Various improvements around previousResult/newData change detection. #4032
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
May help with apollographql/react-apollo#2126, since this code was only making a shallow copy of defaultQueryInfo, so there's a risk that defaultQueryInfo.listeners and .subscriptions could have grown over time, without ever getting garbage collected.
Should help fix #3992. As #3992 (and specifically the reproduction that @OurMajesty provided at https://codesandbox.io/s/q7rkm6y1ow) demonstrates, since the InMemoryCache may return the same object for multiple reads (when the data have not changed), the previousResult is often exactly (===) the same object as newData.result, which was causing the code removed by this commit to return early instead of broadcasting the watch. This early return was unsafe, because the contents of the object may have changed since they were previously broadcast, so we actually do need to report those modifications, even though the object reference is the same. In other words, `previousResult === newData.result` does not imply "nothing has changed," as I mistakenly assumed in this discussion: #3394 (comment) In the longer term, I would like to eliminate the previousResult logic entirely, since it seems redundant now that we can precisely track changes to the store, but for now it's important for backwards compatibility. Specifically, previousResult still provides a way to prefer the previous object if it is structurally identical to the new object, though that preference is moot when `previousResult === newData.result`. The previousResult object should never be used as a basis for skipping broadcasts. Only the application developer can decide whether === object identity means it's safe to skip rendering, likely in the context of a larger design which treats cache results as immutable data. The job of the InMemoryCache is to make no unsound assumptions, and broadcast results whenever they may have changed.
If you go to the trouble of detecting cycles, you should clone them accurately rather than ignoring them!
Should help with #3992. Though deep cloning and isEqual testing are obviously more expensive than relying on === equality, the reality is that lastResult may have been modified since it was previously recorded, so === equality is useless for determining isDifferentResult. A defensive copy is the only way to make sure we get this right.
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #3992, along with a few other miscellaneous improvements.
As #3992 (and specifically the reproduction that @OurMajesty provided) demonstrates, now that the
InMemoryCachemay return the same object for multiple reads (when the data have not changed), thepreviousResultis often exactly (===) the same object asnewData.result, which was causing the code removed by this commit to return early instead of broadcasting the watch.This early return was unsafe, because the contents of the object may have changed since they were previously broadcast, so we actually do need to report those modifications, even though the object reference is the same.
In other words,
previousResult === newData.resultdoes not imply "nothing has changed," as I mistakenly assumed in this discussion: #3394 (comment)In the longer term, I would like to eliminate the
previousResultlogic entirely, since it seems redundant now that we can precisely track changes to the store, but for now it's important for backwards compatibility. Specifically,previousResultstill provides a way to prefer the previous object if it is structurally identical to the new object, though of course that preference is moot whenpreviousResult === newData.result.The
previousResultobject should never be used as a basis for skipping broadcasts. Only the application developer can decide whether===object identity means it's safe to skip rendering, likely in the context of a larger design which treats cache results as immutable data. The job of theInMemoryCacheis to make no unsound assumptions, and to broadcast results whenever they may have changed.