[fixed] Abort pending transition if user navigates away

[gaearon]

Previously, navigating (by clicking Back button or a link) while a transition in progress would put router into an inconsistent state.

Now, a pending transition will be aborted by the router if user navigates away while it was pending.

Fixes #479

[gaearon]

I haven't written test cases for this, here's what I intended to do.
(Will need to test each case again, as I may have broken something midway through it.)

A -> slowly resolving B

• Go. Path will change to /b but it's not ready yet. Now press Back.

Before fix: when B is ready, will navigate from A to B.
After fix: when B is ready, will stay at A because user undecided [sic] going to B.

• Go. Path will change to /b but it's not ready yet. Now press link to C.

Before fix: when B is ready, will navigate from C to B.
After fix: when B is ready, nothing happens because user wanted C. If you press Back now, you will land at A.

• Go. Path will change to /b but it's not ready yet. Now press Back twice.

Before fix: when B is ready, will navigate from A to B.
After fix: when B is ready, will stay at whatever came before A.

A -> slowly resolving and then aborting B

• Go. Path will change to /b but it's not ready yet. Now press Back.

Before fix: when B is ready and aborted, will pop from A to whatever came before A.
After fix: when B is ready and aborted, will stay at A because user wanted to pop B and not A.

• Go. Path will change to /b but it's not ready yet. Now press link to C.

Before fix: when B is ready and aborted, will pop from C to A.
After fix: when B is ready and aborted, nothing happens because user wanted C. If you press Back now, you will land at A.

• Go. Path will change to /b but it's not ready yet. Now press Back twice.

Before fix: when B is ready and aborted, will pop yet further!
After fix: when B is ready and aborted, will stay at whatever came before A.

A -> slowly resolving and then redirecting B

Very similar to first case.

[gaearon]

@rpflorence @mjackson

I’m afraid I don’t have time to write tests for this today, but I verified those scenarios I describe and they work. We need to see if all react-router examples work correctly, and if they do, I think it’s still better to bring this in without tests because previous racing behavior was broken anyway.

[ryanflorence]

f1c84b8

[gaearon]

[gaearon]

@mjackson

I cleaned it up, would you take a look?

I'll be adding tests for now..

[gaearon]

Now that I'm writing tests, I'll write tests for abort as well (they're missing).

[gaearon]

Added tests for basic async transitions (not not the corner cases I'm describing here yet).
Will push corner cases in a couple of hours.

[mjackson]

@gaearon This LGTM so far. Thank you for writing tests as well!

[gaearon]

I added more tests. They cover scenarios I mentioned above.

[mjackson]

Should we have one more step here that basically just asserts "Foo" is still rendered after the delay and calls done?

[mjackson]

Thanks @gaearon ! This looks great. Hopefully I made some useful comments.

[ryanflorence]

@gaearon, this project is way better because of you, thanks :)

[gaearon]

@rpflorence I'm glad to be of use!