Added memoization test for interrupted low-priority renders #8645

bvaughn · 2016-12-27T15:16:59Z

Added a test to verify that a high-priority update can reuse the children from an aborted low-priority update if shouldComponentUpdate returns false.

(Note that this PR is now different from the original. The original description is preserved below so that the subsequent discussion on this PR makes sense.)

Previous Description (no longer valid)

this.props and this.context are not reset when a render is interrupted mid-way and overridden by a subsequent, higher-priority update. The result is that a subsequent calls to shouldComponentUpdate will yield unexpected results¹ when comparing this.props and this.context to the provided nexProps and nexContext.

This PR captures the behavior in a (failing) unit test. Posting for discussion purposes.

¹ It is unclear if this is intentional behavior. I don't think it causes any problems but it was not what I expected initially.

bvaughn · 2016-12-27T15:20:58Z

FYI @gaearon, this is the failing test we were discussing on PR #8631.

The same failure highlighted in this test applies to context as well as props but since context is already in a broken state in master the test focuses on props only.

Do you think it would be useful for me to expand the test to include a check for context as well (or to add a second test for that)? I am kind of on the fence about it.

gaearon · 2016-12-27T16:46:47Z

Hmm maybe I misunderstand but I thought this was intentional. You get "last attempted" as prev in sCU but "last rendered" as prev in cDU. Is this a problem in some scenario?

bvaughn · 2016-12-27T17:15:35Z

Hmm. It's possible it is intentional and I just misunderstood the intent. That would make sense too as a behavior. I suppose I just assumed symmetry of before and after values between those methods (cWRP/sCU/cWU/cDU).

You get "last attempted" as prev in sCU but "last rendered" as prev in cDU

I'm not sure people think of this.props as "last attempted" but I could be wrong. (I don't think most people will be thinking of render attempts or bail-outs to begin with.)

Either way, I guess you're right. I don't think it would cause problems unless users set some sort of expectations in cWRP/sCU/cWU that were checked in cDU which seems unlikely. Looking at the different ways in which context or props might change (or not change) between an interrupted render and the next potential render- I think the current behavior is okay so long as it is intentional.

Maybe @sebmarkbage can clarify or point out a case I'm not thinking about. We talked about this briefly last week in the context of a larger discussion about context but now I'm second-guessing my understanding of this part of that conversation. 😄

If this is indeed the expected behavior then perhaps I could modify my test slightly (so that it passes) and capture it as such.

acdlite · 2016-12-29T22:36:48Z

Thought about this for a bit.

The mental model for shouldComponentUpdate is memoization. But I think it's a memoization of workInProgress.child, not current.child. If sCU returns false, that only means that we don't need to recompute our children. It does not mean that those children were already flushed.

To put this in more practical terms, if a high priority update interrupts a low priority update before it is flushed, we should have the ability to reuse the children computed during the low priority update. Which we can't do if we reset the pointers. I don't think we have a test for this, but we should.

Care to write one? :)

acdlite

Change test to check that a high-priority update can reuse the children from an aborted low-priority update using shouldComponentUpdate.

acdlite · 2016-12-29T22:45:46Z

I don't think it would cause problems unless users set some sort of expectations in cWRP/sCU/cWU that were checked in cDU which seems unlikely.

I agree that this is unlikely. A better name for shouldComponentUpdate might be areInputsEqual, since that makes no promises about the implementation details of memoization.

acdlite · 2016-12-30T02:13:03Z

^ The test I describe above happens to be related to a PR I was already working on, so I added it there. https://github.com/facebook/react/pull/8655/files#diff-6a068060b8f49a9d61ae16cac3be4a9b

bvaughn · 2016-12-31T00:25:31Z

I don't think we have a test for this, but we should.
Care to write one? :)

Wouldn't that essentially be the test I've added, if I only changed an expectation or two (and maybe added a render counter)? That's why I left the previous comment:

If this is indeed the expected behavior then perhaps I could modify my test slightly (so that it passes) and capture it as such.

I'll update the test in this PR.

bvaughn · 2016-12-31T00:38:36Z

Thanks for the input Dan and Andrew. I've updated this PR to capture what I think has been agreed on as the expected/correct behavior (a new passing test instead of a failing one).

sebmarkbage · 2016-12-31T03:04:36Z

src/renderers/shared/fiber/__tests__/ReactIncremental-test.js

+    // Renders: Root
+    ReactNoop.render(<Root>B</Root>);
+    ReactNoop.flushDeferredPri(20);
+


Should you be asserting that what you expected happened? Otherwise you can merge the call below into this one.

sebmarkbage · 2016-12-31T03:06:12Z

src/renderers/shared/fiber/__tests__/ReactIncremental-test.js

+    expect(scuPrevProps.children).toEqual('A');
+    expect(scuNextProps.children).toEqual('B');
+    expect(cduPrevProps).toEqual(null);
+    expect(cduNextProps).toEqual(null);


These could have been called with null if there was a bug. I tend to prefer pushing to an array since that covers that and tests execution order.

Good suggestion.

acdlite · 2017-01-13T21:41:19Z

src/renderers/shared/fiber/__tests__/ReactIncremental-test.js

@@ -2184,4 +2184,87 @@ describe('ReactIncremental', () => {
      'componentDidUpdate',
    ]);
  });
+
+  it('should reuse memoized work after an interrupted low-priority render', () => {


Let's rename the test so it's clear that this is specifically about updating the pointers before calling lifecycles.

acdlite

👍

Test added to verify that a high-priority update can reuse the children from an aborted low-priority update if shouldComponentUpdate returns false.

facebook-github-bot added the GH Review: review-needed label Dec 27, 2016

bvaughn assigned acdlite Dec 27, 2016

bvaughn force-pushed the add-failing-scu-current-props-test branch from 9ee0624 to 382fd04 Compare December 27, 2016 17:13

bvaughn changed the title ~~Add failing test for sCU current props~~ Add failing test for unexpected prevProps passed to cDU Dec 27, 2016

bvaughn changed the title ~~Add failing test for unexpected prevProps passed to cDU~~ Add failing test for unexpected this.props value Dec 27, 2016

facebook-github-bot added the CLA Signed label Dec 29, 2016

acdlite requested changes Dec 29, 2016

View reviewed changes

acdlite added GH Review: needs-revision and removed GH Review: review-needed labels Dec 29, 2016

bvaughn force-pushed the add-failing-scu-current-props-test branch from 382fd04 to d814ef5 Compare December 31, 2016 00:37

facebook-github-bot added GH Review: review-needed and removed GH Review: needs-revision labels Dec 31, 2016

bvaughn changed the title ~~Add failing test for unexpected this.props value~~ Added memoization test for interrupted low-priority renders Dec 31, 2016

sebmarkbage reviewed Dec 31, 2016

View reviewed changes

bvaughn force-pushed the add-failing-scu-current-props-test branch from d814ef5 to 0e954f9 Compare December 31, 2016 04:40

bvaughn force-pushed the add-failing-scu-current-props-test branch from 0e954f9 to e95aaa7 Compare January 13, 2017 21:34

acdlite reviewed Jan 13, 2017

View reviewed changes

acdlite approved these changes Jan 13, 2017

View reviewed changes

acdlite added GH Review: accepted and removed GH Review: review-needed labels Jan 13, 2017

Added memoization test for interrupted low-priority renders

2fb07b9

Test added to verify that a high-priority update can reuse the children from an aborted low-priority update if shouldComponentUpdate returns false.

bvaughn force-pushed the add-failing-scu-current-props-test branch from e95aaa7 to 2fb07b9 Compare January 13, 2017 21:46

bvaughn merged commit c359df7 into facebook:master Jan 13, 2017

bvaughn deleted the add-failing-scu-current-props-test branch January 13, 2017 21:52

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Added memoization test for interrupted low-priority renders #8645

Added memoization test for interrupted low-priority renders #8645

bvaughn commented Dec 27, 2016 •

edited

Loading

bvaughn commented Dec 27, 2016 •

edited

Loading

gaearon commented Dec 27, 2016

bvaughn commented Dec 27, 2016 •

edited

Loading

acdlite commented Dec 29, 2016 •

edited

Loading

acdlite left a comment

acdlite commented Dec 29, 2016

acdlite commented Dec 30, 2016

bvaughn commented Dec 31, 2016

bvaughn commented Dec 31, 2016

sebmarkbage Dec 31, 2016

sebmarkbage Dec 31, 2016

bvaughn Dec 31, 2016

acdlite Jan 13, 2017

acdlite left a comment

Added memoization test for interrupted low-priority renders #8645

Added memoization test for interrupted low-priority renders #8645

Conversation

bvaughn commented Dec 27, 2016 • edited Loading

Previous Description (no longer valid)

bvaughn commented Dec 27, 2016 • edited Loading

gaearon commented Dec 27, 2016

bvaughn commented Dec 27, 2016 • edited Loading

acdlite commented Dec 29, 2016 • edited Loading

acdlite left a comment

Choose a reason for hiding this comment

acdlite commented Dec 29, 2016

acdlite commented Dec 30, 2016

bvaughn commented Dec 31, 2016

bvaughn commented Dec 31, 2016

sebmarkbage Dec 31, 2016

Choose a reason for hiding this comment

sebmarkbage Dec 31, 2016

Choose a reason for hiding this comment

bvaughn Dec 31, 2016

Choose a reason for hiding this comment

acdlite Jan 13, 2017

Choose a reason for hiding this comment

acdlite left a comment

Choose a reason for hiding this comment

bvaughn commented Dec 27, 2016 •

edited

Loading

bvaughn commented Dec 27, 2016 •

edited

Loading

bvaughn commented Dec 27, 2016 •

edited

Loading

acdlite commented Dec 29, 2016 •

edited

Loading