Quick lazy-Google question: let's say `FooQuery` fetches list of items `[a, b, c]` (with `ID`s norma...

# apollo-kotlin

[wasyl]

Quick lazy-Google question: let's say FooQuery fetches list of items [a, b, c] (with `ID`s normalized, so the items have their own entries in the cache). If I manually remove an item from the cache (using apolloClient.apolloStore.remove(CacheKey(itemId), cascade = false)), should I expect FooQuery to
a) trigger refetcher
b) emit null while b is missing in the cache and re-emit non-null value only when b is fetched, either by refetcher or some other query
?

[mbonnin]

remove() doesn't publish by default so you'd at least have to call publish(setOf(a, b, c))

[mbonnin]

If you call publish() , the refetcher will be called and try to fetch according to refetchPolicy

[mbonnin]

Whether it emits null or throws depends a bunch of factors like the refetchThrows parameter you've passed and/or whether you're on 3.x or 4.x

[wasyl]

ah perfect, yeah the emission is less of an issue, I was surprised about the publishing

[wasyl]

Though I'm not sure how I can publish what I want 🤔 In the method to remove an item I don't know which queries depend on that item. Does

apolloClient.apolloStore.remove(CacheKey(contentId), cascade = false)
apolloClient.apolloStore.publish(setOf(contentId))

make sense or the publish wouldn't do anything in such case?

[mbonnin]

As long as cascade is false that should be alright. But there's definitely a blind spot if cascade is true

[wasyl]

Right, because we'd need to publish all cascaded deletions too 🧠 Perfect, that clears things up, thank you 🙂

[mbonnin]

Thanks for raising this.

[mbonnin]

Filed this for tracking: https://github.com/apollographql/apollo-kotlin/issues/4974

[wasyl]

🙇 thanks, just for clarity this is not an issue for us (we use cascade=false), and frankly I'd love to get rid of manipulating cache manually in the first place, only for reasons we can't do that. But the issue looks valid anyway, as the current behavior is just surprising because observing cache for the same query twice can yield different results (if you do observe — delete item — observe)

[wasyl]

🧟 I just stubled upon this again, this time with cascade = false where I think I also can't reliably publish cache changes 🤔 Specifically when I have a query with nested objects and I remove one of the inner objects, it's not (I think) part of the cache key for the encapsulating query, so I can't publish that query's changes, right?

For example query Items would return foo1 = Foo(items = [ bar1 ]) where bar1 = Bar(items = [ baz1 ]. If I remove baz1, then neither bar1 or foo1 can be satisfied, but the cache key for query Items doesn't have baz1 anywhere, so I can't publish the deletion properly, is that right? Let me know if the example isn't clear