PLONK's verification equation was never written: verifyPLONKPairing
refused every proof unconditionally, and above it sat some 200 lines
decoding the proof and the key — exact sizes, subgroup checks, points at
infinity — whose result that refusal discarded. The refusal is the whole
of what the path does, so it is now the whole of what it says.
The aggregate block proof went the same way. VerifyBlockProof re-verified
the very transactions Verify had verified one line above, so it could
only ever agree with them; nothing built a block carrying one; and the
field stood on the wire and in the block id where a reader would take an
aggregate guarantee to be.
Two more second implementations: Verify type-asserted its parent to
*Block after GetBlock had already refused anything that is not one, and
Shutdown asked whether its logger was disabled before logging, which is
what a logger decides.
Coverage 95.5% -> 99.1%. What is left cannot be reached: four bn254.Pair
error returns on one-element slices, two IsInSubGroup refusals on a curve
whose G1 cofactor is 1 and whose Unmarshal already checks G2, and three
boot gates — a store opened through a versiondb view, and fee.Validate on
the floor chain.Floor returns.
Coverage 76.7% -> 95.5%, and three more things the tests found on the way.
A chain that seeds its genesis allocation and has not yet accepted a
block reports itself fresh on every boot, because the store records a tip
when it accepts a block and seeding is not accepting one. Answering that
with a second allocation ends in "UTXO already exists" — a node that
cannot restart until it produces a block. Whether the allocation has been
applied is now asked of the state: Finalize is the last thing it does, so
a committed root that is no longer the empty one is the allocation saying
so.
Three Marshal methods returned an error they never produced, so every
caller wrote an unreachable branch for it — Block.Bytes answered nil in
its one, which is a block with no encoding.
spendTx built every fixture with the same all-zero commitment, so two
different spends created the same output and the set refused the second.
An output commitment now follows what the transaction spends.
The proof tests construct a key and a proof that SATISFY the pairing
equation, so the arithmetic reaches a verified verdict rather than only
refusals — and then show that the same proof bytes on a transaction
spending different notes do not reach it.
zeekay