cdclkit 0.1.0__tar.gz → 0.1.2__tar.gz

{cdclkit-0.1.0 → cdclkit-0.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cdclkit
-Version: 0.1.0
+Version: 0.1.2
 Summary: A CDCL SAT solver, preprocessor, encoding library and modelling layer, written from scratch in readable Python. Every answer comes with a certificate.
 Author: Carlo Perassi
 License-Expression: Apache-2.0
@@ -23,6 +23,8 @@ Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: dratify>=0.1.1
+Provides-Extra: native
+Requires-Dist: cdclkit-native==0.1.2; extra == "native"
 Dynamic: license-file
 
 # cdclkit
@@ -33,9 +35,17 @@ library and a modelling layer — written from scratch in readable Python.**
 Every answer comes with a certificate, and the certificate gets checked.
 
 ```bash
-pip install cdclkit
+pip install cdclkit              # pure Python
+pip install "cdclkit[native]"    # plus the Rust engine, ~18x faster
 ```
 
+Three packages, and no third-party code:
+[`cdclkit`](https://pypi.org/project/cdclkit/),
+[`dratify`](https://pypi.org/project/dratify/) (the proof checker, zero
+dependencies), and optionally
+[`cdclkit-native`](https://pypi.org/project/cdclkit-native/) (abi3 wheels for
+Linux and macOS).
+
 ```bash
 python3 -m cdclkit solve instance.cnf --self-check --check-model
 ```
@@ -77,16 +87,19 @@ puzzle, graph colouring, bounded model checking, circuit equivalence.
 | `portfolio` | parallel configurations |
 | `pyeq` | **experimental** — bounded equivalence of two Python integer functions |
 
-An optional Rust engine (built from source with `make native` -- it is not
-on PyPI yet) is roughly 18x faster
+An optional Rust engine ([`cdclkit-native`](https://pypi.org/project/cdclkit-native/))
+is roughly 18x faster
 and **bit-exact** with the Python one: identical conflicts, decisions and
 propagations on every instance. The pure-Python path has zero third-party
 dependencies and is the one that must never break.
 
 ## Relationship to dratify
 
-Proof checking lives in a separate package, [`dratify`](https://github.com/carlok/dratify),
-which `cdclkit` depends on. That split is deliberate:
+Proof checking lives in a separate package,
+[`dratify`](https://pypi.org/project/dratify/)
+([source](https://github.com/carlok/dratify), also a
+[Rust crate](https://crates.io/crates/dratify)), which `cdclkit` depends on.
+That split is deliberate:
 
 - You should not have to install a SAT solver to verify a proof someone else
   produced.
@@ -95,7 +108,9 @@ which `cdclkit` depends on. That split is deliberate:
   solver rather than only by its own suite.
 
 `dratify` has no dependencies of its own, so installing `cdclkit` pulls in no
-third-party code.
+third-party code. With `[native]`, `cdclkit-native` hands its compiled checker
+to `dratify` through `register_native()`, so proof checking gets the Rust
+implementation too rather than only the solver.
 
 ## Performance
 

{cdclkit-0.1.0 → cdclkit-0.1.2}/README.md

@@ -6,9 +6,17 @@ library and a modelling layer — written from scratch in readable Python.**
 Every answer comes with a certificate, and the certificate gets checked.
 
 ```bash
-pip install cdclkit
+pip install cdclkit              # pure Python
+pip install "cdclkit[native]"    # plus the Rust engine, ~18x faster
 ```
 
+Three packages, and no third-party code:
+[`cdclkit`](https://pypi.org/project/cdclkit/),
+[`dratify`](https://pypi.org/project/dratify/) (the proof checker, zero
+dependencies), and optionally
+[`cdclkit-native`](https://pypi.org/project/cdclkit-native/) (abi3 wheels for
+Linux and macOS).
+
 ```bash
 python3 -m cdclkit solve instance.cnf --self-check --check-model
 ```
@@ -50,16 +58,19 @@ puzzle, graph colouring, bounded model checking, circuit equivalence.
 | `portfolio` | parallel configurations |
 | `pyeq` | **experimental** — bounded equivalence of two Python integer functions |
 
-An optional Rust engine (built from source with `make native` -- it is not
-on PyPI yet) is roughly 18x faster
+An optional Rust engine ([`cdclkit-native`](https://pypi.org/project/cdclkit-native/))
+is roughly 18x faster
 and **bit-exact** with the Python one: identical conflicts, decisions and
 propagations on every instance. The pure-Python path has zero third-party
 dependencies and is the one that must never break.
 
 ## Relationship to dratify
 
-Proof checking lives in a separate package, [`dratify`](https://github.com/carlok/dratify),
-which `cdclkit` depends on. That split is deliberate:
+Proof checking lives in a separate package,
+[`dratify`](https://pypi.org/project/dratify/)
+([source](https://github.com/carlok/dratify), also a
+[Rust crate](https://crates.io/crates/dratify)), which `cdclkit` depends on.
+That split is deliberate:
 
 - You should not have to install a SAT solver to verify a proof someone else
   produced.
@@ -68,7 +79,9 @@ which `cdclkit` depends on. That split is deliberate:
   solver rather than only by its own suite.
 
 `dratify` has no dependencies of its own, so installing `cdclkit` pulls in no
-third-party code.
+third-party code. With `[native]`, `cdclkit-native` hands its compiled checker
+to `dratify` through `register_native()`, so proof checking gets the Rust
+implementation too rather than only the solver.
 
 ## Performance
 

{cdclkit-0.1.0 → cdclkit-0.1.2}/cdclkit/__init__.py

@@ -52,7 +52,7 @@ from .preprocess import Preprocessor, preprocess
 from dratify.proof import DRATChecker, MemoryProof, ProofWriter, check_proof
 from .solver import Config, SAT, Solver, Stats, UNKNOWN, UNSAT
 
-__version__ = "0.1.0"
+__version__ = "0.1.2"
 
 __all__ = [
     "CNF",

{cdclkit-0.1.0 → cdclkit-0.1.2}/cdclkit/pipeline.py

@@ -92,6 +92,7 @@ def _solve(f: CNF, cfg: Config, budget: int | None, engine: str,
             cla_decay=cfg.cla_decay, luby_base=float(cfg.luby_base),
             first_reduce=cfg.first_reduce, reduce_inc=cfg.reduce_inc,
             glue_keep=cfg.glue_keep, block_restart=cfg.block_restart,
+            rnd_freq=cfg.rnd_freq, rnd_seed=cfg.rnd_seed,
         )
         for c in f.clauses:
             if not s.add_clause(list(c)):

{cdclkit-0.1.0 → cdclkit-0.1.2}/cdclkit/portfolio.py

@@ -235,6 +235,7 @@ def _worker_native(index, nvars, clauses, cfg_kwargs, want_proof):
         cla_decay=cfg.cla_decay, luby_base=float(cfg.luby_base),
         first_reduce=cfg.first_reduce, reduce_inc=cfg.reduce_inc,
         glue_keep=cfg.glue_keep, block_restart=cfg.block_restart,
+        rnd_freq=cfg.rnd_freq, rnd_seed=cfg.rnd_seed,
     )
     if want_proof:
         s.enable_proof()  # must precede the first clause
@@ -416,7 +417,8 @@ def _config_tuples(configs):
              var_decay=c.var_decay, var_decay_max=c.var_decay_max,
              cla_decay=c.cla_decay, luby_base=float(c.luby_base),
              first_reduce=c.first_reduce, reduce_inc=c.reduce_inc,
-             glue_keep=c.glue_keep, block_restart=c.block_restart)
+             glue_keep=c.glue_keep, block_restart=c.block_restart,
+             rnd_freq=c.rnd_freq, rnd_seed=c.rnd_seed)
         for c in configs
     ]
 

{cdclkit-0.1.0 → cdclkit-0.1.2}/cdclkit.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cdclkit
-Version: 0.1.0
+Version: 0.1.2
 Summary: A CDCL SAT solver, preprocessor, encoding library and modelling layer, written from scratch in readable Python. Every answer comes with a certificate.
 Author: Carlo Perassi
 License-Expression: Apache-2.0
@@ -23,6 +23,8 @@ Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: dratify>=0.1.1
+Provides-Extra: native
+Requires-Dist: cdclkit-native==0.1.2; extra == "native"
 Dynamic: license-file
 
 # cdclkit
@@ -33,9 +35,17 @@ library and a modelling layer — written from scratch in readable Python.**
 Every answer comes with a certificate, and the certificate gets checked.
 
 ```bash
-pip install cdclkit
+pip install cdclkit              # pure Python
+pip install "cdclkit[native]"    # plus the Rust engine, ~18x faster
 ```
 
+Three packages, and no third-party code:
+[`cdclkit`](https://pypi.org/project/cdclkit/),
+[`dratify`](https://pypi.org/project/dratify/) (the proof checker, zero
+dependencies), and optionally
+[`cdclkit-native`](https://pypi.org/project/cdclkit-native/) (abi3 wheels for
+Linux and macOS).
+
 ```bash
 python3 -m cdclkit solve instance.cnf --self-check --check-model
 ```
@@ -77,16 +87,19 @@ puzzle, graph colouring, bounded model checking, circuit equivalence.
 | `portfolio` | parallel configurations |
 | `pyeq` | **experimental** — bounded equivalence of two Python integer functions |
 
-An optional Rust engine (built from source with `make native` -- it is not
-on PyPI yet) is roughly 18x faster
+An optional Rust engine ([`cdclkit-native`](https://pypi.org/project/cdclkit-native/))
+is roughly 18x faster
 and **bit-exact** with the Python one: identical conflicts, decisions and
 propagations on every instance. The pure-Python path has zero third-party
 dependencies and is the one that must never break.
 
 ## Relationship to dratify
 
-Proof checking lives in a separate package, [`dratify`](https://github.com/carlok/dratify),
-which `cdclkit` depends on. That split is deliberate:
+Proof checking lives in a separate package,
+[`dratify`](https://pypi.org/project/dratify/)
+([source](https://github.com/carlok/dratify), also a
+[Rust crate](https://crates.io/crates/dratify)), which `cdclkit` depends on.
+That split is deliberate:
 
 - You should not have to install a SAT solver to verify a proof someone else
   produced.
@@ -95,7 +108,9 @@ which `cdclkit` depends on. That split is deliberate:
   solver rather than only by its own suite.
 
 `dratify` has no dependencies of its own, so installing `cdclkit` pulls in no
-third-party code.
+third-party code. With `[native]`, `cdclkit-native` hands its compiled checker
+to `dratify` through `register_native()`, so proof checking gets the Rust
+implementation too rather than only the solver.
 
 ## Performance
 

cdclkit-0.1.2/cdclkit.egg-info/requires.txt

@@ -0,0 +1,4 @@
+dratify>=0.1.1
+
+[native]
+cdclkit-native==0.1.2

{cdclkit-0.1.0 → cdclkit-0.1.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "cdclkit"
-version = "0.1.0"
+version = "0.1.2"
 description = "A CDCL SAT solver, preprocessor, encoding library and modelling layer, written from scratch in readable Python. Every answer comes with a certificate."
 readme = "README.md"
 requires-python = ">=3.10"
@@ -31,6 +31,12 @@ classifiers = [
 # test run here as well as by its own suite.
 dependencies = ["dratify>=0.1.1"]
 
+[project.optional-dependencies]
+# The Rust accelerator, as abi3 wheels for Linux and macOS. Pinned exactly:
+# the two are bit-exact by construction and a mismatched pair is a divergence
+# nobody would look for. tests/test_packaging.py enforces the pin.
+native = ["cdclkit-native==0.1.2"]
+
 [project.scripts]
 cdclkit = "cdclkit.__main__:main"
 

{cdclkit-0.1.0 → cdclkit-0.1.2}/tests/test_native_solver.py

@@ -160,6 +160,85 @@ class TestDefaultsAgree(unittest.TestCase):
         self.assertEqual(py.stats.propagations, rs.propagations)
 
 
+@requires_native
+class TestRandomBranchingAgrees(unittest.TestCase):
+    """`rnd_freq` and `rnd_seed` must reach the native engine and match.
+
+    These two knobs had no Rust counterpart at all: the native side accepted
+    the rest of the configuration and silently dropped these, so a Python run
+    at rnd_freq=0.02 took 4,987 conflicts where the native one took 3,171.
+
+    That was not a cosmetic gap. `portfolio.py` builds worker diversity from
+    exactly these two fields, and sets a per-worker seed specifically "so
+    duplicated recipes still diverge" -- which on the native path they did not,
+    because every worker got the same search. Duplicate recipes became
+    bit-identical duplicate workers doing the same work in parallel.
+
+    The existing bit-exactness tests could not have caught this: none of them
+    set either knob.
+    """
+
+    def _native(self, f, rnd_freq, rnd_seed):
+        n = native.require()
+        d = Config()
+        rs = n.Solver(
+            f.nvars, restart=d.restart, ccmin=d.ccmin,
+            phase_saving=d.phase_saving, init_phase=d.init_phase,
+            target_phase=d.target_phase, target_reset=d.target_reset,
+            walk_flips=d.walk_flips, walk_interval=d.walk_interval,
+            walk_patience=d.walk_patience,
+            walk_min_conflicts=d.walk_min_conflicts,
+            var_decay=d.var_decay, var_decay_max=d.var_decay_max,
+            cla_decay=d.cla_decay, luby_base=float(d.luby_base),
+            first_reduce=d.first_reduce, reduce_inc=d.reduce_inc,
+            glue_keep=d.glue_keep, block_restart=d.block_restart,
+            rnd_freq=rnd_freq, rnd_seed=rnd_seed)
+        for c in f.clauses:
+            rs.add_clause(list(c))
+        rs.solve(None)
+        return rs.conflicts, rs.decisions, rs.propagations
+
+    def _python(self, f, rnd_freq, rnd_seed):
+        from cdclkit.solver import Solver as PySolver
+
+        cfg = Config()
+        cfg.rnd_freq = rnd_freq
+        cfg.rnd_seed = rnd_seed
+        py = PySolver(f.nvars, config=cfg)
+        py.add_cnf(f)
+        py.solve()
+        return py.stats.conflicts, py.stats.decisions, py.stats.propagations
+
+    def test_engines_agree_across_frequencies_and_seeds(self):
+        f = php(6)
+        for freq, seed in [(0.0, 91648253), (0.02, 91648253), (0.02, 12345),
+                           (0.10, 7), (0.05, 999983)]:
+            with self.subTest(rnd_freq=freq, rnd_seed=seed):
+                self.assertEqual(
+                    self._python(f, freq, seed), self._native(f, freq, seed),
+                    f"engines diverge at rnd_freq={freq}, rnd_seed={seed}")
+
+    def test_a_zero_frequency_does_not_touch_the_random_stream(self):
+        """The short-circuit is load-bearing, not an optimisation.
+
+        Python tests `rnd_freq > 0.0` before drawing, so the default
+        configuration never advances the PRNG. If the native side drew first,
+        the walk -- which shares the stream -- would desynchronise and the two
+        engines would diverge on instances long enough to reach it.
+        """
+        f = php(6)
+        self.assertEqual(self._python(f, 0.0, 91648253),
+                         self._native(f, 0.0, 91648253))
+
+    def test_the_seed_actually_changes_the_search(self):
+        """A knob that is wired up but inert would pass the tests above."""
+        f = php(6)
+        runs = {self._native(f, 0.05, seed) for seed in (1, 7, 12345, 999983)}
+        self.assertGreater(len(runs), 1,
+                           "every seed produced an identical search, so the "
+                           "seed is not reaching the engine")
+
+
 @requires_native
 class TestTier2Faithfulness(unittest.TestCase):
     """Bit-exact agreement with the Python solver."""

{cdclkit-0.1.0 → cdclkit-0.1.2}/tests/test_packaging.py

@@ -157,6 +157,39 @@ class TestPublicAPISurface(unittest.TestCase):
             self.assertNotIn(internal, cdclkit.__all__)
 
 
+class TestWorkflowsAreWellFormed(unittest.TestCase):
+    """Every top-level line in a workflow must be a known key or a comment.
+
+    A comment in release.yml lost its leading "# " during an edit. The result
+    was still valid YAML -- it simply became a mapping key -- so a
+    `yaml.safe_load` check passed it, and GitHub then refused the whole file
+    with "this run likely failed because of a workflow file issue". Zero jobs
+    ran, and the tag had to be deleted and re-cut.
+
+    Deliberately does not import yaml: the dependency-free test path is the one
+    that must never break, and this failure mode does not need a parser.
+    """
+
+    #: https://docs.github.com/actions/reference/workflow-syntax-for-github-actions
+    ALLOWED = {"name", "on", "permissions", "env", "defaults", "concurrency",
+               "jobs", "run-name"}
+
+    def test_no_stray_top_level_keys(self):
+        workflows = sorted((ROOT / ".github" / "workflows").glob("*.yml"))
+        self.assertTrue(workflows, "no workflows found")
+        for wf in workflows:
+            with self.subTest(workflow=wf.name):
+                for n, line in enumerate(wf.read_text().splitlines(), 1):
+                    if not line or line[0] in " \t#":
+                        continue
+                    key = line.split(":", 1)[0].strip()
+                    self.assertIn(
+                        key, self.ALLOWED,
+                        f"{wf.name}:{n} starts a top-level key {key!r} that "
+                        f"GitHub does not recognise -- most likely a comment "
+                        f"that lost its '# '")
+
+
 class TestLicenceConsistency(unittest.TestCase):
     """The tree is Apache-2.0; every place that states a licence must agree.
 

cdclkit-0.1.0/cdclkit.egg-info/requires.txt

@@ -1 +0,0 @@
-dratify>=0.1.1