Gutenberg: Dirty state is still cleared if post is edited while prolonged save request is made

[westonruter]

Issue Overview

If you make a change to a post, then press update and the update request is slow to respond (either due to connectivity or server issues), and make another change while isPostSaving then when the response finishes the dirty state is erroneously cleared and I am unable to save the new changes I made the save request was pending.

Steps to Reproduce

Tested using master at ab7a429db76c3484da91012714648f592b09fae0.

1. To aid with testing, add sleep(5) to the top of \WP_REST_Posts_Controller::update_item().
2. Open a post draft.
3. Make a change to a post.
4. Hit Save Draft
5. While the save request is happening make another change to the post.
6. Once the save request finishes, try to save the change you just made.

Expected Behavior

The Save Draft button should be re-shown to the user after the save request completes when a change was made during the request. Additionally, there should be the “Are you sure?” dialog when leaving the page at this point so that the user does not lose their changes.

Possible Solution

Introduce a new state flag for isEditedWhileSaving and if when the response completes, check this before resetting isDirty to false.

Screenshots / Video

Screencast of issue: https://youtu.be/HLF7VTd8gjY

[aduth]

Discussed in Slack: https://wordpress.slack.com/archives/C02QB2JS7/p1523341790000232

[aduth]

Should note that this was explicitly accounted for in the original implementation, and if problematic, is a sign of regression.

Previously: #1093, #1092

[aduth]

Actually, I think it's more to do with iterations that have occurred over time in dirty detection: #1996, #3298

The approach in #1996 worked because state.editor.edits unsets values now persisted from its RESET_POST handling, so the selector was, in simplest terms, checking whether any keys existed here (slightly more complex, but general idea).

The withChangeDetection higher-order reducer introduced in #3298 is not so capable however, since it merely resets dirtiness upon a RESET_POST.

[aduth]

Aside: This is a good candidate for E2E testing, since it's very technically challenging and easy to overlook in general usage, and also easy to reproduce in an automated environment.