maude: Phase 5 system modules + rewrite rules (21 tests, 159 total)
Some checks failed
Test, Build, and Deploy / test-build-deploy (push) Failing after 48s
Some checks failed
Test, Build, and Deploy / test-build-deploy (push) Failing after 48s
lib/maude/rewrite.sx: rl/crl transitions interleaved with eq normalisation. mau/rewrite = default strategy (top-down, leftmost-outermost, first rule); mau/rew bounded; mau/search = BFS reachability over all successors. lib/maude/fire.sx: short-circuiting matcher (mau/fire-eq) — finds the first productive match instead of enumerating the whole solution set. Fixes the exponential blowup of AC rewriting on many identical elements (8 coins: 60s+ to <1s). Eager match-multiset kept only for match-all / search. Verified on AC coin-change, traffic light, branching search, crl clock. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -86,10 +86,10 @@ The novel substrate stress: equational matching. Pattern `X + Y` against `1 + 2
|
||||
- [x] Tests: gcd, sorting, conditional simplifications.
|
||||
|
||||
### Phase 5 — System modules + rewrite rules
|
||||
- [ ] `mod ... endm` syntax with `rl` rules.
|
||||
- [ ] Rules apply asymmetrically (`=>` not `=`); fairness across rules.
|
||||
- [ ] Default strategy: top-down, leftmost-outermost, first applicable rule.
|
||||
- [ ] Tests: state-transition systems (puzzle solvers, protocol simulators).
|
||||
- [x] `mod ... endm` syntax with `rl` rules.
|
||||
- [x] Rules apply asymmetrically (`=>` not `=`); fairness across rules.
|
||||
- [x] Default strategy: top-down, leftmost-outermost, first applicable rule.
|
||||
- [x] Tests: state-transition systems (puzzle solvers, protocol simulators).
|
||||
|
||||
### Phase 6 — Strategy language
|
||||
- [ ] Compose strategies: sequential `;`, alternative `|`, iteration `*`, fixed-point.
|
||||
@@ -195,5 +195,30 @@ The novel substrate stress: equational matching. Pattern `X + Y` against `1 + 2
|
||||
right-parenthesized, or add a `gather`/parse-assoc attr later if a test
|
||||
needs bare `a : b : c`.
|
||||
|
||||
- **Phase 5 (system modules + rewrite rules) — DONE, 159/159 total.**
|
||||
`lib/maude/rewrite.sx` + `lib/maude/fire.sx`. Rules (rl/crl) reuse the
|
||||
equation firing machinery (a rule dict is shaped like an eq). `mau/rewrite`
|
||||
is the default strategy: normalise with eqs (`creduce`), fire ONE rule
|
||||
top-down/leftmost-outermost/first-applicable, renormalise, repeat (bounded
|
||||
by fuel). `mau/rew m src n` = bounded `rew [n]`. `mau/search` is BFS over
|
||||
ALL one-step successors (`mau/all-successors`) for reachability — solves the
|
||||
branching `goal` reachable only off the path `rew` takes. Verified: AC
|
||||
multiset coin-change (rule on a sub-multiset), cyclic traffic light (bounded),
|
||||
branching nondeterminism (rew vs search), conditional `crl` clock, eq/rule
|
||||
interleaving.
|
||||
- **PERF (important):** `lib/maude/fire.sx` is the short-circuiting matcher —
|
||||
`mau/fire-eq` finds the FIRST productive match via predicate-threaded
|
||||
`mau/ms-find`/`mau/seq-find` instead of materialising the whole solution
|
||||
set. Without it, AC rewriting on N identical elements is exponential
|
||||
(`q;q;q;q;q;q;q;q` went 60s+ → <1s). The eager `mau/match-multiset` /
|
||||
`mau/eq-candidates` are kept ONLY for `mau/match-all` and `search` (which
|
||||
truly need every solution). Phase 4 `creduce` and Phase 5 rules both fire
|
||||
via `mau/fire-eq`. Keep this split: never route single-step rewriting
|
||||
through the eager enumerator.
|
||||
- Notes: juxtaposition `__` (empty-token mixfix) and `gather` are NOT parsed —
|
||||
use an explicit infix op for multisets and right-parenthesise list literals.
|
||||
`.` can't be an op token (statement terminator). `mau/search` is the prime
|
||||
Phase 7 reflection / Phase 8 extraction target alongside the matcher.
|
||||
|
||||
## Blockers
|
||||
_(none)_
|
||||
|
||||
Reference in New Issue
Block a user