Block Editor: Fix undo level inconsistencies. #17259

epiqueras · 2019-08-29T19:32:06Z

Description

This PR fixes a couple of undo level related inconsistencies that surfaced after merging #16932.

In the Block Editor, withPersistentBlockChange marks any change to an attribute different than the last, as persistent. However, it considers all actions for this comparison, so an explicit persistence change triggered by a timeout like the one that fires after not typing for a while, would break this "chain of comparison", and the next change would be marked as persistent even if targeted at the same attribute as the one before the explicit persistence change. This resulted in scenarios where typing "Hello", waiting a bit, then typing "World", would result in 4 undo levels instead of 3. Namely, there would be an extra unnecessary undo level in the middle, requiring an extra click to undo. This PR fixes this by ignoring explicit persistence changes when comparing the last two attribute changes.

In the Editor, the decision was made to consider initial edits as dirtying operations that create an undo level and enable saving functionality. For some reason, the server sends an empty initial edit with empty strings for the title, excerpt, and content of new posts. This results in another unnecessary undo level. This PR fixes this by ignoring initial edits that would not change any values of the post.

How has this been tested?

It was verified that typing two words with a pause in between does not create an undo level and that new posts don't load with an undo level.

Types of Changes

Bug Fix: Stop inserting unnecessary undo levels between words an on empty posts.

Checklist:

My code is tested.
My code follows the WordPress code style.
My code follows the accessibility standards.
My code has proper inline documentation.
I've included developer documentation if appropriate.

ellatrix · 2019-08-30T15:05:54Z

@epiqueras I added some e2e tests that should pass, and do pass before #16932. They are still failing with these changes though. I think one undo level has been removed, but for some reason there's still one being created with the first letter.

epiqueras · 2019-08-30T15:09:09Z

That's because the Block Editor is marking it as persistent (calling onInput). That was somehow ignored before. Could we make the Block Editor not do that?

ellatrix · 2019-08-30T15:15:43Z

Sure. Where is it doing that?

epiqueras · 2019-08-30T15:25:23Z

with the first letter.

When you type that first character.

ellatrix · 2019-09-02T06:56:31Z

@epiqueras I meant in the code :)

epiqueras · 2019-09-02T18:15:34Z

No idea, I thought you might know, because of your RichText/writing flow expertise.

ellatrix · 2019-09-03T06:51:32Z

@epiqueras RichText doesn't mark the first character as persistent. It continuously calls onChange and createUndoLevel after 1s of no typing. The latter should eventually move somewhere else, so it only calls onChange. The only other case it creates an undo level is when it automatically changes the content based on an input rule, which doesn't seem to be the case here.

epiqueras · 2019-09-03T17:48:19Z

So, the issue is probably coming from somewhere else in the block editor.

In your tests, you are also seeing the first character passed through onInput right?

epiqueras · 2019-09-03T20:45:17Z

@ellatrix

Well, it follows this logic:

gutenberg/packages/block-editor/src/store/reducer.js

Lines 341 to 350 in dd7782a

    
           /** 
        
            * Higher-order reducer intended to augment the blocks reducer, assigning an 
        
            * `isPersistentChange` property value corresponding to whether a change in 
        
            * state can be considered as persistent. All changes are considered persistent 
        
            * except when updating the same block attribute as in the previous action. 
        
            * 
        
            * @param {Function} reducer Original reducer function. 
        
            * 
        
            * @return {Function} Enhanced reducer function. 
        
            */

The first character's edit is not updating the same attribute as the last edit right?

How should we go about this?

I think the logic behind MARK_LAST_CHANGE_AS_PERSISTENT is kind of flawed. It should be a part of the next edit, not a standalone action. So when you type and stop for a while, the next edit will mark the last edit as persistent. With the current flow, the last edit gets marked as persistent, and the next character that comes in is also marked as persistent.

Also, do you have an example of a release where this worked correctly?

ellatrix · 2019-09-05T10:15:28Z

git checkout 256a6745354c80a63de53c62dc0ec107eed00c30 (the commit before merging #16932). Everything worked fine there.

ellatrix · 2019-09-15T12:31:10Z

Should we not make sure this all works before the last Gutenberg release of WP 5.3?

epiqueras · 2019-09-16T01:44:21Z

Should we not make sure this all works before the last Gutenberg release of WP 5.3?

@ellatrix I think we can merge this now with the latest changes?

… against the last action in the persistent block change HOR.

ellatrix · 2019-09-16T06:24:18Z

Rebased the branch as I was getting some PHP errors with the current checkout of WP core.

ellatrix · 2019-09-16T06:49:14Z

Looks way better now.

Still seeing some inconsistencies.

Major: Create a new post. Create a new paragraph and type something in it. Save the draft. Undo the last change. The typing isn't undone. Instead, it seem like the saving is "undone"? The "Save Draft" button appears again, even though the content didn't change.
~~Minor: When I open the demo content, the undo button is active, even though I didn't touch the content. This wasn't the case before.~~

ellatrix

Functionally it looks good. Would be great if the last e2e test case that I added could be fixed. Codewise I don't have any objections, but would be great if someone familiar with the core-data package could provide a second review.

ellatrix · 2019-09-16T07:26:10Z

packages/core-data/src/reducer.js

 			} );
-
+			const comparisonUndoEdits = Object.values( action.meta.undo.edits ).filter( ( edit ) => typeof edit !== 'function' );


What is this checking for? Would be great to inline comment.

That the last two edits are not the same.

I'm more wondering what the filter is for. :)

When an edit is a function it's an optimization to avoid running some expensive operation. We can't rely on the function references being the same so we opt out of comparing them here.

That could use an inline comment.

I'm working on it with another fix.

ellatrix · 2019-09-16T07:28:58Z

packages/editor/src/store/actions.js

+		edits &&
+		Object.keys( edits ).some(
+			( key ) =>
+				edits[ key ] !== ( has( post, [ key, 'raw' ] ) ? post[ key ].raw : post[ key ] )


The raw key seems very API specific. Given all the API abstractions, is this the right place to check for the raw key?

Yeah, it's not ideal. I think we should deprecate initial edits if possible.

ellatrix · 2019-09-16T07:30:06Z

I think the logic behind MARK_LAST_CHANGE_AS_PERSISTENT is kind of flawed. It should be a part of the next edit, not a standalone action. So when you type and stop for a while, the next edit will mark the last edit as persistent. With the current flow, the last edit gets marked as persistent, and the next character that comes in is also marked as persistent.

So this is all fine in the end? Anything that should be changed?

ellatrix · 2019-09-16T07:32:30Z

Since this seems more or less ready, and fixing important issues, I'd suggest we include it in Gutenberg 6.5.

ellatrix · 2019-09-16T08:42:10Z

420ead2 adds a test case for the third item listed in this comment: #16932 (comment). This PR fixes the issue.

youknowriad · 2019-09-16T11:45:03Z

For the demo content being dirty. That was a decision we made on purpose after some discussions with @jasmussen and @epiqueras

ellatrix · 2019-09-16T11:59:07Z

Ok sure, I agree, that's just a minor thing.

ellatrix · 2019-09-16T12:01:35Z

I'm fine with reverting the test case introduced in d1817b6 for now if it would take too long to fix. It's good to have the other fixes merged.

ellatrix · 2019-09-16T12:07:36Z

With that test case I meant the first item in #17259 (comment), not the demo content one. Would still be great to fix for 5.3 separately.

ellatrix

In my opinion, feel free to merge. :)

epiqueras · 2019-09-16T14:37:30Z

@ellatrix This fixes the other issue: #17452.

epiqueras added [Package] Editor /packages/editor [Package] Block editor /packages/block-editor labels Aug 29, 2019

epiqueras added this to the Future milestone Aug 29, 2019

epiqueras requested review from ellatrix, talldan and youknowriad as code owners August 29, 2019 19:32

epiqueras self-assigned this Aug 29, 2019

epiqueras requested review from ajitbohra, gziolo, nerrad and ntwb as code owners August 30, 2019 14:53

This was referenced Sep 7, 2019

Overview: Custom Entity Sources #17368

Closed

Increase Coverage of Undo/History E2E Tests #17434

Closed

epiqueras force-pushed the fix/undo-level-inconsistencies branch 2 times, most recently from d195dae to a40ce33 Compare September 16, 2019 00:37

epiqueras and others added 3 commits September 16, 2019 09:22

Block Editor: Skip explicit persistent marking actions when comparing…

de3559a

… against the last action in the persistent block change HOR.

Editor: Ignore initial edits that wouldn't change post values.

ba9759c

Add e2e tests that should pass

0e2d6be

Core Data: Fix undo approach.

e816e71

ellatrix force-pushed the fix/undo-level-inconsistencies branch from a40ce33 to e816e71 Compare September 16, 2019 06:23

Add e2e test case for undo level creation on save

d1817b6

ellatrix reviewed Sep 16, 2019

View reviewed changes

ellatrix modified the milestones: Future, Gutenberg 6.5 Sep 16, 2019

ellatrix added the [Release] Do Not Punt Used to indicate the issue or pull request must not be moved from the assigned milestone label Sep 16, 2019

Add passing e2e that would fail before

420ead2

Revert d1817b6

7c7239d

ellatrix approved these changes Sep 16, 2019

View reviewed changes

epiqueras merged commit 5fa136b into master Sep 16, 2019

epiqueras deleted the fix/undo-level-inconsistencies branch September 16, 2019 13:53

epiqueras mentioned this pull request Sep 16, 2019

Editor: Ignore save edit in undo history. #17452

Merged

5 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Block Editor: Fix undo level inconsistencies. #17259

Block Editor: Fix undo level inconsistencies. #17259

epiqueras commented Aug 29, 2019

ellatrix commented Aug 30, 2019

epiqueras commented Aug 30, 2019

ellatrix commented Aug 30, 2019

epiqueras commented Aug 30, 2019

ellatrix commented Sep 2, 2019

epiqueras commented Sep 2, 2019

ellatrix commented Sep 3, 2019

epiqueras commented Sep 3, 2019

epiqueras commented Sep 3, 2019

ellatrix commented Sep 5, 2019

ellatrix commented Sep 15, 2019

epiqueras commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019 •

edited

Loading

ellatrix left a comment

ellatrix Sep 16, 2019

epiqueras Sep 16, 2019

ellatrix Sep 16, 2019

epiqueras Sep 16, 2019

mcsf Sep 16, 2019

epiqueras Sep 16, 2019

epiqueras Sep 16, 2019

ellatrix Sep 16, 2019

epiqueras Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019

youknowriad commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix left a comment

epiqueras commented Sep 16, 2019

		} );

		const comparisonUndoEdits = Object.values( action.meta.undo.edits ).filter( ( edit ) => typeof edit !== 'function' );

Block Editor: Fix undo level inconsistencies. #17259

Block Editor: Fix undo level inconsistencies. #17259

Conversation

epiqueras commented Aug 29, 2019

Description

How has this been tested?

Types of Changes

Checklist:

ellatrix commented Aug 30, 2019

epiqueras commented Aug 30, 2019

ellatrix commented Aug 30, 2019

epiqueras commented Aug 30, 2019

ellatrix commented Sep 2, 2019

epiqueras commented Sep 2, 2019

ellatrix commented Sep 3, 2019

epiqueras commented Sep 3, 2019

epiqueras commented Sep 3, 2019

ellatrix commented Sep 5, 2019

ellatrix commented Sep 15, 2019

epiqueras commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019 • edited Loading

ellatrix left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019

youknowriad commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix commented Sep 16, 2019

ellatrix left a comment

Choose a reason for hiding this comment

epiqueras commented Sep 16, 2019

ellatrix commented Sep 16, 2019 •

edited

Loading