v0.9.2

Added

• rvm.fork resolves Flare-family network hardforks (Durango → Shanghai, Etna
  → Cancun, pre-Durango → London) from go-flare's upgrade schedule instead of
  defaulting to the newest spec
• --max-failures N to collect up to N distinct failed assertions (invariant
  mode only) before stopping the campaign, with each one shrunk and reported
  separately
• max_* harness functions: read-only, no-argument uint256 getters where
  reverted or empty results score 0 and any value above 0 is the finding
• Trace decoding falls back to common standard events (ERC20
  Transfer/Approval, ERC721 ApprovalForAll, WETH9 Deposit/Withdrawal,
  Ownable OwnershipTransferred) when no project artifact declares them,
  rendering names and arguments instead of raw emit Log(0x...) lines

Changed

• The fuzzing lifecycle is one fuzz tracing span carrying the harness
  contract and thread count, with consistent started, progress, and
  finished events; the campaign log file records the fuzz-phase duration when
  the span closes
• Per-function statistics log kind function as the message (e.g.
  handler deposit calls=60.2K gas=11.14 G reverts=0) instead of a generic
  Function statistics message with a function field
• Maxxing campaigns that find no improvement log a warn naming the objective
  (e.g.
  objective=max_profit No sequence improved the max value (best stayed at 0))
  instead of an error
• Removed the redundant Called setup log line and the Ripfuzz out. see ya
  farewell line
• Fuzzing progress lines now log structured key=value fields (matching the
  final campaign summary) instead of ·-separated prose
• Terminal log lines print a simple local HH:MM:SS timestamp without the
  module target; the campaign log file keeps the full RFC 3339 timestamp with
  target
• Trace output hangs children, call context, logs, storage, and result lines
  directly under each frame's name (aligned regardless of gas amount),
  replacing the --- Call #N --- header with a [N] counter on the root frame
  line
• --fail-on-revert is replaced by --stop-on-revert: any reverted
  transaction stops the campaign (invariant and maxxing mode), writes the full
  trace to fulltrace.log, dumps a compact trace (call context and storage
  changes omitted) to the log and stderr, names both file paths in the error,
  and exits with a failure instead of shrinking
• A failed setup() after a successful deployment stops the campaign like
  --stop-on-revert: full trace to fulltrace.log, compact trace to log and
  stderr, both paths named in the error
• A failed harness deployment stops the campaign like --stop-on-revert: full
  trace to fulltrace.log, compact trace to log and stderr, both paths named
  in the error
• Failed-assertion and max-value findings now dump their traces like
  --stop-on-revert without failing the campaign: full trace to
  fulltrace.log (per-finding fulltrace-N.log or fulltrace-max-N.log),
  compact trace to log and stderr, both paths named
• Maxxing campaigns no longer track failed assertions or enter the shrinker on
  a revert
• Upgraded solc dependency to v0.1.1
• Fuzzer and shrinker progress logs one compact line every 3 seconds, with the
  full statistics printed after the phase finishes
• Terminal status output now goes through tracing; --disable-log disables
  all log output (terminal and campaign log file)
• Campaign mode is selected automatically: a harness with a max_* function
  runs in max mode, which supports exactly one max function and rejects
  invariant_* functions
• Renamed the maxxing campaign type from MaxCampaign to MaxxingCampaign.
• Fuzzer types now live under fuzzers.
• Shrinker types now live under shrinkers: Shrinker is renamed to
  InvariantShrinker and MaxShrinker to MaxxingShrinker; the max module
  was removed.
• Campaign dispatch moved into commands::run::run; CampaignKind::Max is
  renamed to CampaignKind::Maxxing.

Fixed

• Fork RPC batches that mix cached and missing keys no longer kill the fetcher
  thread with fetcher did not receive all keys, which stalled campaigns after
  new storage slots appeared
• Campaign worker failures are no longer swallowed: any failed or panicked
  fuzzer/shrinker thread exits the campaign after all workers settle, with the
  full cause chain (e.g.
  revm transaction failed: database error: RPC rate limited: …) instead of
  only the outer message
• Skipped build artifacts now warn with the artifact file path and full error
  chain instead of a bare cause message printed twice
• Build artifacts are loaded once per campaign, so trace contexts reuse them
  instead of re-reading the build output directory (which duplicated artifact
  parse errors in the log)
• --stop-on-revert traces stop at the first reverted transaction: only the
  calls up to and including it are re-run and dumped
• Mid-transaction rvm.fork switches no longer drop or leak remote state
  written earlier in the same transaction (e.g. rvm.store/rvm.deal on fork
  A then rvm.fork to B): journaled remote mutations commit to the active fork
  overlay before the switch, and local harness accounts stay shared across
  forks
• Fork transport JSON-RPC payloads are logged at debug instead of info, so
  default runs no longer flood the terminal with full payload lines