Fully eliminate the concept of heuristic fragment matching. #5684
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
Although we removed the FragmentMatcher abstraction in PR #5073, and the HeuristicFragmentMatcher no longer exists, we kept trying to match fragments whenever all their fields matched, even if the fragment type condition was not satisfied by the __typename of the object in question. I recently came to appreciate just how harmful this best-effort matching can be, when I realized that it means we call executeSelectionSet (on the reading side) and processSelectionSet (on the writing side) for every unmatched fragment, and then just throw away the result if any fields are missing, but not before recording those fields as missing in a weaker way than usual, using the ExecResultMissingField.tolerable boolean. Not only is this strategy potentially costly for queries with lots of fragments that really should not match, it also only ever made sense on the writing side, where the fields of the fragment can be compared to the fields of an incoming result object, so the presence of all the fragment's fields is a pretty good indication that the fragment matched. On the reading side, we're comparing the fields of the fragment to all the fields we've written into the cache so far, which means heuristic fragment matching is sensitive to the whole history of cache writes for the entity in question, not just the current query. We could eliminate heuristic fragment matching just for reading, but then we'd have to tell people to pass possibleTypes to the InMemoryCache constructor (the correct, non-heuristic solution), which would make heuristic fragment matching unnecessary on the writing side, too, so we might as well completely eliminate heuristic matching altogether.
When we're writing to an entity ID that exists in the store, and the normalized entity object has a __typename field (as almost all of them should), we can use that __typename as a fallback when the result object that we're writing to the store is missing a __typename field.
Errors due to missing fields from heuristically-matched fragments were previously suppressed because there was a chance the heuristic was wrong. Since we're no longer doing any unsound heuristic fragment matching, there's no longer such a thing as a tolerable missing field. This change should make fixing issue #5614 unnecessary.
|
GitHub appears to be having some serious problems updating in response to |
hwillson
approved these changes
Dec 13, 2019
| @@ -377,10 +377,18 @@ export class Policies { | |||
| public fragmentMatches( | |||
| fragment: InlineFragmentNode | FragmentDefinitionNode, | |||
| typename: string, | |||
| ): boolean | "heuristic" { | |||
| ): boolean { | |||
There was a problem hiding this comment.
Super happy to see this just return a boolean.
There was a problem hiding this comment.
Words like "truthy" and "heuristic" belong in the same circle of word-hell.
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.
Although we removed the
FragmentMatcherabstraction in PR #5073, and theHeuristicFragmentMatcherno longer exists, we kept trying to match fragments whenever all their fields matched, even if the fragment type condition was not satisfied by the__typenameof the object in question.I recently came to appreciate just how harmful this best-effort matching can be, when I realized that it means we call
executeSelectionSet(on the reading side) andprocessSelectionSet(on the writing side) for every unmatched fragment, and then just throw away the result if any fields are missing, but not before recording those fields as missing in a weaker way than usual, using theExecResultMissingField.tolerableboolean.Not only is this strategy potentially costly for queries with lots of fragments that really should not match, it also only ever made sense on the writing side, where the fields of the fragment can be compared to the fields of a single incoming result object, so the presence of all the fragment's fields is a pretty good indication that the fragment matched. On the reading side, we're comparing the fields of the fragment to all the fields we've written into the cache so far, which means heuristic fragment matching is sensitive to the whole history of cache writes for the entity in question, not just the current query.
We could eliminate heuristic fragment matching just for reading, but then we'd have to tell people to start passing
possibleTypesto theInMemoryCacheconstructor (the correct, non-heuristic solution), which would simultaneously make heuristic fragment matching unnecessary on the writing side, so we might as well completely eliminate heuristic matching altogether.