Attempt to fix realtime links

[clehner]

No description provided.

[dominictarr]

what does this fix?

_value is used so that live streams can emit immediately without a lookup.
https://github.com/ssbc/secure-scuttlebutt/blob/dc1c1f8a2000ba622881cb68de049541529eebe0/index.js#L479-L480

Which prevents a race condition in other places... though, I think that was before pull-live which is a better approach in general, so maybe that is now obsolete?

[clehner]

It is to fix NotFoundError being emitted from ssb.links, discussed in %rBu56IqW1RopjM7yOQUI3ckPjVC36b14ax8rLei2jVM=.sha256

_value isn't actually being used until this line is added:
dc1c1f8#diff-168726dbe96b3ce427e7fedce31bb0bcR457

I see db.links uses pull-live but I don't see how that should fix the race. I think what is happening is that before a message is added to the db, the pre hook function (L74) runs, passes a link index op synchronously through realtime(kv) (L82) to the pipeline in db.links (L499), to lookupLinks which calls db.get(op.key, ...) (L481). Trying to get the new message fails because the pre hook function hasn't returned yet and so the message hasn't yet been added.

[dominictarr]

now i remember. at first, i implemented real time views by just creating a live stream of the index,
then mapping that through the lookup. but this meant that realtime view was always delayed (because it would do a get.

but the new code just does the realtime part by filtering db.post(..) so this should now be unnecessary.

merged into 15.2.4

[clehner]

I just realized that the pull request message that I wrote for this wasn't published, because I used a wrong command-line flag (-m msg instead of -F msg) to the hub command. Sorry about that. Anyway, if the errors I was seeing persist, I will revisit this and maybe try to write a test case.

[dominictarr]

ah got it