Optimise (dom (delegs ▷ Set.singleton hk) ◁ stake) #1857
Conversation
(dom (delegs ▷ Set.singleton hk) ◁ stake), motivated by intersectDomPLeft which efficiently computes ((dom rewards' ◁ delegs) ▷ dom poolParams). Altered the Set Algebra operators so that every singleton set ends up in Exp as (SetSingleton x), rather than (Base SetR (Sett _)) or (Base SingleR (SetSingle _)), this means we ony have a single pattern to match. See the smart constructors dRestrict, rRestrict etc. Also added some property tests that test that intersectDomPLeft and intersectDomP meet their specification.
nc6
force-pushed
the
ts/optimiseRewards
branch
from
a891765
to
aa52e8e
Compare
TimSheard
force-pushed
the
ts/optimiseRewards
branch
from
7b34797
to
c656b0e
Compare
…compute.
compute (DRestrict (Dom (RRestrict (Base MapR delegs) (SetSingleton hk))) (Base MapR stake)) = ..
line 1022 of SetAlgebraInternals.hs There are three ways to compute this case
1) run (compile (dom (delegs ▷ Set.singleton hk) ◁ stake)))
2) (intersectDomPLeft (\ _k v2 -> (v2==hk)) stake delegs))
3) materialize MapR (do { (x,y,z) <- delegs 'domEq' stake; when ((y==hk)); one(x,z) }))
Only the first case was correct. When we made the function rRestrict be RRestrict, the
(SetSingleton hk) pattern failed and we fell through to the first case, which is the default case.
Once we fixed the code so case 2 and 3 was correct, everything ran fine, even when we reverted
rRestrict to its optimizing version. We also added some property tests checking that all 3 cases
were the same.
uroboros
force-pushed
the
ts/optimiseRewards
branch
from
98e7994
to
a9c8a44
Compare
|
After @TimSheard 's bug fix on this PR - we were still getting "Insufficient Utxo Balance" errors in the Tx generator - this has been occasionally happening since I increased the trace lengths for "preservation of Ada" properties. Longer traces increase the chance of the Tx generators eventually "running out of utxo". To address this I've increased the |
@TimSheard actually has a fix for the "Insufficient Utxo Balance" as a part of #1815. The fix is a better coin selection algorithm in the trace generation. I think we should separate this fix as a standalone PR. |
Split apart from #1851 as the tests seem to be failing.