Fix SHA3/Keccak large-buffer Absorb OutOfRange (closes #94) - #100
Conversation
Widen TBABytes beyond 64 KiB so {$R+} no longer raises in Absorb for
multi-megabyte messages (HashBenchmark path). HashBenchmark now uses the
correct CalcBuffer(FBenchmarkBuffer[0], Size) form without a spurious @.
Add Keccak-224 1 MiB CalcBuffer regression with known digest.
THash_SHA3Base.Calc used PBABytes for DataPtr; Inc(DataPtr, RoundSize) scaled by SizeOf(TBABytes). Walk with PByte and cast only for Absorb.
Use 64 KiB Absorb rounds so large messages exercise multi-round DataPtr advance. Extend TestCalcBufferLargeMessage to compare one-shot multi-round Calc against incremental small-chunk Init/Calc/Done plus known digest.
d04bfaf to
c57347d
Compare
|
Hallo Markus, kurze Klarstellung: die geänderte Zur Datei selbst: Das ist keine offizielle NIST-CAVS-Datei, sondern eine große DEC-eigene Fixture im CAVS- Beim Fix in #99 stimmte der Ciphertext bereits mit AES-GCM überein, der erwartete Tag in der
habe ich unabhängig mit PyCryptodome (AES-GCM) über denselben Key/IV/PT/AAD nachgerechnet und die Fixture daran angepasst. Der alte Tag Wenn du willst, poste ich dir gerne noch das kleine Python-Snippet dazu. Viele Grüße |
2 participants
For Markus — theme: product bug fix (HashBenchmark / large messages, GitHub #94)
Addresses issue #94 (HashBenchmark OutOfRange in Keccak Absorb) with two root causes — both fixed here.
Cause 1 — Demo API misuse (intermittent)
CalcBuffer(@FBenchmarkBuffer[0], Size)passes a pointer-to-pointer becauseconst Bufferis already by reference. Failures depended on stack layout — hard to reproduce, easy to mis-blame on the library.Fix:
CalcBuffer(FBenchmarkBuffer[0], BufferSize)+ comment; drop thePwdHashBufworkaround so password and non-password hashes share one path.Cause 2 — Library: 64 KiB typed view + multi-round pointer walk
TBABytes = array[0..65535]with{+}in the SHA-3 section →data^[i]past 64 KiB raisedERangeErroron large messages (e.g. 1 MiB).THash_SHA3Base.Calcadvanced aPBABytespointer withInc(..., RoundSize), which scales bySizeOf(TBABytes), not by bytes — multi-round absorption for large inputs walked the wrong addresses.Fix:
TBABytesbound (typed view only; no allocation change)PByte; cast only atAbsorbRegression tests
TestCalcBufferLargeMessage(Keccak-224, 1 MiB, HashBenchmark fill pattern):CalcBuffer/CalcBytesCalcvs incremental small-chunk Init/Calc/DoneFiles
Source/DECHash.pasDemos/HashBenchmark_FMX/MainFormHashBenchmark.pasCalcBufferusageUnit Tests/Tests/TestDECHashSHA3.pasScope boundary
Risk
Medium — core SHA-3 sponge path for all SHA-3/Keccak digests. Mitigated by explicit 1 MiB multi-round regression and existing KATs.
Independence
Independent of DUnitX and GCM. Complements the Keccak test PR; this one is the library large-buffer fix.
Test plan
TestCalcBufferLargeMessagegreenCloses #94 (when merged).