Regression PR 8826 broke cache.reset()

[tm1000]

Intended outcome:
Before #8826 was introduced we used cache.reset() to not only reset the cache but also recall all active query watchers. (much like client.resetStore()).

However, I noticed that #8826 (comment) seems to break this functionality

This seems to be a regression (Or were we using this wrong) as now the store is reset but no active queries are called again.

Actual outcome:
I'd Expect the active query watchers to be called as it worked before

How to reproduce the issue:
Call cache.reset() before and after #8826

[tm1000]

@benjamn if we were using cache.reset() wrong before let me know as we can just use client.resetStore() instead (it just seems strange to call the global client inside fo the client itself). Its just we were using cache.reset() inside of the update callback in some cases instead of having to manually rebuild caches that changes multiple fields.

[benjamn]

Thanks for the quick feedback @tm1000!

I see now how calling this.broadcastWatches() after this.init() was a reliable way for cache.reset to rewatch all watchers (that is, before #8826), so I agree this was a breaking change.

In 3.4.15, we will restore the pre-3.4.14 behavior by default, while supporting

cache.reset({ discardWatches: true })

as a way to opt into the behavior of #8826.

Does that meet your needs/expectations @tm1000? Happy to reopen if not!

[tm1000]

@benjamn Thanks for the quick response. I think it fixes the issues but my engineers have told me that cache.reset() actually broke between 3.3.21 and 3.4.0 which was rather unexpected (they first told me it was related to this and then they ran through downgrading until they got to 3.3.21 which they said worked)

Looking at the code though it looks like the watchers should refetch even in 3.4.0 and especially after your work in #8826 so the investigation continues but without the work in #8826 it definitely wouldn't have worked at all.

Either way this specific issue is resolved. If I find something else I'll open a new issue in regards to that but again #8826 was still needed because it brings back watchers being able to refetch at least

[benjamn]

Alas, 3.3.21 was the last version before 3.4.0, which was a relatively big minor release. Of all the changes in 3.4.0, 526c940 (something I added to #8107, perhaps unwisely) looks the most to me.

Please do open a new issue if you find anything more. Now that we have the Cache.ResetOptions interface, we can add new options to cache.reset relatively easily, to accommodate more use cases simultaneously.

[tm1000]

@benjamn Thanks for pointing me in the right direction!