StrictNullChecks and throw/return with incorrect when used with closure scope

[tlong-dev]

TypeScript Version: 2.1.5

Code

function other() {
  return new Promise((resolve, reject) => {
    resolve("OK!");
  });
}

function test(hosts: {whatever: boolean}[]) {
  let host = hosts.filter((f) => f.whatever).shift();

  if (!host) {
    throw new Error("Crap!");
  }

  return other().then(() => {
    console.log(`${host.whatever} selected`);
  });
}

Playground

Expected behavior:

Do not emit an error about host possibly being null, since the method exits early if that's the case.

Actual behavior:

Typically when using strictnullchecks Typescript is intelligent enough to understand that when you throw an error from a method based on a type check/guard it narrows the type after the statement. As indicated in this short example:

function test(hosts: {whatever: boolean}[]) {
  let host = hosts.filter((f) => f.whatever).shift();

  // Correctly calls out hose as possibly being null
  console.log(`${host.whatever} selected`);
  
  if (!host) {
    throw new Error("Crap!");
  }

  // Does not complain here because of the above check/throw
  console.log(`${host.whatever} selected`);
}

For some reason when you add the return of the function and use host through a closure it starts complaining.

I'm sure it's the closure, but I don't see how this could error.

[tlong-dev]

Note: re-assigning the variable to a new variable after the throw seems to re-bind the type information and solves the issue... Seems like a bug.

function other() {
  return new Promise((resolve, reject) => {
    resolve("OK!");
  });
}

function test(hosts: {whatever: boolean}[]) {
  let host = hosts.filter((f) => f.whatever).shift();

  if (!host) {
    throw new Error("Crap!");
  }

  let ho = host;  // WAVES MAGIC WAND

  return other().then(() => {
    console.log(`${ho.whatever} selected`);
  });
}

[HerringtonDarkholme]

#9998 / #11498

Changing let host to const host also solves this.

[tlong-dev]

Wow, the perfect comment!

Found the threads I knew existed by couldn't locate in search, and offered clean workaround!

Thanks. Closing in favor of #9998 and #11498