What is the difference between then and finally in

2019-02-18 06:03发布

问题:

I see the docs for Bluebird's finally but I still don't quite understand the difference vs. then.

To be clear: I know exactly why then gets called after a catch. I want it to be called after a catch. That's the intent. My question is: If I want code to always be executed regardless of the promise state, what's the difference between then vs finally for that?

I built this test:

var Promise = require("bluebird");

function test1 () {
    console.log("RESOLVE + THEN + CATCH + THEN");
    return new Promise((resolve, reject) => resolve())
       .then(() => console.log("then"))
       .catch(err => console.log("error:", err.message))
       .then(() => console.log("end"));
}

function test2 () {
    console.log("REJECT + THEN + CATCH + THEN");
    return new Promise((resolve, reject) => reject(new Error("rejected")))
       .then(() => console.log("then"))
       .catch(err => console.log("error:", err.message))
       .then(() => console.log("end"));
}

function test3 () {
    console.log("RESOLVE + THEN + CATCH + FINALLY");
    return new Promise((resolve, reject) => resolve())
       .then(() => console.log("then"))
       .catch(err => console.log("error:", err.message))
       .finally(() => console.log("end"));
}

function test4 () {
    console.log("REJECT + THEN + CATCH + FINALLY");
    return new Promise((resolve, reject) => reject(new Error("rejected")))
       .then(() => console.log("then"))
       .catch(err => console.log("error:", err.message))
       .finally(() => console.log("end"));
}

// run tests "sequentially" so console output doesn't get blended
setTimeout(test1, 500);
setTimeout(test2, 1000);
setTimeout(test3, 1500);
setTimeout(test4, 2000);

This tests four cases:

  1. .then(...).catch(...).then(...) with a resolved promise.
  2. .then(...).catch(...).then(...) with a rejected promise.
  3. .then(...).catch(...).finally(...) with a resolved promise.
  4. .then(...).catch(...).finally(...) with a rejected promise.

The results I see are cases 1+2 behaves identically to 3+4: The last bit (then or finally depending on test) executes no matter what happens before it, as intended. The output of that program is:

RESOLVE + THEN + CATCH + THEN
then
end
REJECT + THEN + CATCH + THEN
error: rejected
end
RESOLVE + THEN + CATCH + FINALLY
then
end
REJECT + THEN + CATCH + FINALLY
error: rejected
end

Now, the reason I'm asking is because I saw a comment on this other question I asked:

Not sure if your promises support it, but you should change the last .then to .finally so that the busy always gets cleared.

From my very limited knowledge of then, and the tests above, it seems like then is sufficient. But after that comment I'm questioning myself and the safety of using then to execute "finally" code.

So my question is: What's the difference between then and finally? They look like they behave the same, but when would I need to use finally instead of then?

回答1:

First difference: Sometimes you don't want to catch errors at the place they arise, but in the code that uses this function, so you don't catch them. In that case you can't substitute then() and finally().

Sometimes you have to clean something up whether there was an error or not (nulling references, clearing timeouts ... stuff like that). That's where you use finally().

Second difference: The function you pass to catch() could also throw, then you would have a rejected Promise and the following then() would not be called.

(so a finally before a catch will still execute on an error, didn't know that)

Yeah, that's the point of finally(). It will be executed under any circumstance without changing the resolved value.

You might want to read/google a bit about try {} finally {}, without catch.



回答2:

.then and .finally are not the same.

.then is the main promise primitive. It's what is defined, thoroughly, in the Promises/A+ spec and all promise libraries will implement it.

A Bluebird .finally handler will "be called regardless of the promise's fate". So an unhandled exception still triggers a .finally.

new Promise((resolve, reject) => reject(false))
  .finally(a => console.log('finally', a))
// finally undefined
// Unhandled rejection false

new Promise((resolve, reject) => reject(false))
  .then(a => console.log('then', a))
// Unhandled rejection false

.finally won't change the resolved value of the promise and doesn't receive the result of the promise chain.

new Promise((resolve, reject) => reject(false))
  .catch(e => {
    console.log(e)
    return 2
  })
  .finally(a => {
    console.log('finally', a)
    return 1
  })
  .then(res => console.log('res', res))
// finally undefined
// res 2

The methods look similar in your test cases as the tests catch all errors and you are only using promises for flow control, not relying on the values being resolve/rejected along the promise chain.



回答3:

All right, after some chatting and a lot of help from KevinB, I figured out at least one difference. Consider the following two new tests:

function test5 () {
    console.log("REJECT + THEN + CATCH/THROW + THEN");
    return new Promise((resolve, reject) => reject(new Error("rejected")))
       .then(() => console.log("then"))
       .catch(function(err) { throw new Error("error in catch"); })
       .then(() => console.log("end"));
}

function test6 () {
    console.log("REJECT + THEN + CATCH/THROW + FINALLY");
    return new Promise((resolve, reject) => reject(new Error("rejected")))
       .then(() => console.log("then"))
       .catch(function(err) { throw new Error("error in catch"); })
       .finally(() => console.log("end"));
}

In these the promise is rejected, but an error is thrown from catch.

The promise ends up ultimately rejected in both cases but for the finally case the finally is still executed, the then is not.

So that's the difference. They're almost the same with the sole exception that when an error is thrown from the catch handler, finally executes, and then does not.

This means the comment I quoted also does have merit: If in my error handler another error occurred, a then wouldn't guarantee clean-up, but a finally would. That's the case I was missing.