#dat

a peer cannot replicate and say a feed is zero length in a trusted way

right

so might as well just say progress == 0

we might consider putting this progress method on hyperdrive

for now, just know I think I've got a correct method in archiver-api

sgtm

mafintosh: strong indications that RAF closed-files is the main bug, and bad progress reporting was a secondary issue

looking for a permanent fix

need to isolate when the RAF *should* be closed vs when it is now

mafintosh: hey this looks off https://github.com/mafintosh/hypercore-archiver...

is it possible that all those references to opened[..] in the done() method ought to be to opened[..] cnt?

because opened[..] references an object, not a number, with {feed:obj, cnt:number}

ahhhh

yes

ok that may be our winner. testing

nice

mafintosh: https://github.com/mafintosh/hypercore-archiver...

pfrazee: are you an owner on that?

mafintosh: yeah I think so

pfrazee: cool, merge and release it :)

mafintosh: will do. Sadly Im now finding that if I run the tests without debug logging on, I still get failures, which makes me think there's a timing issue that's affected by the logging

well that seems like a great fix no matter what

yeah definitely

Im not seeing signs of that bug, or the RAF issue, now - so we may have a THIRD bug at play. Dun dun dunnn

mafintosh: Im looking forward to the day we get to reimplement the dat stack in rust, and then compile it into wasm and/or native

yea

agreed

pfrazee: i'm gonna redeploy the archiver with your fix

mafintosh: ok. Turns out there's still cases where the RAF is closed when the write is attempted, so I should have another patch on the way

pfrazee: thats fixes the remaining bug?

mafintosh: it's the same failing test - sometimes sync doesnt complete

mafintosh: my patch fixed one case that caused that, but there still appears to be another cause

mafintosh: is it possible that RAF instances are shared between the meta and content hypercores of an archive?

Im trying to wrap my head around why the RAF instance is getting closed prematurely

ok nm I'm onto something

yeah ok Ive got a lead

mafintosh: the opened[] entries are never cleared

waaait guh nm