Rule suggestion: restrict + operands

[myitcv]

With reference to the binary + operator in the spec

I find it rather disconcerting that this is allowed:

let x = 5 + "test";
let y = true + "test";

Would there be interest in a rule that forces the two operands to + to either:

• both be of type string
• both be of type number

and fail otherwise?

Do we have the necessary type information available during linting to do this?

[myitcv]

I've made a start, but just run into a problem

[adidahiya]

sounds like a good lint rule, thanks for looking into this!

[adidahiya]

TSLint currently only gets type information on a per-file basis, so to do this properly, I think we're blocked on #680

[myitcv]

Ah. Is that why the TypeChecker returns Any?

[myitcv]

Sorry, to be more explicit: that question relates to the TypeScript issue I created

[adidahiya]

No, I'm not sure why the type checker returns any in that issue you created -- intuitively it feels like the checker has enough information about the built-in .toString() to report the correct return type.

Perhaps we're not instantiating the checker correctly?

[myitcv]

Hmm, having done a quick test I think it is related. In the following example the TypeChecker (in the way we are using it with a single file) reports the type of Banana() is Any:

import {Banana} from "./somewhere_else";

// Banana has type signature: Banana(): string

let y = "test" + Banana();

My guess would be that when it hits an identifier defined outside of the scope of what is being checked it assumes an Any type

[jkillian]

Right, there's obviously no way for the TypeChecker to know what Banana() is if it's only given that one file. Maybe the same thing applies for the built-in definitions: possibly since lib.core.d.ts isn't included in the program it has no awareness of core typings.

[myitcv]

@jkillian exactly. So #680 will definitely allow us to write such rules. Look forward to it landing!

[alexeagle]

This one is approved to be implemented in tsc now. microsoft/TypeScript#7989 (comment)

[myitcv]

@alexeagle but only if you use --strictNullChecks

[adidahiya]

@alexeagle thanks for the link. Agreed with @myitcv that it can still be a TSLint rule that is a more granular opt-in than --strictNullChecks in the compiler.

[OliverJAsh]

Can this be closed now we have https://palantir.github.io/tslint/rules/restrict-plus-operands/?