…ndaries

Promotes TestResult.other_failures from Vec<String> to Vec<InvariantOtherFailure>
carrying name, reason, optional counterexample, and persisted path. Display
renders each secondary symmetrically with [FAIL: reason] + [Sequence] block
when a counterexample is available, falling back to the terse 'name: reason'
one-liner otherwise.

Amp-Thread-ID: https://ampcode.com/threads/T-019dcd68-66ac-76ed-ac5c-7ea722a9c9ae
Co-authored-by: Amp <amp@ampcode.com>

…-up, opt-mode warning

Three small UX wins for assert_all campaigns. No behavior change, no
new dependencies.

1. Shrink progress bar gets an [i/N] queue counter when more than one
   invariant needs shrinking, so users see how many shrinkers are
   queued behind the current one (e.g. '[2/3] Shrink: invariant_X').
   reset_shrink_progress and replay_error gain a position parameter;
   single-invariant call sites pass None.

2. Suite-level roll-up footer: when assert_all exercised >1 invariant
   and the test failed, render 'Suite assert_all: <broken>/<total>
   invariants broken' above the per-invariant blocks. Gives CI logs
   and Slack pastes a glanceable health line. New Option<usize>
   field on TestResult, populated only when meaningful.

3. Startup warning when assert_all + optimization-mode are combined.
   Optimization mode tracks one int256 return value, so any boolean
   secondary invariants in the same contract are filtered out before
   the campaign — previously silent. Now emits a once-per-suite warning
   naming the optimization invariant and every dropped boolean so users
   can move them to a separate contract.

E2E tests: extend assert_all to assert the new 4/5 roll-up;
assert_all_only_primary covers the 1/2 case; new
assert_all_optimization_mode_warning verifies the warning fires with
the dropped invariant names.

Co-authored-by: Amp <amp@ampcode.com>
Amp-Thread-ID: https://ampcode.com/threads/T-019dcdd3-53f5-76b6-ac36-d59f06b58280

… failures

Symmetric with the primary's existing persisted-replay warning. Echidna
and Medusa never silently drop properties between runs — properties are
re-evaluated every campaign and a previous failure doesn't suppress
them. Foundry's per-property failure file model meant secondaries with
a stale persisted counterexample were filtered out of the campaign with
no acknowledgment, so users coming from Echidna/Medusa would see fewer
invariants in the report than their contract defines.

Now emits one stderr line listing every skipped name and the cache dir
to clean, e.g.:

  Warning: test/X.t.sol:Suite: 3 invariant(s) skipped due to persisted
  failures: invariant_a, invariant_b, invariant_c. Run `forge clean`
  or delete files in cache/invariant/failures/Suite to re-include.

E2E: extends assert_all re-run case with stderr_eq snapshot asserting
the warning fires with all 3 skipped names.

Co-authored-by: Amp <amp@ampcode.com>
Amp-Thread-ID: https://ampcode.com/threads/T-019dcdd3-53f5-76b6-ac36-d59f06b58280

Previously afterInvariant was gated on failures.errors.is_empty()
campaign-wide. Under assert_all that gate stayed closed for the rest
of the campaign once any invariant broke, silently skipping the
afterInvariant hook on every subsequent run. Any assertions or
cleanup logic in afterInvariant therefore stopped running after the
first unrelated invariant failure.

Now snapshot failures.errors.len() at the start of each run and only
skip afterInvariant when the current run produced a new failure.
Preserves the legacy 'don't run afterInvariant on a run that already
failed' semantics while letting it run on subsequent runs once an
earlier invariant has broken.

E2E: new assert_all_after_invariant_runs_after_earlier_failure case
breaks invariant_first in run 1, keeps the campaign alive with a
second never-breaking invariant, and asserts an always-reverting
afterInvariant surfaces its marker in failure output.

Co-authored-by: Amp <amp@ampcode.com>
Amp-Thread-ID: https://ampcode.com/threads/T-019dce2d-57c7-734a-bbc6-6fa5e34b25de

…assert_all

When the selected invariant test passes but a secondary breaks under
assert_all, the report previously rendered a hollow '[FAIL]' header
(no reason, no counterexample) for the primary and the suite roll-up
overcounted broken invariants as '1 + other_failures.len()',
attributing a non-existent primary failure.

Now key the primary header on whether the primary actually broke
(`reason.is_some() || counterexample.is_some()`) and skip the
header when it didn't. Roll-up uses the same flag so the count
reflects only invariants that actually broke (e.g., 1/2 instead of
2/2). JSON shape is unchanged: top-level reason/counterexample stay
null when the selected primary didn't break, with full secondary
detail in other_failures.

E2E: new assert_all_secondary_only_failure_no_hollow_fail asserts a
secondary-only break renders 'Suite assert_all: 1/2 invariants broken'
followed by the secondary's '[FAIL: ...] <name>' block, with no
hollow primary header.

Co-authored-by: Amp <amp@ampcode.com>
Amp-Thread-ID: https://ampcode.com/threads/T-019dce2d-57c7-734a-bbc6-6fa5e34b25de

…ry-only case

Previous commit suppressed the '[FAIL]' header whenever the primary had
no top-level reason or counterexample, which also matches DS-style
failures (they signal via the 'failed' flag and log events rather than
through TestResult.reason). That regressed failure_assertions::ds_style_test_failing
and test_cmd::core::legacy_assertions in CI.

Now the suppression is scoped strictly to the assert_all secondary-only
case: skip the primary header only when no primary failure AND
assert_all is in play AND there is at least one secondary to render.
DS-style, plain unit and single-invariant failures keep the original
'[FAIL]'/'[FAIL: ...]' rendering.

Co-authored-by: Amp <amp@ampcode.com>
Amp-Thread-ID: https://ampcode.com/threads/T-019dce2d-57c7-734a-bbc6-6fa5e34b25de

…declaration order

The structured JSON 'failure' event emitted to stderr at campaign end
(consumed by benchmark and CI tooling) used 'errors.values().next()'
on a HashMap to pick its 'reason' field, while hardcoding the
'invariant' field to the primary's name. With HashMap RandomState,
the same broken set of invariants produced a different reason string
across runs, and the event was self-inconsistent (e.g.,
'invariant: invariant_balance, reason: fee miscalculation'). Three
sites used this pattern: in-run break path, afterInvariant break
path, and the preflight check fallback.

Now they walk 'invariant_contract.invariant_fns' in declaration
order (a Vec, deterministic) and pick the first one with a recorded
failure. Both 'invariant' and 'reason' fields refer to the same
function, and the event is stable across runs.

A new 'first_broken_event' helper centralizes the lookup.

E2E: assert_all_failure_event_uses_declaration_order declares three
invariants (a, b, c) that all break on the same call, runs with
'--mt invariant_c' (primary is the last declared) and asserts the
emitted event names invariant_a with reason 'a broken'.

Co-authored-by: Amp <amp@ampcode.com>
Amp-Thread-ID: https://ampcode.com/threads/T-019dce2d-57c7-734a-bbc6-6fa5e34b25de

…secondary_failures and serialize sparsely

Renames TestResult.other_failures -> invariant_secondary_failures and
the underlying InvariantOtherFailure struct -> InvariantSecondaryFailure.
The previous names were generic ('other relative to what?'); the new
names align with the existing 'primary/secondary' terminology used
throughout the assert_all rollout and follow the Rust Vec<Foo>/foos
plural-of-singular convention.

Also marks the field with #[serde(default, skip_serializing_if =
'Vec::is_empty')] so it is omitted from JSON output for any test that
has no secondary failure data — plain unit tests, fuzz tests, passing
tests. Pre-PR JSON consumers continue to see the same shape on those
results. invariant_failure_dir and assert_all_invariant_count already
had Option::is_none guards.

Updates the SimpleContractTest{NonVerbose,Verbose}.json fixtures to
drop the now-skipped empty field.

Co-authored-by: Amp <amp@ampcode.com>
Amp-Thread-ID: https://ampcode.com/threads/T-019dce2d-57c7-734a-bbc6-6fa5e34b25de