FIX wrong success for an erronous fetch, after an abort of a legit fetch

[basos9]

This bug can be manifested when a pending fetch of a legit resource (200 response) is aborted and afterwards a new fetch that leads to error (e.g. 404) is issued. At the time, the second request finishes the lastValue has the aborted data of the previous fetch (callback is executed even on aborted), YIELDING the second (normally erroneus response) to be signaled as successfull.

The effect of this bug is demonstrated with a test, One can remove the jsonp patch and see what happens.

[jaubourg]

Isn't this due to the fact browsers won't stop the execution of a script when the script tag is removed from the DOM?

[basos9]

Yes, this seems to be the basic cause. Of course this applies when the callback name is the same across requests. If it is random the odds decrease. I think that this fix leaves a corner case, but it definitely narrows the error cases. For example; this will work (with the fix), 1.request, 2. abort, 3. aborted 200 response (modifies lastValue), 4. request, 5. response 404, 6. notify error (since lastValue is reset at 4). But this still does not work 1. request, 2.abort, 3. request (reset lastValue), 4. aborted 200 response (modifies lastValue), 5. response 404, 6. notify success (since lastValue is considered legit from 4). But i think that the last case is more rare or maybe impossible. The last will be valid if one thing that I'm not sure if it applies in this case, is valid: script execution is done in the order included and serially (anyway). This applies at least for handwritten script tags. But I think it should apply for dynamic scripts as well.