File size: 4,866 Bytes
1b950d4 ed2bd8a 1b950d4 ed2bd8a 1b950d4 ed2bd8a 1b950d4 ed2bd8a 1b950d4 ed2bd8a 1b950d4 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 | # Transcript rewrite + Multiplexer + Differential Validation
This replaces the earlier tagged-union `raw/transcript/transcript.go`
(nativeEvent/tracePayload + StreamProjection + deterministic hash) with a
flat, single-struct `RawEvent` design, per an explicit decision to trade
away the deterministic hash and immutable-projection guarantees for lower
per-event allocation cost. If you want the hash/projection version back,
it's recoverable from git history (commit `ed2bd8a`).
## What was extended beyond the pasted draft
The pasted `RawEvent`/`EventKind` only covered `TxStart/TxEnd/Enter/Exit/
Fault/BalanceChange`, with only `AppendEnter`/`AppendExit` actually
implemented. That's not enough to capture what EMES-V1
(`docs/emes_profile.md`) needs to reconstruct an SSR -- nonce, code, and
storage mutations, self-destruct, and account creation were missing
entirely. This version adds `KindNonceChange`, `KindCodeChange`,
`KindStorageChange`, `KindSelfDestruct`, `KindAccountCreated` and their
`Append*` methods, following the same flat-struct design.
One inconsistency kept rather than silently fixed: the original design's
stated goal was eliminating string allocations, but `RawEvent.ErrType` is
still a `string`. Noted in the source rather than either silently dropping
error type info or silently ignoring the stated goal.
## Bugs found and fixed (multiplexer/broadcast.go)
Checked against the real `core/tracing/hooks.go` (not assumed), three
signature mismatches:
1. `OnTxStart` was missing the `from common.Address` parameter
`TxStartHook` requires, and used a value `tracing.VMContext` instead of
the required pointer `*tracing.VMContext`.
2. `OnEnter`'s `value` parameter was typed `*uint256.Int`; `EnterHook`
requires `*big.Int`.
3. `OnFault` didn't match `FaultHook` at all -- the real signature is
`func(pc uint64, op byte, gas, cost uint64, scope OpContext, depth int,
err error)`, missing the `scope OpContext` parameter entirely and in a
different parameter order (pasted had `depth` first; the real hook has
it second-to-last).
Also: the pasted `BroadcastSink.WiringToGethHooks()` only implemented
`OnEnter` and left the rest as a comment ("mirror this exact wrapper
structure"). All five remaining hooks (`OnTxStart`, `OnTxEnd`, `OnExit`,
`OnFault`, `OnBalanceChange`) are now actually implemented, not left as a
placeholder, and the method was renamed `Hooks()` for consistency with
`tracer.Tracer.Hooks()` elsewhere in this repo.
## Other fixes (validation/normalizer.go, validation/engine.go)
- `normalizer.go` imported `kay-sentinel/raw/transcript` (hyphenated
module name); this repo's actual module is `kaysentinel` (see `go.mod`).
Fixed to `kaysentinel/raw/transcript`.
- `engine.go`'s comparison method was unexported
(`compareSemanticFields`), making it uncallable from outside package
`validation` -- presumably an oversight for something meant to be a
differential-testing API. Exported as `CompareSemanticFields`.
- Noted, not fixed: `SemanticEvent` only carries the fields shown
(Sequence/Opcode/Depth/From/To/Value/GasUsed/Reverted/ErrType), so a
differential check can't currently catch a mismatch in, say, a specific
storage slot's before/after value -- only that *some* event of a given
Kind/Sequence differs on the fields that do exist.
## Verification performed
```bash
go build ./raw/... ./validation/... ./multiplexer/...
go vet ./raw/... ./validation/... ./multiplexer/...
```
Both clean. Beyond that, an actual end-to-end run:
- Registered two `HookSink`s on a `BroadcastSink`: one that records into an
`ExecutionTranscript`, one that deliberately panics on `OnEnter`.
- Drove a fake transaction through `bus.Hooks()`. Confirmed: the panicking
sink's panic was caught, recorded as a `PanicReport` naming the right
sink and callback, and did **not** propagate (strictMode=false) --
and confirmed it was evicted: a second round of hook calls produced no
new panic and no new report, proving eviction actually removes the sink
rather than just logging and continuing to call it.
- The recording sink's transcript was fed through `EventNormalizer` and
`DifferentialValidator`: comparing a stream against itself matched on
every event, and a deliberately injected field mismatch (`Depth: 99`) was
correctly caught and reported with the expected tabular diff format.
## Status
Compiles, vets clean, and the panic-isolation + differential-validation
behavior was actually exercised, not just compiled. Not yet wired into
`tracer.Tracer` from the earlier EMES-V1 work -- this and that tracer are
currently two separate, non-integrated ways of collecting events from
Geth's `tracing.Hooks`. Deciding whether to merge them or keep both for
different purposes (e.g. this one for a legacy-vs-new differential period)
is an open decision, not resolved here.
|