domain.py
python
| 1 | """MuseDomainPlugin — the six-interface protocol that defines a Muse domain. |
| 2 | |
| 3 | Muse provides the DAG engine, content-addressed object store, branching, |
| 4 | lineage walking, topological log graph, and merge base finder. A domain plugin |
| 5 | implements these six interfaces and Muse does the rest. |
| 6 | |
| 7 | The music plugin (``muse.plugins.midi``) is the reference implementation. |
| 8 | Every other domain — scientific simulation, genomics, 3D spatial design, |
| 9 | spacetime — is a new plugin. |
| 10 | |
| 11 | Typed Delta Algebra |
| 12 | ------------------- |
| 13 | ``StateDelta`` is a ``StructuredDelta`` carrying a typed operation list rather |
| 14 | than an opaque path list. Each operation knows its kind (insert / delete / |
| 15 | move / replace / patch), the address it touched, and a content-addressed ID |
| 16 | for the before/after content. |
| 17 | |
| 18 | Domain Schema |
| 19 | ------------- |
| 20 | ``schema()`` is the sixth protocol method. Plugins return a ``DomainSchema`` |
| 21 | declaring their data structure. The core engine uses this declaration to drive |
| 22 | diff algorithm selection via :func:`~muse.core.diff_algorithms.diff_by_schema`. |
| 23 | |
| 24 | Operational Transformation Merge |
| 25 | --------------------------------- |
| 26 | Plugins may optionally implement :class:`StructuredMergePlugin`, a sub-protocol |
| 27 | that adds ``merge_ops()``. When both branches have produced ``StructuredDelta`` |
| 28 | from ``diff()``, the merge engine checks |
| 29 | ``isinstance(plugin, StructuredMergePlugin)`` and calls ``merge_ops()`` for |
| 30 | fine-grained, operation-level conflict detection. Non-supporting plugins fall |
| 31 | back to the existing file-level ``merge()`` path. |
| 32 | |
| 33 | CRDT Convergent Merge |
| 34 | --------------------- |
| 35 | Plugins may optionally implement :class:`CRDTPlugin`, a sub-protocol that |
| 36 | replaces ``merge()`` with ``join()``. ``join`` always succeeds — no conflict |
| 37 | state ever exists. Given any two :class:`CRDTSnapshotManifest` values, |
| 38 | ``join`` produces a deterministic merged result regardless of message delivery |
| 39 | order. |
| 40 | |
| 41 | The core engine detects ``CRDTPlugin`` via ``isinstance`` at merge time. |
| 42 | ``DomainSchema.merge_mode == "crdt"`` signals that the CRDT path should be |
| 43 | taken. |
| 44 | """ |
| 45 | |
| 46 | import pathlib |
| 47 | from dataclasses import dataclass, field |
| 48 | from typing import TYPE_CHECKING, Literal, Protocol, TypedDict, runtime_checkable |
| 49 | |
| 50 | # Public re-exports so callers can do ``from muse.domain import MutateOp`` etc. |
| 51 | __all__ = [ |
| 52 | "SnapshotManifest", |
| 53 | "DomainAddress", |
| 54 | "InsertOp", |
| 55 | "DeleteOp", |
| 56 | "MoveOp", |
| 57 | "ReplaceOp", |
| 58 | "FieldMutation", |
| 59 | "MutateOp", |
| 60 | "EntityProvenance", |
| 61 | "LeafDomainOp", |
| 62 | "PatchOp", |
| 63 | "DomainOp", |
| 64 | "SemVerBump", |
| 65 | "StructuredDelta", |
| 66 | "infer_sem_ver_bump", |
| 67 | "LiveState", |
| 68 | "StateSnapshot", |
| 69 | "StateDelta", |
| 70 | "ConflictRecord", |
| 71 | "MergeResult", |
| 72 | "DriftReport", |
| 73 | "MuseDomainPlugin", |
| 74 | "StructuredMergePlugin", |
| 75 | "CRDTSnapshotManifest", |
| 76 | "CRDTPlugin", |
| 77 | ] |
| 78 | |
| 79 | if TYPE_CHECKING: |
| 80 | from muse.core.schema import CRDTDimensionSpec, DomainSchema |
| 81 | |
| 82 | |
| 83 | # --------------------------------------------------------------------------- |
| 84 | # Snapshot types (unchanged from pre-Phase-1) |
| 85 | # --------------------------------------------------------------------------- |
| 86 | |
| 87 | |
| 88 | class SnapshotManifest(TypedDict): |
| 89 | """Content-addressed snapshot of domain state. |
| 90 | |
| 91 | ``files`` maps workspace-relative POSIX paths to their SHA-256 content |
| 92 | digests. ``domain`` identifies which plugin produced this snapshot. |
| 93 | """ |
| 94 | |
| 95 | files: dict[str, str] |
| 96 | domain: str |
| 97 | |
| 98 | |
| 99 | # --------------------------------------------------------------------------- |
| 100 | # Typed delta algebra |
| 101 | # --------------------------------------------------------------------------- |
| 102 | |
| 103 | #: A domain-specific address identifying a location within the state graph. |
| 104 | #: For file-level ops this is a workspace-relative POSIX path. |
| 105 | #: For sub-file ops this is a domain-specific coordinate (e.g. "note:42"). |
| 106 | DomainAddress = str |
| 107 | |
| 108 | |
| 109 | class InsertOp(TypedDict): |
| 110 | """An element was inserted into a collection. |
| 111 | |
| 112 | For ordered sequences ``position`` is the integer index at which the |
| 113 | element was inserted. For unordered sets ``position`` is ``None``. |
| 114 | ``content_id`` is the SHA-256 of the inserted content — either a blob |
| 115 | already in the object store (for file-level ops) or a deterministic hash |
| 116 | of the element's canonical serialisation (for sub-file ops). |
| 117 | """ |
| 118 | |
| 119 | op: Literal["insert"] |
| 120 | address: DomainAddress |
| 121 | position: int | None |
| 122 | content_id: str |
| 123 | content_summary: str |
| 124 | |
| 125 | |
| 126 | class DeleteOp(TypedDict): |
| 127 | """An element was removed from a collection. |
| 128 | |
| 129 | ``position`` is the integer index that was removed for ordered sequences, |
| 130 | or ``None`` for unordered sets. ``content_id`` is the SHA-256 of the |
| 131 | deleted content so that the operation can be applied idempotently (already- |
| 132 | absent elements can be skipped). ``content_summary`` is the human-readable |
| 133 | description of what was removed, for ``muse show``. |
| 134 | """ |
| 135 | |
| 136 | op: Literal["delete"] |
| 137 | address: DomainAddress |
| 138 | position: int | None |
| 139 | content_id: str |
| 140 | content_summary: str |
| 141 | |
| 142 | |
| 143 | class MoveOp(TypedDict): |
| 144 | """An element was repositioned within an ordered sequence. |
| 145 | |
| 146 | ``from_position`` is the source index (in the pre-move sequence) and |
| 147 | ``to_position`` is the destination index (in the post-move sequence). |
| 148 | Both are mandatory — moves are only meaningful in ordered collections. |
| 149 | ``content_id`` identifies the element being moved so that the operation |
| 150 | can be validated during replay. |
| 151 | """ |
| 152 | |
| 153 | op: Literal["move"] |
| 154 | address: DomainAddress |
| 155 | from_position: int |
| 156 | to_position: int |
| 157 | content_id: str |
| 158 | |
| 159 | |
| 160 | class ReplaceOp(TypedDict): |
| 161 | """An element's value changed (atomic, leaf-level replacement). |
| 162 | |
| 163 | ``old_content_id`` and ``new_content_id`` are SHA-256 hashes of the |
| 164 | before- and after-content. They enable three-way merge engines to detect |
| 165 | concurrent conflicting modifications (both changed from the same |
| 166 | ``old_content_id`` to different ``new_content_id`` values). |
| 167 | ``old_summary`` and ``new_summary`` are human-readable strings for display, |
| 168 | analogous to ``content_summary`` on :class:`InsertOp`. |
| 169 | ``position`` is the index within the container (``None`` for unordered). |
| 170 | """ |
| 171 | |
| 172 | op: Literal["replace"] |
| 173 | address: DomainAddress |
| 174 | position: int | None |
| 175 | old_content_id: str |
| 176 | new_content_id: str |
| 177 | old_summary: str |
| 178 | new_summary: str |
| 179 | |
| 180 | |
| 181 | class FieldMutation(TypedDict): |
| 182 | """The string-serialised before/after of a single field in a :class:`MutateOp`. |
| 183 | |
| 184 | Values are always strings so that typed primitives (int, float, bool) can |
| 185 | be compared uniformly without carrying domain-specific type information in |
| 186 | the generic delta algebra. Plugins format them according to their domain |
| 187 | conventions (e.g. ``"80"`` for a MIDI velocity, ``"C4"`` for a pitch name). |
| 188 | """ |
| 189 | |
| 190 | old: str |
| 191 | new: str |
| 192 | |
| 193 | |
| 194 | class MutateOp(TypedDict): |
| 195 | """A named entity's specific fields were updated. |
| 196 | |
| 197 | Unlike :class:`ReplaceOp` — which replaces an entire element atomically — |
| 198 | ``MutateOp`` records *which* specific fields of a domain entity changed. |
| 199 | This enables mutation tracking for domains that maintain stable entity |
| 200 | identity separate from content equality. |
| 201 | |
| 202 | Example: a MIDI note's velocity changed from 80 to 100. Under a pure |
| 203 | content-hash model that becomes ``DeleteOp + InsertOp`` (two different |
| 204 | content hashes). With ``MutateOp`` and a stable ``entity_id`` the diff |
| 205 | reports "velocity 80→100 on entity C4@bar4" — lineage is preserved. |
| 206 | |
| 207 | ``entity_id`` |
| 208 | Stable identifier for the mutated entity, assigned at first insertion |
| 209 | and reused across all subsequent mutations (regardless of content |
| 210 | changes). |
| 211 | ``fields`` |
| 212 | Mapping from field name (e.g. ``"velocity"``, ``"start_tick"``) to a |
| 213 | :class:`FieldMutation` recording the serialised old and new values. |
| 214 | ``old_content_id`` / ``new_content_id`` |
| 215 | SHA-256 of the full element state before and after the mutation, |
| 216 | enabling three-way merge conflict detection identical to |
| 217 | :class:`ReplaceOp`. |
| 218 | ``position`` |
| 219 | Index within the containing ordered sequence (``None`` for unordered). |
| 220 | """ |
| 221 | |
| 222 | op: Literal["mutate"] |
| 223 | address: DomainAddress |
| 224 | entity_id: str |
| 225 | old_content_id: str |
| 226 | new_content_id: str |
| 227 | fields: dict[str, FieldMutation] |
| 228 | old_summary: str |
| 229 | new_summary: str |
| 230 | position: int | None |
| 231 | |
| 232 | |
| 233 | class EntityProvenance(TypedDict, total=False): |
| 234 | """Causal metadata attached to ops that create or modify tracked entities. |
| 235 | |
| 236 | All fields are optional (``total=False``) because entity tracking is an |
| 237 | opt-in capability. Plugins that implement stable entity identity populate |
| 238 | these fields when constructing :class:`InsertOp`, :class:`MutateOp`, or |
| 239 | :class:`DeleteOp` entries. Consumers that do not understand entity |
| 240 | provenance can safely ignore them. |
| 241 | |
| 242 | ``entity_id`` |
| 243 | Stable domain-specific identifier for the entity (e.g. a UUID assigned |
| 244 | at the note's first insertion). |
| 245 | ``origin_op_id`` |
| 246 | The ``op_id`` of the op that first created this entity. |
| 247 | ``last_modified_op_id`` |
| 248 | The ``op_id`` of the most recent op that touched this entity. |
| 249 | ``created_at_commit`` |
| 250 | Short-form commit ID where this entity was first introduced. |
| 251 | ``actor_id`` |
| 252 | The agent or human identity that performed this op. |
| 253 | """ |
| 254 | |
| 255 | entity_id: str |
| 256 | origin_op_id: str |
| 257 | last_modified_op_id: str |
| 258 | created_at_commit: str |
| 259 | actor_id: str |
| 260 | |
| 261 | |
| 262 | #: The five non-recursive (leaf) operation types. |
| 263 | LeafDomainOp = InsertOp | DeleteOp | MoveOp | ReplaceOp | MutateOp |
| 264 | |
| 265 | |
| 266 | class PatchOp(TypedDict): |
| 267 | """A container element was internally modified. |
| 268 | |
| 269 | ``address`` names the container (e.g. a file path). ``child_ops`` lists |
| 270 | the sub-element changes inside that container. These are always |
| 271 | leaf ops in the current implementation; true recursion via a nested |
| 272 | ``StructuredDelta`` is reserved for a future release. |
| 273 | |
| 274 | ``child_domain`` identifies the sub-element domain (e.g. ``"midi_notes"`` |
| 275 | for note-level ops inside a ``.mid`` file). ``child_summary`` is a |
| 276 | human-readable description of the child changes for ``muse show``. |
| 277 | """ |
| 278 | |
| 279 | op: Literal["patch"] |
| 280 | address: DomainAddress |
| 281 | child_ops: list[DomainOp] |
| 282 | child_domain: str |
| 283 | child_summary: str |
| 284 | |
| 285 | |
| 286 | #: Union of all operation types — the atoms of a ``StructuredDelta``. |
| 287 | type DomainOp = LeafDomainOp | PatchOp |
| 288 | |
| 289 | |
| 290 | SemVerBump = Literal["major", "minor", "patch", "none"] |
| 291 | """Semantic version impact of a delta. |
| 292 | |
| 293 | ``major`` Breaking change: public symbol deleted, renamed, or signature changed. |
| 294 | ``minor`` Additive: new public symbol inserted. |
| 295 | ``patch`` Implementation-only change: body changed, signature stable. |
| 296 | ``none`` No semantic change (formatting, whitespace, metadata only). |
| 297 | """ |
| 298 | |
| 299 | class StructuredDelta(TypedDict, total=False): |
| 300 | """Rich, composable delta between two domain snapshots. |
| 301 | |
| 302 | ``ops`` is an ordered list of operations that transforms ``base`` into |
| 303 | ``target`` when applied in sequence. The core engine stores this alongside |
| 304 | commit records so that ``muse show`` and ``muse diff`` can display it |
| 305 | without reloading full blobs. |
| 306 | |
| 307 | ``summary`` is a precomputed human-readable string — for example |
| 308 | ``"3 notes added, 1 note removed"``. Plugins compute it because only they |
| 309 | understand their domain semantics. |
| 310 | |
| 311 | ``sem_ver_bump`` (v2, optional) is the semantic version impact of this |
| 312 | delta, computed by :func:`infer_sem_ver_bump`. Absent for legacy records |
| 313 | or non-code domains that do not compute it. |
| 314 | |
| 315 | ``breaking_changes`` (v2, optional) lists the symbol addresses whose |
| 316 | public interface was removed or incompatibly changed. |
| 317 | """ |
| 318 | |
| 319 | domain: str |
| 320 | ops: list[DomainOp] |
| 321 | summary: str |
| 322 | sem_ver_bump: SemVerBump |
| 323 | breaking_changes: list[str] |
| 324 | |
| 325 | |
| 326 | # --------------------------------------------------------------------------- |
| 327 | # SemVer inference helper |
| 328 | # --------------------------------------------------------------------------- |
| 329 | |
| 330 | |
| 331 | def infer_sem_ver_bump(delta: "StructuredDelta") -> tuple[SemVerBump, list[str]]: |
| 332 | """Infer the semantic version bump and breaking-change list from a delta. |
| 333 | |
| 334 | Reads the ``ops`` list and applies the following rules: |
| 335 | |
| 336 | * Any public symbol (name not starting with ``_``) that is deleted or |
| 337 | renamed → **major** (breaking: callers will fail). |
| 338 | * Any public symbol whose ``signature_id`` changed (signature_only or |
| 339 | full_rewrite with new signature) → **major** (breaking: call-site |
| 340 | compatibility broken). |
| 341 | * Any public symbol inserted → **minor** (additive). |
| 342 | * Any symbol whose only change is the body (``impl_only``) → **patch**. |
| 343 | * No semantic ops → **none**. |
| 344 | |
| 345 | Returns: |
| 346 | A ``(bump, breaking_changes)`` tuple where ``breaking_changes`` is a |
| 347 | sorted list of symbol addresses whose public contract changed. |
| 348 | |
| 349 | This function is domain-agnostic; it relies on the op address format used |
| 350 | by code plugins (``<file>::<symbol>``) and the ``new_summary`` / ``old_summary`` |
| 351 | conventions from :func:`~muse.plugins.code.symbol_diff.diff_symbol_trees`. |
| 352 | For non-code domains the heuristics may not apply — plugins should override |
| 353 | by setting ``sem_ver_bump`` directly when constructing the delta. |
| 354 | """ |
| 355 | ops = delta.get("ops", []) |
| 356 | bump: SemVerBump = "none" |
| 357 | breaking: list[str] = [] |
| 358 | |
| 359 | def _is_public(address: str) -> bool: |
| 360 | """Return True if the innermost symbol name does not start with ``_``.""" |
| 361 | parts = address.split("::") |
| 362 | name = parts[-1].split(".")[-1] if parts else "" |
| 363 | return not name.startswith("_") |
| 364 | |
| 365 | def _promote(current: SemVerBump, candidate: SemVerBump) -> SemVerBump: |
| 366 | order: list[SemVerBump] = ["none", "patch", "minor", "major"] |
| 367 | return candidate if order.index(candidate) > order.index(current) else current |
| 368 | |
| 369 | for op in ops: |
| 370 | op_type = op.get("op", "") |
| 371 | address = str(op.get("address", "")) |
| 372 | |
| 373 | if op_type == "patch": |
| 374 | # Recurse into child_ops. We know op is a PatchOp here. |
| 375 | if op["op"] == "patch": |
| 376 | child_ops_raw: list[DomainOp] = op["child_ops"] |
| 377 | sub_delta: StructuredDelta = {"domain": "", "ops": child_ops_raw, "summary": ""} |
| 378 | sub_bump, sub_breaking = infer_sem_ver_bump(sub_delta) |
| 379 | bump = _promote(bump, sub_bump) |
| 380 | breaking.extend(sub_breaking) |
| 381 | continue |
| 382 | |
| 383 | if not _is_public(address): |
| 384 | continue |
| 385 | |
| 386 | if op_type == "delete": |
| 387 | bump = _promote(bump, "major") |
| 388 | breaking.append(address) |
| 389 | |
| 390 | elif op_type == "insert": |
| 391 | bump = _promote(bump, "minor") |
| 392 | |
| 393 | elif op_type == "replace": |
| 394 | new_summary: str = str(op.get("new_summary", "")) |
| 395 | old_summary: str = str(op.get("old_summary", "")) |
| 396 | if ( |
| 397 | new_summary.startswith("renamed to ") |
| 398 | or "signature" in new_summary |
| 399 | or "signature" in old_summary |
| 400 | ): |
| 401 | bump = _promote(bump, "major") |
| 402 | breaking.append(address) |
| 403 | elif "implementation" in new_summary or "implementation" in old_summary: |
| 404 | bump = _promote(bump, "patch") |
| 405 | else: |
| 406 | bump = _promote(bump, "major") |
| 407 | breaking.append(address) |
| 408 | |
| 409 | return bump, sorted(set(breaking)) |
| 410 | |
| 411 | |
| 412 | # --------------------------------------------------------------------------- |
| 413 | # Type aliases used in the protocol signatures |
| 414 | # --------------------------------------------------------------------------- |
| 415 | |
| 416 | #: Live state is either an already-snapshotted manifest dict or a workdir path. |
| 417 | #: The music plugin accepts both: a Path (for CLI commit/status) and a |
| 418 | #: SnapshotManifest dict (for in-memory merge and diff operations). |
| 419 | type LiveState = SnapshotManifest | pathlib.Path |
| 420 | |
| 421 | #: A content-addressed, immutable snapshot of state at a point in time. |
| 422 | type StateSnapshot = SnapshotManifest |
| 423 | |
| 424 | #: The minimal change between two snapshots — a list of typed domain operations. |
| 425 | type StateDelta = StructuredDelta |
| 426 | |
| 427 | |
| 428 | # --------------------------------------------------------------------------- |
| 429 | # Merge and drift result types |
| 430 | # --------------------------------------------------------------------------- |
| 431 | |
| 432 | |
| 433 | @dataclass |
| 434 | class ConflictRecord: |
| 435 | """Structured conflict record in a merge result (v2 taxonomy). |
| 436 | |
| 437 | ``path`` The workspace-relative file path in conflict. |
| 438 | ``conflict_type`` One of: ``symbol_edit_overlap``, ``rename_edit``, |
| 439 | ``move_edit``, ``delete_use``, ``dependency_conflict``, |
| 440 | ``file_level`` (legacy, no symbol info). |
| 441 | ``ours_summary`` Short description of ours-side change. |
| 442 | ``theirs_summary`` Short description of theirs-side change. |
| 443 | ``addresses`` Symbol addresses involved (empty for file-level). |
| 444 | """ |
| 445 | |
| 446 | path: str |
| 447 | conflict_type: str = "file_level" |
| 448 | ours_summary: str = "" |
| 449 | theirs_summary: str = "" |
| 450 | addresses: list[str] = field(default_factory=list) |
| 451 | |
| 452 | def to_dict(self) -> dict[str, str | list[str]]: |
| 453 | return { |
| 454 | "path": self.path, |
| 455 | "conflict_type": self.conflict_type, |
| 456 | "ours_summary": self.ours_summary, |
| 457 | "theirs_summary": self.theirs_summary, |
| 458 | "addresses": self.addresses, |
| 459 | } |
| 460 | |
| 461 | |
| 462 | @dataclass |
| 463 | class MergeResult: |
| 464 | """Outcome of a three-way merge between two divergent state lines. |
| 465 | |
| 466 | ``merged`` is the reconciled snapshot. ``conflicts`` is a list of |
| 467 | workspace-relative file paths that could not be auto-merged and require |
| 468 | manual resolution. An empty ``conflicts`` list means the merge was clean. |
| 469 | The CLI is responsible for formatting user-facing messages from these paths. |
| 470 | |
| 471 | ``applied_strategies`` maps each path where a ``.museattributes`` rule |
| 472 | overrode the default conflict behaviour to the strategy that was applied. |
| 473 | |
| 474 | ``dimension_reports`` maps conflicting paths to their per-dimension |
| 475 | resolution detail. |
| 476 | |
| 477 | ``op_log`` is the ordered list of ``DomainOp`` entries applied to produce |
| 478 | the merged snapshot. Empty for file-level merges; populated by plugins |
| 479 | that implement operation-level OT merge. |
| 480 | |
| 481 | ``conflict_records`` (v2) provides structured conflict metadata with a |
| 482 | semantic taxonomy per conflicting path. Populated by plugins that |
| 483 | implement :class:`StructuredMergePlugin`. May be empty even when |
| 484 | ``conflicts`` is non-empty (legacy file-level conflict). |
| 485 | """ |
| 486 | |
| 487 | merged: StateSnapshot |
| 488 | conflicts: list[str] = field(default_factory=list) |
| 489 | applied_strategies: dict[str, str] = field(default_factory=dict) |
| 490 | dimension_reports: dict[str, dict[str, str]] = field(default_factory=dict) |
| 491 | op_log: list[DomainOp] = field(default_factory=list) |
| 492 | conflict_records: list[ConflictRecord] = field(default_factory=list) |
| 493 | |
| 494 | @property |
| 495 | def is_clean(self) -> bool: |
| 496 | """``True`` when no unresolvable conflicts remain.""" |
| 497 | return len(self.conflicts) == 0 |
| 498 | |
| 499 | |
| 500 | @dataclass |
| 501 | class DriftReport: |
| 502 | """Gap between committed state and current live state. |
| 503 | |
| 504 | ``has_drift`` is ``True`` when the live state differs from the committed |
| 505 | snapshot. ``summary`` is a human-readable description of what changed. |
| 506 | ``delta`` is the machine-readable structured delta for programmatic consumers. |
| 507 | """ |
| 508 | |
| 509 | has_drift: bool |
| 510 | summary: str = "" |
| 511 | delta: StateDelta = field(default_factory=lambda: StructuredDelta( |
| 512 | domain="", ops=[], summary="working tree clean", |
| 513 | )) |
| 514 | |
| 515 | |
| 516 | # --------------------------------------------------------------------------- |
| 517 | # The plugin protocol |
| 518 | # --------------------------------------------------------------------------- |
| 519 | |
| 520 | |
| 521 | @runtime_checkable |
| 522 | class MuseDomainPlugin(Protocol): |
| 523 | """The six interfaces a domain plugin must implement. |
| 524 | |
| 525 | Muse provides everything else: the DAG, branching, checkout, lineage |
| 526 | walking, ASCII log graph, and merge base finder. Implement these six |
| 527 | methods and your domain gets the full Muse VCS for free. |
| 528 | |
| 529 | Music is the reference implementation (``muse.plugins.midi``). |
| 530 | """ |
| 531 | |
| 532 | def snapshot(self, live_state: LiveState) -> StateSnapshot: |
| 533 | """Capture current live state as a serialisable, hashable snapshot. |
| 534 | |
| 535 | The returned ``SnapshotManifest`` must be JSON-serialisable. Muse will |
| 536 | compute a SHA-256 content address from the canonical JSON form and |
| 537 | store the snapshot as a blob in ``.muse/objects/``. |
| 538 | |
| 539 | **``.museignore`` contract** — when *live_state* is a |
| 540 | ``pathlib.Path`` (the ``muse-work/`` directory), domain plugin |
| 541 | implementations **must** honour ``.museignore`` by calling |
| 542 | :func:`muse.core.ignore.load_patterns` on the repository root and |
| 543 | filtering out paths matched by :func:`muse.core.ignore.is_ignored`. |
| 544 | """ |
| 545 | ... |
| 546 | |
| 547 | def diff( |
| 548 | self, |
| 549 | base: StateSnapshot, |
| 550 | target: StateSnapshot, |
| 551 | *, |
| 552 | repo_root: pathlib.Path | None = None, |
| 553 | ) -> StateDelta: |
| 554 | """Compute the structured delta between two snapshots. |
| 555 | |
| 556 | Returns a ``StructuredDelta`` where ``ops`` is a minimal list of |
| 557 | typed operations that transforms ``base`` into ``target``. Plugins |
| 558 | should: |
| 559 | |
| 560 | 1. Compute ops at the finest granularity they can interpret. |
| 561 | 2. Assign meaningful ``content_summary`` strings to each op. |
| 562 | 3. When ``repo_root`` is provided, load sub-file content from the |
| 563 | object store and produce ``PatchOp`` entries with note/element-level |
| 564 | ``child_ops`` instead of coarse ``ReplaceOp`` entries. |
| 565 | 4. Compute a human-readable ``summary`` across all ops. |
| 566 | |
| 567 | The core engine stores this delta alongside the commit record so that |
| 568 | ``muse show`` and ``muse diff`` can display it without reloading blobs. |
| 569 | """ |
| 570 | ... |
| 571 | |
| 572 | def merge( |
| 573 | self, |
| 574 | base: StateSnapshot, |
| 575 | left: StateSnapshot, |
| 576 | right: StateSnapshot, |
| 577 | *, |
| 578 | repo_root: pathlib.Path | None = None, |
| 579 | ) -> MergeResult: |
| 580 | """Three-way merge two divergent state lines against a common base. |
| 581 | |
| 582 | ``base`` is the common ancestor (merge base). ``left`` and ``right`` |
| 583 | are the two divergent snapshots. Returns a ``MergeResult`` with the |
| 584 | reconciled snapshot and any unresolvable conflicts. |
| 585 | |
| 586 | **``.museattributes`` and multidimensional merge contract** — when |
| 587 | *repo_root* is provided, domain plugin implementations should: |
| 588 | |
| 589 | 1. Load ``.museattributes`` via |
| 590 | :func:`muse.core.attributes.load_attributes`. |
| 591 | 2. For each conflicting path, call |
| 592 | :func:`muse.core.attributes.resolve_strategy` with the relevant |
| 593 | dimension name (or ``"*"`` for file-level resolution). |
| 594 | 3. Apply the returned strategy: |
| 595 | |
| 596 | - ``"ours"`` — take the *left* version; remove from conflict list. |
| 597 | - ``"theirs"`` — take the *right* version; remove from conflict list. |
| 598 | - ``"manual"`` — force into conflict list even if the engine would |
| 599 | auto-resolve. |
| 600 | - ``"auto"`` / ``"union"`` — defer to the engine's default logic. |
| 601 | |
| 602 | 4. For domain formats that support true multidimensional content (e.g. |
| 603 | MIDI: notes, pitch_bend, cc_volume, track_structure), attempt |
| 604 | sub-file dimension merge before falling back to a file-level conflict. |
| 605 | """ |
| 606 | ... |
| 607 | |
| 608 | def drift( |
| 609 | self, |
| 610 | committed: StateSnapshot, |
| 611 | live: LiveState, |
| 612 | ) -> DriftReport: |
| 613 | """Compare committed state against current live state. |
| 614 | |
| 615 | Used by ``muse status`` to detect uncommitted changes. Returns a |
| 616 | ``DriftReport`` describing whether the live state has diverged from |
| 617 | the last committed snapshot and, if so, by how much. |
| 618 | """ |
| 619 | ... |
| 620 | |
| 621 | def apply(self, delta: StateDelta, live_state: LiveState) -> LiveState: |
| 622 | """Apply a delta to produce a new live state. |
| 623 | |
| 624 | Used by ``muse checkout`` to reconstruct a historical state. Applies |
| 625 | ``delta`` on top of ``live_state`` and returns the resulting state. |
| 626 | |
| 627 | For ``InsertOp`` and ``ReplaceOp``, the new content is identified by |
| 628 | ``content_id`` (a SHA-256 hash). When ``live_state`` is a |
| 629 | ``pathlib.Path``, the plugin reads the content from the object store. |
| 630 | When ``live_state`` is a ``SnapshotManifest``, only ``DeleteOp`` and |
| 631 | ``ReplaceOp`` at the file level can be applied in-memory. |
| 632 | """ |
| 633 | ... |
| 634 | |
| 635 | def schema(self) -> DomainSchema: |
| 636 | """Declare the structural schema of this domain's state. |
| 637 | |
| 638 | The core engine calls this once at plugin registration time. Plugins |
| 639 | must return a stable, deterministic :class:`~muse.core.schema.DomainSchema` |
| 640 | describing: |
| 641 | |
| 642 | - ``top_level`` — the primary collection structure (e.g. a set of |
| 643 | files, a map of chromosome names to sequences). |
| 644 | - ``dimensions`` — the semantic sub-dimensions of state (e.g. notes, pitch_bend, cc_volume, track_structure for MIDI). |
| 645 | - ``merge_mode`` — ``"three_way"`` (OT merge) or ``"crdt"`` (CRDT convergent join). |
| 646 | |
| 647 | The schema drives :func:`~muse.core.diff_algorithms.diff_by_schema` |
| 648 | algorithm selection and the OT merge engine's conflict detection. |
| 649 | |
| 650 | See :mod:`muse.core.schema` for all available element schema types. |
| 651 | """ |
| 652 | ... |
| 653 | |
| 654 | |
| 655 | # --------------------------------------------------------------------------- |
| 656 | # Operational Transformation optional extension — structured (operation-level) merge |
| 657 | # --------------------------------------------------------------------------- |
| 658 | |
| 659 | |
| 660 | @runtime_checkable |
| 661 | class StructuredMergePlugin(MuseDomainPlugin, Protocol): |
| 662 | """Optional extension for plugins that support operation-level merging. |
| 663 | |
| 664 | Plugins that implement this sub-protocol gain sub-file auto-merge: two |
| 665 | agents inserting notes at non-overlapping bars never produce a conflict, |
| 666 | because the merge engine reasons over ``DomainOp`` trees rather than file |
| 667 | paths. |
| 668 | |
| 669 | The merge engine detects support at runtime via:: |
| 670 | |
| 671 | isinstance(plugin, StructuredMergePlugin) |
| 672 | |
| 673 | Plugins that do not implement ``merge_ops`` fall back to the existing |
| 674 | file-level ``merge()`` path automatically — no changes required. |
| 675 | |
| 676 | The :class:`~muse.plugins.midi.plugin.MidiPlugin` is the reference |
| 677 | implementation for OT-based merge. |
| 678 | """ |
| 679 | |
| 680 | def merge_ops( |
| 681 | self, |
| 682 | base: StateSnapshot, |
| 683 | ours_snap: StateSnapshot, |
| 684 | theirs_snap: StateSnapshot, |
| 685 | ours_ops: list[DomainOp], |
| 686 | theirs_ops: list[DomainOp], |
| 687 | *, |
| 688 | repo_root: pathlib.Path | None = None, |
| 689 | ) -> MergeResult: |
| 690 | """Merge two op lists against a common base using domain knowledge. |
| 691 | |
| 692 | The core merge engine calls this when both branches have produced |
| 693 | ``StructuredDelta`` from ``diff()``. The plugin: |
| 694 | |
| 695 | 1. Calls :func:`muse.core.op_transform.merge_op_lists` to detect |
| 696 | conflicting ``DomainOp`` pairs. |
| 697 | 2. For clean pairs, builds the merged ``SnapshotManifest`` by applying |
| 698 | the adjusted merged ops to *base*. The plugin uses *ours_snap* and |
| 699 | *theirs_snap* to look up the final content IDs for files touched only |
| 700 | by one side (necessary for ``PatchOp`` entries, which do not carry a |
| 701 | ``new_content_id`` directly). |
| 702 | 3. For conflicting pairs, consults ``.museattributes`` (when |
| 703 | *repo_root* is provided) and either auto-resolves via the declared |
| 704 | strategy or adds the address to ``MergeResult.conflicts``. |
| 705 | |
| 706 | Implementations must be domain-aware: a ``.museattributes`` rule of |
| 707 | ``merge=ours`` should take this plugin's understanding of "ours" (the |
| 708 | left branch content), not a raw file-level copy. |
| 709 | |
| 710 | Args: |
| 711 | base: Common ancestor snapshot. |
| 712 | ours_snap: Final snapshot of our branch. |
| 713 | theirs_snap: Final snapshot of their branch. |
| 714 | ours_ops: Operations from our branch delta (base → ours). |
| 715 | theirs_ops: Operations from their branch delta (base → theirs). |
| 716 | repo_root: Repository root for ``.museattributes`` lookup. |
| 717 | |
| 718 | Returns: |
| 719 | A :class:`MergeResult` with the reconciled snapshot and any |
| 720 | remaining unresolvable conflicts. |
| 721 | """ |
| 722 | ... |
| 723 | |
| 724 | |
| 725 | # --------------------------------------------------------------------------- |
| 726 | # CRDT convergent merge — snapshot manifest and CRDTPlugin protocol |
| 727 | # --------------------------------------------------------------------------- |
| 728 | |
| 729 | |
| 730 | class CRDTSnapshotManifest(TypedDict): |
| 731 | """Extended snapshot manifest for CRDT-mode plugins. |
| 732 | |
| 733 | Carries all the fields of a standard snapshot manifest plus CRDT-specific |
| 734 | metadata. The ``files`` mapping has the same semantics as |
| 735 | :class:`SnapshotManifest` — path → content hash. The additional fields |
| 736 | persist CRDT state between commits. |
| 737 | |
| 738 | ``vclock`` records the causal state of the snapshot as a vector clock |
| 739 | ``{agent_id: event_count}``. It is used to detect concurrent writes and |
| 740 | to resolve LWW tiebreaks when two agents write at the same logical time. |
| 741 | |
| 742 | ``crdt_state`` maps per-file-path CRDT state blobs to their SHA-256 hashes |
| 743 | in the object store. CRDT metadata (tombstones, RGA element IDs, OR-Set |
| 744 | tokens) lives here, separate from content hashes, so the content-addressed |
| 745 | store remains valid. |
| 746 | |
| 747 | ``schema_version`` is always ``1``. |
| 748 | """ |
| 749 | |
| 750 | files: dict[str, str] |
| 751 | domain: str |
| 752 | vclock: dict[str, int] |
| 753 | crdt_state: dict[str, str] |
| 754 | schema_version: Literal[1] |
| 755 | |
| 756 | |
| 757 | @runtime_checkable |
| 758 | class CRDTPlugin(MuseDomainPlugin, Protocol): |
| 759 | """Optional extension for plugins that want convergent CRDT merge semantics. |
| 760 | |
| 761 | Plugins implementing this protocol replace the three-way ``merge()`` with |
| 762 | a mathematical ``join()`` on a lattice. ``join`` always succeeds: |
| 763 | |
| 764 | - **No conflict state ever exists.** |
| 765 | - Any two replicas that have received the same set of writes converge to |
| 766 | the same state, regardless of delivery order. |
| 767 | - Millions of agents can write concurrently without coordination. |
| 768 | |
| 769 | The three lattice laws guaranteed by ``join``: |
| 770 | |
| 771 | 1. **Commutativity**: ``join(a, b) == join(b, a)`` |
| 772 | 2. **Associativity**: ``join(join(a, b), c) == join(a, join(b, c))`` |
| 773 | 3. **Idempotency**: ``join(a, a) == a`` |
| 774 | |
| 775 | The core engine detects support at runtime via:: |
| 776 | |
| 777 | isinstance(plugin, CRDTPlugin) |
| 778 | |
| 779 | and routes to ``join`` when ``DomainSchema.merge_mode == "crdt"``. |
| 780 | Plugins that do not implement ``CRDTPlugin`` fall back to the existing |
| 781 | three-way ``merge()`` path. |
| 782 | |
| 783 | Implementation checklist for plugin authors |
| 784 | ------------------------------------------- |
| 785 | 1. Override ``schema()`` to return a :class:`~muse.core.schema.DomainSchema` |
| 786 | with ``merge_mode="crdt"`` and :class:`~muse.core.schema.CRDTDimensionSpec` |
| 787 | for each CRDT dimension. |
| 788 | 2. Implement ``crdt_schema()`` to declare which CRDT primitive maps to each |
| 789 | dimension. |
| 790 | 3. Implement ``join(a, b)`` using the CRDT primitives in |
| 791 | :mod:`muse.core.crdts`. |
| 792 | 4. Implement ``to_crdt_state(snapshot)`` to lift a plain snapshot into |
| 793 | CRDT state. |
| 794 | 5. Implement ``from_crdt_state(crdt)`` to materialise a CRDT state back to |
| 795 | a plain snapshot for ``muse show`` and CLI display. |
| 796 | """ |
| 797 | |
| 798 | def crdt_schema(self) -> list[CRDTDimensionSpec]: |
| 799 | """Declare the CRDT type used for each dimension. |
| 800 | |
| 801 | Returns a list of :class:`~muse.core.schema.CRDTDimensionSpec` — one |
| 802 | per dimension that uses CRDT semantics. Dimensions not listed here |
| 803 | fall back to three-way merge. |
| 804 | |
| 805 | Returns: |
| 806 | List of CRDT dimension declarations. |
| 807 | """ |
| 808 | ... |
| 809 | |
| 810 | def join( |
| 811 | self, |
| 812 | a: CRDTSnapshotManifest, |
| 813 | b: CRDTSnapshotManifest, |
| 814 | ) -> CRDTSnapshotManifest: |
| 815 | """Merge two CRDT snapshots by computing their lattice join. |
| 816 | |
| 817 | This operation is: |
| 818 | |
| 819 | - Commutative: ``join(a, b) == join(b, a)`` |
| 820 | - Associative: ``join(join(a, b), c) == join(a, join(b, c))`` |
| 821 | - Idempotent: ``join(a, a) == a`` |
| 822 | |
| 823 | These three properties guarantee convergence regardless of message |
| 824 | order or delivery count. |
| 825 | |
| 826 | The implementation should use the CRDT primitives in |
| 827 | :mod:`muse.core.crdts` (one primitive per declared CRDT dimension), |
| 828 | compute the per-dimension joins, then rebuild the ``files`` manifest |
| 829 | and ``vclock`` from the results. |
| 830 | |
| 831 | Args: |
| 832 | a: First CRDT snapshot manifest. |
| 833 | b: Second CRDT snapshot manifest. |
| 834 | |
| 835 | Returns: |
| 836 | A new :class:`CRDTSnapshotManifest` that is the join of *a* and *b*. |
| 837 | """ |
| 838 | ... |
| 839 | |
| 840 | def to_crdt_state(self, snapshot: StateSnapshot) -> CRDTSnapshotManifest: |
| 841 | """Lift a plain snapshot into CRDT state representation. |
| 842 | |
| 843 | Called when importing a snapshot that was created before this plugin |
| 844 | opted into CRDT mode. The implementation should initialise fresh CRDT |
| 845 | primitives from the snapshot content, with an empty vector clock. |
| 846 | |
| 847 | Args: |
| 848 | snapshot: A plain :class:`StateSnapshot` to lift. |
| 849 | |
| 850 | Returns: |
| 851 | A :class:`CRDTSnapshotManifest` with the same content and empty |
| 852 | CRDT metadata (zero vector clock, empty ``crdt_state``). |
| 853 | """ |
| 854 | ... |
| 855 | |
| 856 | def from_crdt_state(self, crdt: CRDTSnapshotManifest) -> StateSnapshot: |
| 857 | """Materialise a CRDT state back to a plain snapshot. |
| 858 | |
| 859 | Used by ``muse show``, ``muse status``, and CLI commands that need a |
| 860 | standard :class:`StateSnapshot` view of a CRDT-mode snapshot. |
| 861 | |
| 862 | Args: |
| 863 | crdt: A :class:`CRDTSnapshotManifest` to materialise. |
| 864 | |
| 865 | Returns: |
| 866 | A plain :class:`StateSnapshot` with the visible (non-tombstoned) |
| 867 | content. |
| 868 | """ |
| 869 | ... |