead-of-time compilation resulting in efficient code.
- Guarantees on the absence of specific classes of bugs. (This requires soundness.)
- Aiding verification. Types provide a rough classification of program behaviour which drastically simplifies verification. If you don't have types, your program logic needs to make typing information explicit in assertions.
2. Type systems have always been super limited in the kinds of bugs they can eliminate, even with "soundness".
3. This sounds like a type theorists point that is completely detached from most programming experiences.
Anyway JS's popularity is orthogonal to questions about efficient compilation. Types drastically simplify those, for you can work out at compile time how much memory needs to be allocated for each sub-expression of the program under execution. If you don't have types, you can only do that at run-time, and need extremely complicated JIT compiler technology to make this moderately performant.
(2) Sure. Still extremely helpful, all the more so, the bigger the software. If you don't have types to do this, you will have to find those bugs by testing -- much more labour intensive.
(3) Detached, maybe, but still true. To paraphrase Sutton's law: I'm detached from the toils of contemporary bread/butter programmers and look at verification because that's where the interesting open problems are.