pub struct CheckOptions {Show 23 fields
pub inputs: Option<Arc<dyn InputSource>>,
pub seen_disk: Option<SeenOptions>,
pub frontier_memory_bytes: Option<usize>,
pub frontier_directory: Option<PathBuf>,
pub candidate_memory_bytes: Option<usize>,
pub candidate_directory: Option<PathBuf>,
pub store_compact_bytes: Option<usize>,
pub predecessor_memory_bytes: Option<usize>,
pub predecessor_directory: Option<PathBuf>,
pub profile: Profile,
pub symmetry_liveness: SymmetryLiveness,
pub eval: Options,
pub store: StoreMode,
pub fp_index: usize,
pub tlc_fingerprints: bool,
pub seed: i64,
pub no_deadlock: bool,
pub workers: usize,
pub sequential: bool,
pub chaos: Option<u64>,
pub budget: Budget,
pub max_states: Option<u64>,
pub max_depth: Option<u64>,
}Expand description
How to run a check.
Fields§
§inputs: Option<Arc<dyn InputSource>>Optional owned external inputs for source binding. Runtime reads and checkpoint/recovery are not implemented; evaluator grants stay explicit.
seen_disk: Option<SeenOptions>Opt-in transient fallible disk seen-state retention.
None keeps the seen set resident in memory. Disk runs use an explicitly
reported inline fallback; this is not a total-memory or recovery budget.
frontier_memory_bytes: Option<usize>Opt-in encoded resident budget per current/next frontier queue. TLC uses
its disk queue; None keeps the in-memory equivalence reference. This
bounds frontier bytes only, not seen sets, candidates, traces or graphs.
frontier_directory: Option<PathBuf>Parent directory for private transient frontier files; system temp by default. Files are explicitly cleaned and failures reported.
candidate_memory_bytes: Option<usize>Encoded candidate bytes per worker under explicit Core safety BFS. None is the resident TLC-equivalent reference; zero immediately spills. Active evaluator roots, codec scratch and accepted Store are outside it.
candidate_directory: Option<PathBuf>Parent for private transient candidate files; requires a candidate budget.
store_compact_bytes: Option<usize>Completed-level measured Store-byte trigger for explicit pure Core safety. Zero forces each boundary; None is TLC-equivalent. This is not a hard cap.
predecessor_memory_bytes: Option<usize>Separate retained predecessor capacity budget. None preserves the
in-memory TLC counterpart; zero immediately spills native records.
Seen sets, frontiers, candidates, graphs and reconstructed paths are
outside this limit. Only ordinary safety BFS is admitted.
predecessor_directory: Option<PathBuf>Parent of private transient predecessor directories. Requires a budget; omitted uses the system temporary directory. Not a checkpoint location.
profile: ProfileThe profile: dive’s default or --compat=tlc.
symmetry_liveness: SymmetryLivenessWhole-run symmetry policy for genuine temporal properties. TLC checks a
symmetry-reduced graph, which can miss or invent liveness violations.
TLC-equivalent: Tlc; unsupported quotient checking still refuses.
eval: OptionsEvaluation options (Options::tlc() under --compat=tlc).
store: StoreModeFingerprint mode (the default, as in TLC) or exact mode
(--store exact, full states with an unconditional exhaustiveness
claim). TLC setting: fingerprint.
fp_index: usizeTLC’s fingerprint polynomial index (-fp), 0..131. TLC setting:
random per run unless given; the default profile fixes it at 0.
tlc_fingerprints: boolWhether to key the fingerprint set on TLC’s own state fingerprints in
TLC’s variable order (StateFingerprinter::tlc), or on dive’s cheaper
two-level fingerprints (StateFingerprinter::new: cached FP64 value
fingerprints combined by a non-linear, position-keyed mixer chain).
TLC setting: true, which --compat=tlc uses so that
the fingerprint set and the “actual” collision figure are TLC’s; the
default profile uses dive’s. Result-preserving: only a collision could
tell them apart.
seed: i64The run’s seed (always reported, so --seed replays the run).
no_deadlock: boolTLC’s -deadlock flag: do not check deadlock, whatever the
configuration says.
workers: usizeWorker threads (--workers). Results are identical at every count.
TLC setting: 1, TLC’s default; a spec that depends on
TLC’s sequential side effects runs with one logical worker whatever
this says (sequential mode), and the result says so.
Final/short graph capture expands at most
crate::capture::MAX_EXPANSION_UNITS source units concurrently even
with a wider pool; its scratch admission bound is reported separately.
sequential: boolRun every level inline on one thread, as sequential mode does, even
when the specification does not need it. For tests and diagnosis
(--sequential); results are identical either way.
chaos: Option<u64>Test-only chaos (--chaos SEED, hidden): randomize block sizes, the
order in which workers claim blocks and shards, the split of narrow
levels, whether small phases run inline, and insert random yields.
Results must not change; any change is a determinism bug.
budget: BudgetThe run’s time and memory limits. A run arms an unarmed
budget for itself; a caller that runs several stages
(capture, then temporal analysis) arms it once with
crate::budget::Budget::start so the stages share one clock.
Unlimited here; dive_api resolves the profile’s defaults. TLC
setting: no time limit, memory bounded by its heap.
max_states: Option<u64>Maximum admitted distinct in-model identities; None is TLC-equivalent.
Duplicate states do not consume this budget.
max_depth: Option<u64>Maximum breadth-first discovery depth; initial states have depth one.
None is TLC-equivalent; TLC’s simulation -depth is unrelated.