--- Log opened Tue Jul 06 00:00:59 2021
00:14 < kcalvinalvin> Ok one more swapless win that we didn't realize: Leaf remember algo is much simpler
00:16 < kcalvinalvin> After a transform, we need to re-evaluate which leaves need to be remembered. A leaf might have been remembered as a proof. But if this leaf gets re-positioned, we need to re-evaluate if it still needs to be remembered as its sibling changed
00:17 < kcalvinalvin> So like if 14 & 15 were siblings and only 15 had the remember thing on, 14 would still be remembered as it's the proof for 15
00:17 < kcalvinalvin> but since 15 is remembered, it's gonna be deleted soon. After a transform, we need to re-evaluate if 14 still needs to be remembered
00:18 < kcalvinalvin> Like if the new 14's sibling also needs to be remembered. If it doesn't need to be, we can remove 14
00:18 < kcalvinalvin> But we're not doing this check at the moment
00:18 < kcalvinalvin> So random leaves stick around
00:18 < kcalvinalvin> I *think* this is why we have that memory issue. Well, maybe there might be more but this is def one of them
11:13 < adiabat> yeah in general things moving around is super annoying :)
11:14 < adiabat> with swapless, things don't *really* move around, they just move up.  And in some cases you can pretend that they didn't
23:19 < kcalvinalvin> https://github.com/mit-dci/utreexo/issues/287 I wrote a thing about this problem
--- Log closed Wed Jul 07 00:01:00 2021