askrobots / dbbasic-object-server
Public
Runs live, versioned Python application objects on one VM, with a suite of apps: projects, notes, tasks, files and more.
danq9
signup: the person chooses the password and name on the confirm page
83c7aa6
2 weeks, 5 days ago
History
test_object_records.py · 124.1 KB · 3064 lines
| 1 | import csv |
| 2 | import json |
| 3 | import random |
| 4 | import unittest.mock |
| 5 | from pathlib import Path |
| 6 | from uuid import UUID |
| 7 | |
| 8 | import pytest |
| 9 | |
| 10 | import object_collections |
| 11 | import object_permissions |
| 12 | import object_record_changes |
| 13 | import object_records |
| 14 | import object_schemas |
| 15 | |
| 16 | |
| 17 | def write_source(path: Path, content: str = "def GET(request):\n return {}\n") -> Path: |
| 18 | path.parent.mkdir(parents=True, exist_ok=True) |
| 19 | path.write_text(content) |
| 20 | return path |
| 21 | |
| 22 | |
| 23 | def write_records(data_dir: Path, collection: str, content: str) -> Path: |
| 24 | path = data_dir / "collections" / collection / "records.tsv" |
| 25 | path.parent.mkdir(parents=True, exist_ok=True) |
| 26 | path.write_text(content) |
| 27 | return path |
| 28 | |
| 29 | |
| 30 | def write_schema(data_dir: Path, collection: str, fields: list[dict] | None = None) -> Path: |
| 31 | path = data_dir / "schemas" / f"{collection}.json" |
| 32 | path.parent.mkdir(parents=True, exist_ok=True) |
| 33 | path.write_text(json.dumps({"fields": fields or [{"name": "id"}]})) |
| 34 | return path |
| 35 | |
| 36 | |
| 37 | def test_list_collection_records_reads_tsv_with_pagination(tmp_path): |
| 38 | data_dir = tmp_path / "data" |
| 39 | write_records( |
| 40 | data_dir, |
| 41 | "contacts", |
| 42 | "id\tfirst_name\tlast_name\n" |
| 43 | "c1\tAda\tLovelace\n" |
| 44 | "c2\tGrace\tHopper\n" |
| 45 | "c3\tKatherine\tJohnson\n", |
| 46 | ) |
| 47 | |
| 48 | result = object_records.list_collection_records( |
| 49 | "contacts", |
| 50 | base_dir=data_dir, |
| 51 | roots=[], |
| 52 | limit=2, |
| 53 | offset=1, |
| 54 | ) |
| 55 | |
| 56 | assert result == { |
| 57 | "collection": "contacts", |
| 58 | "records": [ |
| 59 | {"id": "c2", "first_name": "Grace", "last_name": "Hopper"}, |
| 60 | {"id": "c3", "first_name": "Katherine", "last_name": "Johnson"}, |
| 61 | ], |
| 62 | "count": 2, |
| 63 | "total": 3, |
| 64 | "limit": 2, |
| 65 | "offset": 1, |
| 66 | "has_more": False, |
| 67 | } |
| 68 | |
| 69 | |
| 70 | def test_get_collection_record_returns_record_by_id(tmp_path): |
| 71 | data_dir = tmp_path / "data" |
| 72 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\nc2\tGrace\n") |
| 73 | |
| 74 | record = object_records.get_collection_record("contacts", "c2", base_dir=data_dir, roots=[]) |
| 75 | |
| 76 | assert record == {"id": "c2", "name": "Grace"} |
| 77 | |
| 78 | |
| 79 | def test_create_collection_record_appends_row_and_new_fields(tmp_path): |
| 80 | data_dir = tmp_path / "data" |
| 81 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 82 | |
| 83 | record = object_records.create_collection_record( |
| 84 | "contacts", |
| 85 | {"id": "c2", "name": "Grace", "email": "grace@example.com"}, |
| 86 | base_dir=data_dir, |
| 87 | roots=[], |
| 88 | ) |
| 89 | |
| 90 | assert record == {"id": "c2", "name": "Grace", "email": "grace@example.com"} |
| 91 | assert object_records.list_collection_records("contacts", base_dir=data_dir, roots=[])["records"] == [ |
| 92 | {"id": "c1", "name": "Ada", "email": ""}, |
| 93 | {"id": "c2", "name": "Grace", "email": "grace@example.com"}, |
| 94 | ] |
| 95 | |
| 96 | |
| 97 | def test_create_collection_record_generates_uuid_id_when_missing(tmp_path): |
| 98 | data_dir = tmp_path / "data" |
| 99 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 100 | |
| 101 | record = object_records.create_collection_record( |
| 102 | "contacts", |
| 103 | {"name": "Grace"}, |
| 104 | base_dir=data_dir, |
| 105 | roots=[], |
| 106 | ) |
| 107 | |
| 108 | parsed = UUID(record["id"]) |
| 109 | assert parsed.version == 4 |
| 110 | assert record["name"] == "Grace" |
| 111 | assert object_records.get_collection_record( |
| 112 | "contacts", |
| 113 | record["id"], |
| 114 | base_dir=data_dir, |
| 115 | roots=[], |
| 116 | ) == record |
| 117 | |
| 118 | |
| 119 | def test_create_collection_record_can_start_schema_backed_collection(tmp_path): |
| 120 | data_dir = tmp_path / "data" |
| 121 | write_schema(data_dir, "invoices") |
| 122 | |
| 123 | record = object_records.create_collection_record( |
| 124 | "invoices", |
| 125 | {"id": "i1", "total": 120}, |
| 126 | base_dir=data_dir, |
| 127 | roots=[], |
| 128 | ) |
| 129 | |
| 130 | assert record == {"id": "i1", "total": "120"} |
| 131 | assert (data_dir / "collections" / "invoices" / "records.tsv").read_text() == ( |
| 132 | "id\ttotal\n" |
| 133 | "i1\t120\n" |
| 134 | ) |
| 135 | |
| 136 | |
| 137 | def test_create_collection_record_applies_schema_defaults_and_allows_extra_fields(tmp_path): |
| 138 | data_dir = tmp_path / "data" |
| 139 | write_schema( |
| 140 | data_dir, |
| 141 | "invoices", |
| 142 | [ |
| 143 | {"name": "id"}, |
| 144 | {"name": "invoice_date", "type": "date", "required": True}, |
| 145 | {"name": "status", "type": "enum", "enum": ["draft", "sent"], "default": "draft"}, |
| 146 | {"name": "quantity", "type": "integer", "validation": {"min": 1, "max": 20}}, |
| 147 | {"name": "notes", "validation": {"min_length": 3, "max_length": 20}}, |
| 148 | ], |
| 149 | ) |
| 150 | |
| 151 | record = object_records.create_collection_record( |
| 152 | "invoices", |
| 153 | { |
| 154 | "id": "i1", |
| 155 | "invoice_date": "2026-04-08", |
| 156 | "quantity": 3, |
| 157 | "notes": "Thanks", |
| 158 | "extra_field": "still allowed", |
| 159 | }, |
| 160 | base_dir=data_dir, |
| 161 | roots=[], |
| 162 | ) |
| 163 | |
| 164 | assert record == { |
| 165 | "id": "i1", |
| 166 | "invoice_date": "2026-04-08", |
| 167 | "quantity": "3", |
| 168 | "notes": "Thanks", |
| 169 | "extra_field": "still allowed", |
| 170 | "status": "draft", |
| 171 | } |
| 172 | |
| 173 | |
| 174 | @pytest.mark.parametrize( |
| 175 | ("payload", "message"), |
| 176 | [ |
| 177 | ({"id": "i1", "status": "draft"}, "invoice_date' is required"), |
| 178 | ({"id": "i1", "invoice_date": "bad-date", "status": "draft"}, "invoice_date' must be a date"), |
| 179 | ({"id": "i1", "invoice_date": "2026-04-08", "status": "paid"}, "status' must be one of"), |
| 180 | ({"id": "i1", "invoice_date": "2026-04-08", "quantity": "0"}, "quantity' is below min"), |
| 181 | ({"id": "i1", "invoice_date": "2026-04-08", "quantity": "3.5"}, "quantity' must be an integer"), |
| 182 | ({"id": "i1", "invoice_date": "2026-04-08", "total": "100"}, "total' is computed"), |
| 183 | ], |
| 184 | ) |
| 185 | def test_create_collection_record_validates_schema_fields(tmp_path, payload, message): |
| 186 | data_dir = tmp_path / "data" |
| 187 | write_schema( |
| 188 | data_dir, |
| 189 | "invoices", |
| 190 | [ |
| 191 | {"name": "id"}, |
| 192 | {"name": "invoice_date", "type": "date", "required": True}, |
| 193 | {"name": "status", "type": "enum", "enum": ["draft", "sent"]}, |
| 194 | {"name": "quantity", "type": "integer", "validation": {"min": 1, "max": 20}}, |
| 195 | {"name": "total", "type": "computed", "computed": "sum(line_items)"}, |
| 196 | ], |
| 197 | ) |
| 198 | |
| 199 | with pytest.raises(object_records.InvalidRecordPayloadError, match=message): |
| 200 | object_records.create_collection_record( |
| 201 | "invoices", |
| 202 | payload, |
| 203 | base_dir=data_dir, |
| 204 | roots=[], |
| 205 | ) |
| 206 | |
| 207 | |
| 208 | def test_create_collection_record_rejects_duplicate_id(tmp_path): |
| 209 | data_dir = tmp_path / "data" |
| 210 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 211 | |
| 212 | with pytest.raises(object_records.DuplicateRecordIdError): |
| 213 | object_records.create_collection_record( |
| 214 | "contacts", |
| 215 | {"id": "c1", "name": "Duplicate"}, |
| 216 | base_dir=data_dir, |
| 217 | roots=[], |
| 218 | ) |
| 219 | |
| 220 | |
| 221 | def test_update_collection_record_merges_changes_and_preserves_fields(tmp_path): |
| 222 | data_dir = tmp_path / "data" |
| 223 | write_records(data_dir, "contacts", "id\tname\temail\nc1\tAda\tada@example.com\n") |
| 224 | |
| 225 | record = object_records.update_collection_record( |
| 226 | "contacts", |
| 227 | "c1", |
| 228 | {"name": "Ada Lovelace", "phone": "555-0100"}, |
| 229 | base_dir=data_dir, |
| 230 | roots=[], |
| 231 | ) |
| 232 | |
| 233 | assert record == { |
| 234 | "id": "c1", |
| 235 | "name": "Ada Lovelace", |
| 236 | "email": "ada@example.com", |
| 237 | "phone": "555-0100", |
| 238 | } |
| 239 | assert object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) == record |
| 240 | |
| 241 | |
| 242 | def test_boolean_fields_are_stored_canonically(tmp_path): |
| 243 | data_dir = tmp_path / "data" |
| 244 | write_schema( |
| 245 | data_dir, |
| 246 | "notes", |
| 247 | [ |
| 248 | {"name": "id"}, |
| 249 | {"name": "content"}, |
| 250 | {"name": "is_public", "type": "boolean", "default": "false"}, |
| 251 | ], |
| 252 | ) |
| 253 | write_records(data_dir, "notes", "id\tcontent\tis_public\n") |
| 254 | |
| 255 | for record_id, submitted in (("n1", "True"), ("n2", "YES"), ("n3", "1")): |
| 256 | record = object_records.create_collection_record( |
| 257 | "notes", |
| 258 | {"id": record_id, "content": "x", "is_public": submitted}, |
| 259 | base_dir=data_dir, |
| 260 | roots=[], |
| 261 | ) |
| 262 | assert record["is_public"] == "true" |
| 263 | |
| 264 | defaulted = object_records.create_collection_record( |
| 265 | "notes", {"id": "n4", "content": "x"}, base_dir=data_dir, roots=[] |
| 266 | ) |
| 267 | assert defaulted["is_public"] == "false" |
| 268 | |
| 269 | updated = object_records.update_collection_record( |
| 270 | "notes", "n1", {"is_public": "OFF"}, base_dir=data_dir, roots=[] |
| 271 | ) |
| 272 | assert updated["is_public"] == "false" |
| 273 | |
| 274 | |
| 275 | def test_enum_transitions_are_enforced_on_update(tmp_path): |
| 276 | data_dir = tmp_path / "data" |
| 277 | write_schema( |
| 278 | data_dir, |
| 279 | "tasks", |
| 280 | [ |
| 281 | {"name": "id"}, |
| 282 | {"name": "title"}, |
| 283 | { |
| 284 | "name": "status", |
| 285 | "type": "enum", |
| 286 | "default": "open", |
| 287 | "enum": ["open", "assigned", "done", "cancelled"], |
| 288 | "transitions": { |
| 289 | "open": ["assigned", "cancelled"], |
| 290 | "assigned": ["done", "open", "cancelled"], |
| 291 | }, |
| 292 | }, |
| 293 | ], |
| 294 | ) |
| 295 | write_records(data_dir, "tasks", "id\ttitle\tstatus\nt1\tShip it\topen\n") |
| 296 | |
| 297 | with pytest.raises( |
| 298 | object_records.InvalidRecordPayloadError, |
| 299 | match="cannot move from 'open' to 'done'", |
| 300 | ): |
| 301 | object_records.update_collection_record( |
| 302 | "tasks", "t1", {"status": "done"}, base_dir=data_dir, roots=[] |
| 303 | ) |
| 304 | |
| 305 | record = object_records.update_collection_record( |
| 306 | "tasks", "t1", {"status": "assigned"}, base_dir=data_dir, roots=[] |
| 307 | ) |
| 308 | assert record["status"] == "assigned" |
| 309 | |
| 310 | record = object_records.update_collection_record( |
| 311 | "tasks", "t1", {"status": "done"}, base_dir=data_dir, roots=[] |
| 312 | ) |
| 313 | assert record["status"] == "done" |
| 314 | |
| 315 | # done is not in the transitions map, so it is terminal. |
| 316 | with pytest.raises(object_records.InvalidRecordPayloadError, match="allowed: none"): |
| 317 | object_records.update_collection_record( |
| 318 | "tasks", "t1", {"status": "open"}, base_dir=data_dir, roots=[] |
| 319 | ) |
| 320 | |
| 321 | # Unrelated updates to a record sitting in a terminal state stay fine. |
| 322 | record = object_records.update_collection_record( |
| 323 | "tasks", "t1", {"title": "Shipped"}, base_dir=data_dir, roots=[] |
| 324 | ) |
| 325 | assert record["title"] == "Shipped" |
| 326 | |
| 327 | |
| 328 | def _write_guarded_ticket_schema(data_dir): |
| 329 | write_schema( |
| 330 | data_dir, |
| 331 | "tickets", |
| 332 | [ |
| 333 | {"name": "id"}, |
| 334 | {"name": "owner_id"}, |
| 335 | { |
| 336 | "name": "status", |
| 337 | "type": "enum", |
| 338 | "enum": ["open", "closed", "reopened"], |
| 339 | "transitions": { |
| 340 | "open": ["closed"], |
| 341 | "closed": [{"to": "reopened", "when": {"owner_id": "$user_id"}}], |
| 342 | }, |
| 343 | }, |
| 344 | ], |
| 345 | ) |
| 346 | |
| 347 | |
| 348 | def test_guarded_transition_without_subject_enforces_validity_but_not_guard(tmp_path): |
| 349 | """Direct library callers (daemon, CLI, tests) that don't pass a |
| 350 | transition_subject are trusted callers: a guarded move still has to be |
| 351 | in the list, but the guard's 'when' clause is not enforced.""" |
| 352 | data_dir = tmp_path / "data" |
| 353 | _write_guarded_ticket_schema(data_dir) |
| 354 | write_records(data_dir, "tickets", "id\towner_id\tstatus\nt1\t7\tclosed\n") |
| 355 | |
| 356 | # No subject: the guarded move is valid (present in the list) even |
| 357 | # though the record's owner_id doesn't match anyone in particular. |
| 358 | record = object_records.update_collection_record( |
| 359 | "tickets", "t1", {"status": "reopened"}, base_dir=data_dir, roots=[] |
| 360 | ) |
| 361 | assert record["status"] == "reopened" |
| 362 | |
| 363 | # A move that isn't in the list at all is still rejected outright. |
| 364 | with pytest.raises(object_records.InvalidRecordPayloadError): |
| 365 | object_records.update_collection_record( |
| 366 | "tickets", "t1", {"status": "closed"}, base_dir=data_dir, roots=[] |
| 367 | ) |
| 368 | |
| 369 | |
| 370 | def test_guarded_transition_with_subject_allows_matching_and_denies_others(tmp_path): |
| 371 | data_dir = tmp_path / "data" |
| 372 | _write_guarded_ticket_schema(data_dir) |
| 373 | write_records(data_dir, "tickets", "id\towner_id\tstatus\nt1\t7\tclosed\nt2\t7\tclosed\n") |
| 374 | |
| 375 | owner = object_permissions.PermissionSubject(user_id="7") |
| 376 | other = object_permissions.PermissionSubject(user_id="8") |
| 377 | |
| 378 | with pytest.raises( |
| 379 | object_records.TransitionNotAllowedError, |
| 380 | match="cannot move from 'closed' to 'reopened'", |
| 381 | ): |
| 382 | object_records.update_collection_record( |
| 383 | "tickets", "t1", {"status": "reopened"}, base_dir=data_dir, roots=[], |
| 384 | transition_subject=other, |
| 385 | ) |
| 386 | |
| 387 | record = object_records.update_collection_record( |
| 388 | "tickets", "t2", {"status": "reopened"}, base_dir=data_dir, roots=[], |
| 389 | transition_subject=owner, |
| 390 | ) |
| 391 | assert record["status"] == "reopened" |
| 392 | |
| 393 | # Plain string entries (open -> closed here) stay unguarded regardless |
| 394 | # of the subject passed in. |
| 395 | object_records.create_collection_record( |
| 396 | "tickets", {"id": "t3", "owner_id": "7", "status": "open"}, base_dir=data_dir, roots=[] |
| 397 | ) |
| 398 | record = object_records.update_collection_record( |
| 399 | "tickets", "t3", {"status": "closed"}, base_dir=data_dir, roots=[], |
| 400 | transition_subject=other, |
| 401 | ) |
| 402 | assert record["status"] == "closed" |
| 403 | |
| 404 | |
| 405 | def test_update_collection_record_validates_final_schema_record(tmp_path): |
| 406 | data_dir = tmp_path / "data" |
| 407 | write_schema( |
| 408 | data_dir, |
| 409 | "invoices", |
| 410 | [ |
| 411 | {"name": "id"}, |
| 412 | {"name": "customer_id", "type": "relation", "required": True}, |
| 413 | {"name": "paid", "type": "boolean"}, |
| 414 | {"name": "total", "type": "computed", "computed": "sum(line_items)"}, |
| 415 | ], |
| 416 | ) |
| 417 | write_records(data_dir, "invoices", "id\tcustomer_id\tpaid\n" |
| 418 | "i1\tc1\tfalse\n") |
| 419 | |
| 420 | record = object_records.update_collection_record( |
| 421 | "invoices", |
| 422 | "i1", |
| 423 | {"paid": "true"}, |
| 424 | base_dir=data_dir, |
| 425 | roots=[], |
| 426 | ) |
| 427 | |
| 428 | assert record == {"id": "i1", "customer_id": "c1", "paid": "true"} |
| 429 | with pytest.raises(object_records.InvalidRecordPayloadError, match="customer_id' is required"): |
| 430 | object_records.update_collection_record( |
| 431 | "invoices", |
| 432 | "i1", |
| 433 | {"customer_id": ""}, |
| 434 | base_dir=data_dir, |
| 435 | roots=[], |
| 436 | ) |
| 437 | with pytest.raises(object_records.InvalidRecordPayloadError, match="total' is computed"): |
| 438 | object_records.update_collection_record( |
| 439 | "invoices", |
| 440 | "i1", |
| 441 | {"total": "100"}, |
| 442 | base_dir=data_dir, |
| 443 | roots=[], |
| 444 | ) |
| 445 | |
| 446 | |
| 447 | def test_update_collection_record_rejects_id_change(tmp_path): |
| 448 | data_dir = tmp_path / "data" |
| 449 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 450 | |
| 451 | with pytest.raises(object_records.InvalidRecordPayloadError, match="id cannot be changed"): |
| 452 | object_records.update_collection_record( |
| 453 | "contacts", |
| 454 | "c1", |
| 455 | {"id": "c2", "name": "Grace"}, |
| 456 | base_dir=data_dir, |
| 457 | roots=[], |
| 458 | ) |
| 459 | |
| 460 | |
| 461 | def test_delete_collection_record_removes_row_and_returns_deleted_record(tmp_path): |
| 462 | data_dir = tmp_path / "data" |
| 463 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\nc2\tGrace\n") |
| 464 | |
| 465 | record = object_records.delete_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 466 | |
| 467 | assert record == {"id": "c1", "name": "Ada"} |
| 468 | assert object_records.list_collection_records("contacts", base_dir=data_dir, roots=[])["records"] == [ |
| 469 | {"id": "c2", "name": "Grace"}, |
| 470 | ] |
| 471 | |
| 472 | |
| 473 | def test_record_payload_rejects_unsafe_fields_and_non_scalar_values(): |
| 474 | with pytest.raises(object_records.InvalidRecordPayloadError, match="field name"): |
| 475 | object_records.normalize_record_payload({"bad\tfield": "value"}) |
| 476 | |
| 477 | with pytest.raises(object_records.InvalidRecordPayloadError, match="scalar"): |
| 478 | object_records.normalize_record_payload({"id": "c1", "tags": ["vip"]}) |
| 479 | |
| 480 | |
| 481 | def test_schema_backed_collection_without_records_returns_empty_list(tmp_path): |
| 482 | data_dir = tmp_path / "data" |
| 483 | write_schema(data_dir, "invoices") |
| 484 | |
| 485 | result = object_records.list_collection_records("invoices", base_dir=data_dir, roots=[]) |
| 486 | |
| 487 | assert result == { |
| 488 | "collection": "invoices", |
| 489 | "records": [], |
| 490 | "count": 0, |
| 491 | "total": 0, |
| 492 | "limit": 100, |
| 493 | "offset": 0, |
| 494 | "has_more": False, |
| 495 | } |
| 496 | |
| 497 | |
| 498 | def test_object_backed_collection_without_records_returns_empty_list(tmp_path): |
| 499 | data_dir = tmp_path / "data" |
| 500 | root = tmp_path / "objects" |
| 501 | write_source(root / "contacts" / "directory.py") |
| 502 | |
| 503 | result = object_records.list_collection_records("contacts", base_dir=data_dir, roots=[root]) |
| 504 | |
| 505 | assert result["records"] == [] |
| 506 | assert result["total"] == 0 |
| 507 | |
| 508 | |
| 509 | def test_get_collection_record_rejects_missing_record(tmp_path): |
| 510 | data_dir = tmp_path / "data" |
| 511 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 512 | |
| 513 | with pytest.raises(object_records.RecordNotFoundError): |
| 514 | object_records.get_collection_record("contacts", "missing", base_dir=data_dir, roots=[]) |
| 515 | |
| 516 | |
| 517 | def test_list_collection_records_rejects_missing_collection(tmp_path): |
| 518 | with pytest.raises(object_collections.CollectionNotFoundError): |
| 519 | object_records.list_collection_records("missing", base_dir=tmp_path / "data", roots=[]) |
| 520 | |
| 521 | |
| 522 | def test_collection_records_reject_unsafe_names(tmp_path): |
| 523 | with pytest.raises(object_collections.InvalidCollectionNameError): |
| 524 | object_records.list_collection_records("../bad", base_dir=tmp_path / "data", roots=[]) |
| 525 | |
| 526 | data_dir = tmp_path / "data" |
| 527 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 528 | with pytest.raises(object_records.InvalidRecordIdError): |
| 529 | object_records.get_collection_record("contacts", "../bad", base_dir=data_dir, roots=[]) |
| 530 | |
| 531 | |
| 532 | def test_records_file_requires_id_column(tmp_path): |
| 533 | data_dir = tmp_path / "data" |
| 534 | write_records(data_dir, "contacts", "name\nAda\n") |
| 535 | |
| 536 | with pytest.raises(ValueError, match="id column"): |
| 537 | object_records.list_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 538 | |
| 539 | |
| 540 | def test_records_file_rejects_duplicate_fields(tmp_path): |
| 541 | data_dir = tmp_path / "data" |
| 542 | write_records(data_dir, "contacts", "id\tname\tname\nc1\tAda\tLovelace\n") |
| 543 | |
| 544 | with pytest.raises(ValueError, match="duplicate"): |
| 545 | object_records.list_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 546 | |
| 547 | |
| 548 | def test_records_file_rejects_extra_row_fields(tmp_path): |
| 549 | data_dir = tmp_path / "data" |
| 550 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\textra\n") |
| 551 | |
| 552 | with pytest.raises(ValueError, match="extra fields"): |
| 553 | object_records.list_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 554 | |
| 555 | |
| 556 | def test_iter_record_collections_lists_safe_record_directories(tmp_path): |
| 557 | data_dir = tmp_path / "data" |
| 558 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 559 | write_records(data_dir, "bad-name", "id\tname\nbad\tBad\n") |
| 560 | |
| 561 | assert object_records.iter_record_collections(data_dir) == ["contacts"] |
| 562 | |
| 563 | |
| 564 | def test_relation_fields_require_existing_target_records(tmp_path): |
| 565 | write_records(tmp_path, "projects", "id\tname\np1\tWebsite\n") |
| 566 | schema_file = tmp_path / "schemas" / "tasks.json" |
| 567 | schema_file.parent.mkdir(parents=True, exist_ok=True) |
| 568 | schema_file.write_text( |
| 569 | json.dumps( |
| 570 | { |
| 571 | "fields": [ |
| 572 | {"name": "id"}, |
| 573 | {"name": "title", "required": True}, |
| 574 | {"name": "project_id", "relation": {"collection": "projects", "display_field": "name"}}, |
| 575 | ] |
| 576 | } |
| 577 | ) |
| 578 | ) |
| 579 | |
| 580 | created = object_records.create_collection_record( |
| 581 | "tasks", |
| 582 | {"id": "t1", "title": "Launch", "project_id": "p1"}, |
| 583 | base_dir=tmp_path, |
| 584 | ) |
| 585 | assert created["project_id"] == "p1" |
| 586 | |
| 587 | with pytest.raises(object_records.InvalidRecordPayloadError) as missing: |
| 588 | object_records.create_collection_record( |
| 589 | "tasks", |
| 590 | {"id": "t2", "title": "Broken", "project_id": "p999"}, |
| 591 | base_dir=tmp_path, |
| 592 | ) |
| 593 | assert "missing record: projects/p999" in str(missing.value) |
| 594 | |
| 595 | # Empty relation values stay allowed unless the field is required. |
| 596 | optional = object_records.create_collection_record( |
| 597 | "tasks", |
| 598 | {"id": "t3", "title": "No project"}, |
| 599 | base_dir=tmp_path, |
| 600 | ) |
| 601 | assert optional["project_id"] == "" |
| 602 | |
| 603 | |
| 604 | def test_relation_accepts_string_shorthand_and_rejects_bad_shapes(tmp_path): |
| 605 | write_records(tmp_path, "projects", "id\tname\np1\tWebsite\n") |
| 606 | schema_file = tmp_path / "schemas" / "tasks.json" |
| 607 | schema_file.parent.mkdir(parents=True, exist_ok=True) |
| 608 | schema_file.write_text( |
| 609 | json.dumps({"fields": [{"name": "id"}, {"name": "project_id", "relation": "projects"}]}) |
| 610 | ) |
| 611 | |
| 612 | created = object_records.create_collection_record( |
| 613 | "tasks", {"id": "t1", "project_id": "p1"}, base_dir=tmp_path |
| 614 | ) |
| 615 | assert created["project_id"] == "p1" |
| 616 | |
| 617 | schema_file.write_text( |
| 618 | json.dumps({"fields": [{"name": "id"}, {"name": "project_id", "relation": 42}]}) |
| 619 | ) |
| 620 | with pytest.raises(object_records.InvalidRecordPayloadError): |
| 621 | object_records.create_collection_record( |
| 622 | "tasks", {"id": "t2", "project_id": "p1"}, base_dir=tmp_path |
| 623 | ) |
| 624 | |
| 625 | |
| 626 | def test_created_at_is_server_set_on_create(tmp_path): |
| 627 | data_dir = tmp_path / "data" |
| 628 | write_schema( |
| 629 | data_dir, |
| 630 | "links", |
| 631 | [ |
| 632 | {"name": "id"}, |
| 633 | {"name": "title", "required": True}, |
| 634 | {"name": "created_at", "type": "datetime", "read_only": True}, |
| 635 | ], |
| 636 | ) |
| 637 | write_records(data_dir, "links", "id\ttitle\tcreated_at\n") |
| 638 | |
| 639 | # client omits created_at -> server fills it, read_only is not violated |
| 640 | rec = object_records.create_collection_record( |
| 641 | "links", {"id": "l1", "title": "x"}, base_dir=data_dir, roots=[] |
| 642 | ) |
| 643 | assert rec["created_at"].endswith("Z") and "T" in rec["created_at"] |
| 644 | |
| 645 | # client cannot spoof a read_only created_at |
| 646 | import pytest |
| 647 | with pytest.raises(object_records.InvalidRecordPayloadError, match="read-only"): |
| 648 | object_records.create_collection_record( |
| 649 | "links", {"id": "l2", "title": "y", "created_at": "2000-01-01T00:00:00Z"}, |
| 650 | base_dir=data_dir, roots=[], |
| 651 | ) |
| 652 | |
| 653 | |
| 654 | # --- Phase 4b: the `extra` extension field (docs/upgrade-and-customization.md Rule 3) --- |
| 655 | |
| 656 | |
| 657 | def test_backward_compat_no_store_extra_fields_means_no_extra_column(tmp_path): |
| 658 | """A collection with no store:extra fields and no submitted `extra` key |
| 659 | must behave byte-identically to before the `extra` feature existed: no |
| 660 | `extra` column ever appears on disk.""" |
| 661 | data_dir = tmp_path / "data" |
| 662 | write_schema( |
| 663 | data_dir, |
| 664 | "contacts", |
| 665 | [{"name": "id"}, {"name": "name", "required": True}, {"name": "email"}], |
| 666 | ) |
| 667 | |
| 668 | created = object_records.create_collection_record( |
| 669 | "contacts", |
| 670 | {"id": "c1", "name": "Ada", "email": "ada@example.com"}, |
| 671 | base_dir=data_dir, |
| 672 | roots=[], |
| 673 | ) |
| 674 | assert created == {"id": "c1", "name": "Ada", "email": "ada@example.com"} |
| 675 | |
| 676 | updated = object_records.update_collection_record( |
| 677 | "contacts", "c1", {"email": "ada@example.org"}, base_dir=data_dir, roots=[] |
| 678 | ) |
| 679 | assert updated == {"id": "c1", "name": "Ada", "email": "ada@example.org"} |
| 680 | |
| 681 | fetched = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 682 | assert fetched == updated |
| 683 | |
| 684 | listed = object_records.list_collection_records("contacts", base_dir=data_dir, roots=[])["records"] |
| 685 | assert listed == [updated] |
| 686 | |
| 687 | tsv_text = (data_dir / "collections" / "contacts" / "records.tsv").read_text() |
| 688 | assert tsv_text == "id\tname\temail\nc1\tAda\tada@example.org\n" |
| 689 | assert "extra" not in tsv_text.splitlines()[0].split("\t") |
| 690 | |
| 691 | |
| 692 | def test_store_extra_view_field_is_stored_in_extra_column_not_own_column(tmp_path): |
| 693 | data_dir = tmp_path / "data" |
| 694 | write_schema( |
| 695 | data_dir, |
| 696 | "notes", |
| 697 | [ |
| 698 | {"name": "id"}, |
| 699 | {"name": "title", "required": True}, |
| 700 | {"name": "x_mood", "type": "text", "store": "extra"}, |
| 701 | ], |
| 702 | ) |
| 703 | |
| 704 | record = object_records.create_collection_record( |
| 705 | "notes", |
| 706 | {"id": "n1", "title": "Hi", "x_mood": "curious"}, |
| 707 | base_dir=data_dir, |
| 708 | roots=[], |
| 709 | ) |
| 710 | |
| 711 | assert record["title"] == "Hi" |
| 712 | assert record["x_mood"] == "curious" |
| 713 | |
| 714 | path = data_dir / "collections" / "notes" / "records.tsv" |
| 715 | with path.open(newline="") as handle: |
| 716 | rows = list(csv.reader(handle, delimiter="\t")) |
| 717 | header = rows[0] |
| 718 | assert "extra" in header |
| 719 | assert "x_mood" not in header |
| 720 | |
| 721 | extra_cell = rows[1][header.index("extra")] |
| 722 | assert json.loads(extra_cell) == {"x_mood": "curious"} |
| 723 | |
| 724 | fetched = object_records.get_collection_record("notes", "n1", base_dir=data_dir, roots=[]) |
| 725 | assert fetched["x_mood"] == "curious" |
| 726 | assert json.loads(fetched["extra"]) == {"x_mood": "curious"} |
| 727 | |
| 728 | listed = object_records.list_collection_records("notes", base_dir=data_dir, roots=[])["records"] |
| 729 | assert listed[0]["x_mood"] == "curious" |
| 730 | |
| 731 | |
| 732 | def test_store_extra_view_field_still_validates(tmp_path): |
| 733 | data_dir = tmp_path / "data" |
| 734 | write_schema( |
| 735 | data_dir, |
| 736 | "notes", |
| 737 | [ |
| 738 | {"name": "id"}, |
| 739 | {"name": "title", "required": True}, |
| 740 | { |
| 741 | "name": "x_mood", |
| 742 | "type": "enum", |
| 743 | "enum": ["happy", "curious", "grumpy"], |
| 744 | "required": True, |
| 745 | "store": "extra", |
| 746 | }, |
| 747 | ], |
| 748 | ) |
| 749 | |
| 750 | with pytest.raises(object_records.InvalidRecordPayloadError, match="x_mood' is required"): |
| 751 | object_records.create_collection_record( |
| 752 | "notes", {"id": "n1", "title": "Hi"}, base_dir=data_dir, roots=[] |
| 753 | ) |
| 754 | |
| 755 | with pytest.raises(object_records.InvalidRecordPayloadError, match="x_mood' must be one of"): |
| 756 | object_records.create_collection_record( |
| 757 | "notes", |
| 758 | {"id": "n1", "title": "Hi", "x_mood": "furious"}, |
| 759 | base_dir=data_dir, |
| 760 | roots=[], |
| 761 | ) |
| 762 | |
| 763 | record = object_records.create_collection_record( |
| 764 | "notes", |
| 765 | {"id": "n1", "title": "Hi", "x_mood": "curious"}, |
| 766 | base_dir=data_dir, |
| 767 | roots=[], |
| 768 | ) |
| 769 | assert record["x_mood"] == "curious" |
| 770 | |
| 771 | |
| 772 | def test_partial_update_preserves_untouched_extra_keys(tmp_path): |
| 773 | data_dir = tmp_path / "data" |
| 774 | write_schema( |
| 775 | data_dir, |
| 776 | "notes", |
| 777 | [ |
| 778 | {"name": "id"}, |
| 779 | {"name": "title"}, |
| 780 | {"name": "x_mood", "type": "text", "store": "extra"}, |
| 781 | ], |
| 782 | ) |
| 783 | |
| 784 | created = object_records.create_collection_record( |
| 785 | "notes", |
| 786 | {"id": "n1", "title": "Hi", "x_mood": "curious", "extra": {"note": "hi"}}, |
| 787 | base_dir=data_dir, |
| 788 | roots=[], |
| 789 | ) |
| 790 | assert created["x_mood"] == "curious" |
| 791 | assert json.loads(created["extra"]) == {"x_mood": "curious", "note": "hi"} |
| 792 | |
| 793 | # Updating only `title` must not wipe x_mood or the undeclared `note` key. |
| 794 | after_title = object_records.update_collection_record( |
| 795 | "notes", "n1", {"title": "Hello"}, base_dir=data_dir, roots=[] |
| 796 | ) |
| 797 | assert after_title["title"] == "Hello" |
| 798 | assert after_title["x_mood"] == "curious" |
| 799 | assert json.loads(after_title["extra"]) == {"x_mood": "curious", "note": "hi"} |
| 800 | |
| 801 | # Updating only x_mood must not wipe `note`, and must not touch title. |
| 802 | after_mood = object_records.update_collection_record( |
| 803 | "notes", "n1", {"x_mood": "grumpy"}, base_dir=data_dir, roots=[] |
| 804 | ) |
| 805 | assert after_mood["title"] == "Hello" |
| 806 | assert after_mood["x_mood"] == "grumpy" |
| 807 | assert json.loads(after_mood["extra"]) == {"x_mood": "grumpy", "note": "hi"} |
| 808 | |
| 809 | fetched = object_records.get_collection_record("notes", "n1", base_dir=data_dir, roots=[]) |
| 810 | assert fetched["x_mood"] == "grumpy" |
| 811 | assert json.loads(fetched["extra"]) == {"x_mood": "grumpy", "note": "hi"} |
| 812 | |
| 813 | |
| 814 | def test_raw_extra_passthrough_accepts_dict_and_json_string(tmp_path): |
| 815 | data_dir = tmp_path / "data" |
| 816 | write_records(data_dir, "widgets", "id\tname\n") |
| 817 | |
| 818 | from_dict = object_records.create_collection_record( |
| 819 | "widgets", |
| 820 | {"id": "w1", "name": "Gadget", "extra": {"a": "1"}}, |
| 821 | base_dir=data_dir, |
| 822 | roots=[], |
| 823 | ) |
| 824 | assert json.loads(from_dict["extra"]) == {"a": "1"} |
| 825 | |
| 826 | from_string = object_records.create_collection_record( |
| 827 | "widgets", |
| 828 | {"id": "w2", "name": "Gizmo", "extra": json.dumps({"a": "1"})}, |
| 829 | base_dir=data_dir, |
| 830 | roots=[], |
| 831 | ) |
| 832 | assert json.loads(from_string["extra"]) == {"a": "1"} |
| 833 | |
| 834 | fetched = object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 835 | assert json.loads(fetched["extra"]) == {"a": "1"} |
| 836 | |
| 837 | |
| 838 | @pytest.mark.parametrize("bad_extra", [["a", "1"], "[1, 2]", "not json", 5, True]) |
| 839 | def test_invalid_extra_payload_is_rejected(tmp_path, bad_extra): |
| 840 | data_dir = tmp_path / "data" |
| 841 | write_records(data_dir, "widgets", "id\tname\n") |
| 842 | |
| 843 | with pytest.raises(object_records.InvalidRecordPayloadError, match="extra"): |
| 844 | object_records.create_collection_record( |
| 845 | "widgets", |
| 846 | {"id": "w1", "name": "Gadget", "extra": bad_extra}, |
| 847 | base_dir=data_dir, |
| 848 | roots=[], |
| 849 | ) |
| 850 | |
| 851 | |
| 852 | # --- Perf pass: records cache, id index, csv.reader-based parsing --- |
| 853 | # |
| 854 | # These tests describe behavior that must hold both BEFORE and AFTER the |
| 855 | # caching/parsing change (there is no cache pre-change, so "a mutation |
| 856 | # never leaks" and "external changes are always seen" hold trivially; the |
| 857 | # point is that they must ALSO hold once a cache is introduced). |
| 858 | |
| 859 | |
| 860 | def test_short_row_missing_fields_become_empty_string(tmp_path): |
| 861 | """A data row with fewer tab-separated values than the header (e.g. a |
| 862 | hand-edited or partially-written TSV) must fill the missing trailing |
| 863 | fields with "" -- this mirrors csv.DictReader's restval=None behavior, |
| 864 | which the surrounding code already converts to "" (see the |
| 865 | `value if value is not None else ""` normalization).""" |
| 866 | data_dir = tmp_path / "data" |
| 867 | write_records(data_dir, "contacts", "id\tname\temail\tphone\nc1\tAda\n") |
| 868 | |
| 869 | result = object_records.list_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 870 | assert result["records"] == [{"id": "c1", "name": "Ada", "email": "", "phone": ""}] |
| 871 | |
| 872 | record = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 873 | assert record == {"id": "c1", "name": "Ada", "email": "", "phone": ""} |
| 874 | |
| 875 | |
| 876 | def test_long_row_extra_fields_raise_matching_original_error(tmp_path): |
| 877 | data_dir = tmp_path / "data" |
| 878 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\tLovelace\textra\n") |
| 879 | |
| 880 | with pytest.raises(ValueError, match="extra fields on row 2"): |
| 881 | object_records.list_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 882 | |
| 883 | |
| 884 | def test_quoted_value_containing_tab_round_trips(tmp_path): |
| 885 | """A field value containing a literal tab (and newline) is legitimate -- |
| 886 | only field NAMES/ids are restricted, not values (see |
| 887 | _normalize_record_payload). csv.DictWriter's default QUOTE_MINIMAL |
| 888 | quotes such a value on write; the read path must undo that quoting |
| 889 | exactly, never naively split on tab.""" |
| 890 | data_dir = tmp_path / "data" |
| 891 | write_records(data_dir, "notes", "id\tbody\n") |
| 892 | |
| 893 | tricky_value = "line one\tline two\nline three" |
| 894 | created = object_records.create_collection_record( |
| 895 | "notes", |
| 896 | {"id": "n1", "body": tricky_value}, |
| 897 | base_dir=data_dir, |
| 898 | roots=[], |
| 899 | ) |
| 900 | assert created["body"] == tricky_value |
| 901 | |
| 902 | fetched = object_records.get_collection_record("notes", "n1", base_dir=data_dir, roots=[]) |
| 903 | assert fetched["body"] == tricky_value |
| 904 | |
| 905 | listed = object_records.list_collection_records("notes", base_dir=data_dir, roots=[])["records"] |
| 906 | assert listed[0]["body"] == tricky_value |
| 907 | |
| 908 | |
| 909 | def test_get_collection_record_with_duplicate_ids_returns_first_match(tmp_path): |
| 910 | """Malformed/hand-edited data can contain a duplicate id. Any id index |
| 911 | used for O(1) lookup must replicate the original linear scan's |
| 912 | first-match-wins semantics exactly.""" |
| 913 | data_dir = tmp_path / "data" |
| 914 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\nc1\tGrace\n") |
| 915 | |
| 916 | record = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 917 | assert record == {"id": "c1", "name": "Ada"} |
| 918 | |
| 919 | |
| 920 | def test_mutating_returned_record_does_not_corrupt_next_read(tmp_path): |
| 921 | """Every record dict/list handed back across the public API must be a |
| 922 | fresh copy: a caller mutating what it got back must never affect a |
| 923 | later, independent call. This is the aliasing guard for the records |
| 924 | cache -- cached objects must never be returned by reference.""" |
| 925 | data_dir = tmp_path / "data" |
| 926 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\nc2\tGrace\n") |
| 927 | |
| 928 | first = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 929 | first["name"] = "CORRUPTED" |
| 930 | first["new_field"] = "also corrupted" |
| 931 | |
| 932 | second = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 933 | assert second == {"id": "c1", "name": "Ada"} |
| 934 | |
| 935 | records = object_records.list_collection_records("contacts", base_dir=data_dir, roots=[])["records"] |
| 936 | records[0]["name"] = "ALSO CORRUPTED" |
| 937 | |
| 938 | still_clean = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 939 | assert still_clean["name"] == "Ada" |
| 940 | still_listed = object_records.list_collection_records("contacts", base_dir=data_dir, roots=[])["records"] |
| 941 | assert still_listed[0]["name"] == "Ada" |
| 942 | |
| 943 | |
| 944 | def test_cache_reflects_external_file_change(tmp_path): |
| 945 | """Another process (or the same process writing directly, bypassing the |
| 946 | API) can rewrite records.tsv on disk. The next API read must observe |
| 947 | the new content -- the stat-signature check the records cache relies |
| 948 | on for cross-process consistency.""" |
| 949 | data_dir = tmp_path / "data" |
| 950 | path = write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 951 | |
| 952 | warm = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 953 | assert warm["name"] == "Ada" |
| 954 | |
| 955 | # Rewrite the file directly, outside the object_records API. |
| 956 | path.write_text("id\tname\nc1\tAda Lovelace\nc2\tGrace Hopper\n") |
| 957 | |
| 958 | refreshed = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 959 | assert refreshed["name"] == "Ada Lovelace" |
| 960 | |
| 961 | listed = object_records.list_collection_records("contacts", base_dir=data_dir, roots=[])["records"] |
| 962 | assert listed == [ |
| 963 | {"id": "c1", "name": "Ada Lovelace"}, |
| 964 | {"id": "c2", "name": "Grace Hopper"}, |
| 965 | ] |
| 966 | |
| 967 | |
| 968 | def test_update_after_cached_read_returns_fresh_data(tmp_path): |
| 969 | """A get/list right after a create/update must never return a value |
| 970 | that predates the write, whether or not an earlier read happened to |
| 971 | warm the cache first.""" |
| 972 | data_dir = tmp_path / "data" |
| 973 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 974 | |
| 975 | warm = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 976 | assert warm["name"] == "Ada" |
| 977 | |
| 978 | object_records.update_collection_record( |
| 979 | "contacts", "c1", {"name": "Ada Lovelace"}, base_dir=data_dir, roots=[] |
| 980 | ) |
| 981 | |
| 982 | refreshed = object_records.get_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 983 | assert refreshed["name"] == "Ada Lovelace" |
| 984 | |
| 985 | listed = object_records.list_collection_records("contacts", base_dir=data_dir, roots=[])["records"] |
| 986 | assert listed == [{"id": "c1", "name": "Ada Lovelace"}] |
| 987 | |
| 988 | |
| 989 | def test_classic_write_keeps_an_open_readers_fd_on_a_complete_version(tmp_path): |
| 990 | """The "Q2 property" that live cross-process editing depends on, pinned |
| 991 | by a test -- a code comment (object_records.py, the CONCURRENCY block) |
| 992 | can't fail CI. Classic create/update/delete each rewrite via a temp file |
| 993 | + atomic replace (_write_collection_records), so a reader in another |
| 994 | process that already has the file open keeps seeing the COMPLETE |
| 995 | pre-write version -- the atomic rename swaps the directory entry to a new |
| 996 | inode while the reader's fd holds the old one -- and a reader that opens |
| 997 | AFTER the write sees the COMPLETE post-write version. Never a torn mix. |
| 998 | |
| 999 | This is load-bearing for any consumer that edits a collection live from |
| 1000 | more than one process at once -- e.g. a request handler and a background |
| 1001 | daemon mutating the same objects while clients are actively reading them. |
| 1002 | The regression this guards against is a future "optimization" that |
| 1003 | rewrites in place via open(path, "w"): that truncates the very inode a |
| 1004 | concurrent reader's fd points at, turning live editing into live |
| 1005 | corruption. Such a writer fails the mid-write assertions below. |
| 1006 | |
| 1007 | (Append mode deliberately mutates the same inode and upholds the same |
| 1008 | guarantee via the torn-tail rule instead -- see |
| 1009 | test_append_mode_torn_tail_is_ignored_and_self_heals.) |
| 1010 | """ |
| 1011 | data_dir = tmp_path / "data" |
| 1012 | path = write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 1013 | |
| 1014 | # UPDATE: a reader fd opened before the write still reads the whole |
| 1015 | # pre-write file; a fresh read afterward sees the whole post-write file. |
| 1016 | with path.open("rb") as held: |
| 1017 | object_records.update_collection_record( |
| 1018 | "contacts", "c1", {"name": "Ada Lovelace"}, base_dir=data_dir, roots=[] |
| 1019 | ) |
| 1020 | assert held.read() == b"id\tname\nc1\tAda\n" |
| 1021 | assert path.read_text() == "id\tname\nc1\tAda Lovelace\n" |
| 1022 | |
| 1023 | # CREATE |
| 1024 | with path.open("rb") as held: |
| 1025 | object_records.create_collection_record( |
| 1026 | "contacts", {"id": "c2", "name": "Grace"}, base_dir=data_dir, roots=[] |
| 1027 | ) |
| 1028 | assert held.read() == b"id\tname\nc1\tAda Lovelace\n" |
| 1029 | assert path.read_text() == "id\tname\nc1\tAda Lovelace\nc2\tGrace\n" |
| 1030 | |
| 1031 | # DELETE |
| 1032 | with path.open("rb") as held: |
| 1033 | object_records.delete_collection_record("contacts", "c2", base_dir=data_dir, roots=[]) |
| 1034 | assert held.read() == b"id\tname\nc1\tAda Lovelace\nc2\tGrace\n" |
| 1035 | assert path.read_text() == "id\tname\nc1\tAda Lovelace\n" |
| 1036 | |
| 1037 | |
| 1038 | def test_delete_collection_record_unaffected_by_extra(tmp_path): |
| 1039 | data_dir = tmp_path / "data" |
| 1040 | write_schema( |
| 1041 | data_dir, |
| 1042 | "notes", |
| 1043 | [ |
| 1044 | {"name": "id"}, |
| 1045 | {"name": "title"}, |
| 1046 | {"name": "x_mood", "type": "text", "store": "extra"}, |
| 1047 | ], |
| 1048 | ) |
| 1049 | object_records.create_collection_record( |
| 1050 | "notes", |
| 1051 | {"id": "n1", "title": "Hi", "x_mood": "curious"}, |
| 1052 | base_dir=data_dir, |
| 1053 | roots=[], |
| 1054 | ) |
| 1055 | |
| 1056 | removed = object_records.delete_collection_record("notes", "n1", base_dir=data_dir, roots=[]) |
| 1057 | assert removed["id"] == "n1" |
| 1058 | |
| 1059 | assert object_records.list_collection_records("notes", base_dir=data_dir, roots=[])["records"] == [] |
| 1060 | |
| 1061 | |
| 1062 | def _cache_key(collection: str, data_dir: Path) -> str: |
| 1063 | path = object_records.collection_records_file(collection, base_dir=data_dir) |
| 1064 | return str(path.resolve(strict=False)) |
| 1065 | |
| 1066 | |
| 1067 | def test_records_cache_lru_evicts_least_recently_used(tmp_path, monkeypatch): |
| 1068 | """With capacity 2, reading a third collection must evict the least |
| 1069 | recently touched one (not necessarily the first-read one), and |
| 1070 | re-reading the evicted collection must still return correct data via |
| 1071 | the reparse path.""" |
| 1072 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ENTRIES", "2") |
| 1073 | data_dir = tmp_path / "data" |
| 1074 | write_records(data_dir, "coll_a", "id\tname\na1\tAda\n") |
| 1075 | write_records(data_dir, "coll_b", "id\tname\nb1\tBob\n") |
| 1076 | write_records(data_dir, "coll_c", "id\tname\nc1\tCleo\n") |
| 1077 | |
| 1078 | object_records._RECORDS_CACHE.clear() |
| 1079 | object_records.read_collection_records("coll_a", base_dir=data_dir, roots=[]) |
| 1080 | object_records.read_collection_records("coll_b", base_dir=data_dir, roots=[]) |
| 1081 | # coll_a is now the least-recently-used of {coll_a, coll_b}; reading |
| 1082 | # coll_c should push the cache over capacity and evict coll_a. |
| 1083 | object_records.read_collection_records("coll_c", base_dir=data_dir, roots=[]) |
| 1084 | |
| 1085 | keys = set(object_records._RECORDS_CACHE.keys()) |
| 1086 | assert keys == {_cache_key("coll_b", data_dir), _cache_key("coll_c", data_dir)} |
| 1087 | assert _cache_key("coll_a", data_dir) not in object_records._RECORDS_CACHE |
| 1088 | |
| 1089 | # Re-reading the evicted collection must still work correctly. |
| 1090 | result = object_records.read_collection_records("coll_a", base_dir=data_dir, roots=[]) |
| 1091 | assert result == [{"id": "a1", "name": "Ada"}] |
| 1092 | assert _cache_key("coll_a", data_dir) in object_records._RECORDS_CACHE |
| 1093 | |
| 1094 | |
| 1095 | def test_records_cache_lru_hit_refreshes_recency(tmp_path, monkeypatch): |
| 1096 | """A cache HIT (not just a store) must count as use: touching coll_a |
| 1097 | again before reading coll_c should save it from eviction.""" |
| 1098 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ENTRIES", "2") |
| 1099 | data_dir = tmp_path / "data" |
| 1100 | write_records(data_dir, "coll_a", "id\tname\na1\tAda\n") |
| 1101 | write_records(data_dir, "coll_b", "id\tname\nb1\tBob\n") |
| 1102 | write_records(data_dir, "coll_c", "id\tname\nc1\tCleo\n") |
| 1103 | |
| 1104 | object_records._RECORDS_CACHE.clear() |
| 1105 | object_records.read_collection_records("coll_a", base_dir=data_dir, roots=[]) |
| 1106 | object_records.read_collection_records("coll_b", base_dir=data_dir, roots=[]) |
| 1107 | # Touch coll_a again -- it is now the most-recently-used, so coll_b |
| 1108 | # (untouched since its initial read) should be evicted instead. |
| 1109 | object_records.read_collection_records("coll_a", base_dir=data_dir, roots=[]) |
| 1110 | object_records.read_collection_records("coll_c", base_dir=data_dir, roots=[]) |
| 1111 | |
| 1112 | keys = set(object_records._RECORDS_CACHE.keys()) |
| 1113 | assert keys == {_cache_key("coll_a", data_dir), _cache_key("coll_c", data_dir)} |
| 1114 | |
| 1115 | |
| 1116 | def test_records_cache_row_threshold_excludes_large_collections(tmp_path, monkeypatch): |
| 1117 | """A collection parsing to more rows than DBBASIC_RECORDS_CACHE_MAX_ROWS |
| 1118 | is returned correctly but never stored -- and must not evict a |
| 1119 | small, hot collection just because it was read once.""" |
| 1120 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "10") |
| 1121 | data_dir = tmp_path / "data" |
| 1122 | small_rows = "".join(f"s{i}\tName{i}\n" for i in range(5)) |
| 1123 | write_records(data_dir, "small", "id\tname\n" + small_rows) |
| 1124 | large_rows = "".join(f"l{i}\tName{i}\n" for i in range(20)) |
| 1125 | write_records(data_dir, "large", "id\tname\n" + large_rows) |
| 1126 | |
| 1127 | object_records._RECORDS_CACHE.clear() |
| 1128 | |
| 1129 | small_result = object_records.read_collection_records("small", base_dir=data_dir, roots=[]) |
| 1130 | assert len(small_result) == 5 |
| 1131 | assert _cache_key("small", data_dir) in object_records._RECORDS_CACHE |
| 1132 | |
| 1133 | large_result = object_records.read_collection_records("large", base_dir=data_dir, roots=[]) |
| 1134 | assert len(large_result) == 20 |
| 1135 | assert large_result[0]["id"] == "l0" |
| 1136 | assert large_result[-1]["id"] == "l19" |
| 1137 | |
| 1138 | assert _cache_key("large", data_dir) not in object_records._RECORDS_CACHE |
| 1139 | # The small collection's entry must have survived reading the large one. |
| 1140 | assert _cache_key("small", data_dir) in object_records._RECORDS_CACHE |
| 1141 | |
| 1142 | |
| 1143 | def test_records_cache_row_threshold_reparses_correctly_every_time(tmp_path, monkeypatch): |
| 1144 | """A collection over the row threshold is never cached, so repeated |
| 1145 | reads (and a read after an update) must each reparse from disk and |
| 1146 | return correct, current data -- not a stale first result.""" |
| 1147 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "10") |
| 1148 | data_dir = tmp_path / "data" |
| 1149 | large_rows = "".join(f"l{i}\tName{i}\n" for i in range(20)) |
| 1150 | write_records(data_dir, "large", "id\tname\n" + large_rows) |
| 1151 | |
| 1152 | object_records._RECORDS_CACHE.clear() |
| 1153 | first = object_records.get_collection_record("large", "l5", base_dir=data_dir, roots=[]) |
| 1154 | assert first["name"] == "Name5" |
| 1155 | assert _cache_key("large", data_dir) not in object_records._RECORDS_CACHE |
| 1156 | |
| 1157 | object_records.update_collection_record( |
| 1158 | "large", "l5", {"name": "Updated"}, base_dir=data_dir, roots=[] |
| 1159 | ) |
| 1160 | second = object_records.get_collection_record("large", "l5", base_dir=data_dir, roots=[]) |
| 1161 | assert second["name"] == "Updated" |
| 1162 | |
| 1163 | |
| 1164 | def test_list_collection_records_window_copy_and_boundaries(tmp_path): |
| 1165 | """The pagination window copy path: correct slice, correct total/ |
| 1166 | has_more at both interior and edge offsets, and mutating a returned |
| 1167 | record must not leak into the cache or a later independent call.""" |
| 1168 | data_dir = tmp_path / "data" |
| 1169 | rows = "".join(f"r{i}\tName{i}\n" for i in range(10)) |
| 1170 | write_records(data_dir, "items", "id\tname\n" + rows) |
| 1171 | |
| 1172 | # Warm the cache first. |
| 1173 | object_records.read_collection_records("items", base_dir=data_dir, roots=[]) |
| 1174 | |
| 1175 | page = object_records.list_collection_records( |
| 1176 | "items", base_dir=data_dir, roots=[], limit=3, offset=2 |
| 1177 | ) |
| 1178 | assert page["records"] == [ |
| 1179 | {"id": "r2", "name": "Name2"}, |
| 1180 | {"id": "r3", "name": "Name3"}, |
| 1181 | {"id": "r4", "name": "Name4"}, |
| 1182 | ] |
| 1183 | assert page["total"] == 10 |
| 1184 | assert page["count"] == 3 |
| 1185 | assert page["has_more"] is True |
| 1186 | |
| 1187 | # Aliasing guard: mutating a returned record must not affect a later, |
| 1188 | # independent get/list call. |
| 1189 | page["records"][0]["name"] = "CORRUPTED" |
| 1190 | page["records"][0]["new_field"] = "also corrupted" |
| 1191 | again = object_records.get_collection_record("items", "r2", base_dir=data_dir, roots=[]) |
| 1192 | assert again == {"id": "r2", "name": "Name2"} |
| 1193 | relisted = object_records.list_collection_records( |
| 1194 | "items", base_dir=data_dir, roots=[], limit=3, offset=2 |
| 1195 | ) |
| 1196 | assert relisted["records"][0] == {"id": "r2", "name": "Name2"} |
| 1197 | |
| 1198 | # offset beyond the end of the collection. |
| 1199 | empty_page = object_records.list_collection_records( |
| 1200 | "items", base_dir=data_dir, roots=[], limit=5, offset=100 |
| 1201 | ) |
| 1202 | assert empty_page == { |
| 1203 | "collection": "items", |
| 1204 | "records": [], |
| 1205 | "count": 0, |
| 1206 | "total": 10, |
| 1207 | "limit": 5, |
| 1208 | "offset": 100, |
| 1209 | "has_more": False, |
| 1210 | } |
| 1211 | |
| 1212 | # limit reaching past the end of the collection. |
| 1213 | tail_page = object_records.list_collection_records( |
| 1214 | "items", base_dir=data_dir, roots=[], limit=100, offset=8 |
| 1215 | ) |
| 1216 | assert [record["id"] for record in tail_page["records"]] == ["r8", "r9"] |
| 1217 | assert tail_page["count"] == 2 |
| 1218 | assert tail_page["total"] == 10 |
| 1219 | assert tail_page["has_more"] is False |
| 1220 | |
| 1221 | |
| 1222 | def test_list_collection_records_warm_copies_bounded_by_window(tmp_path, monkeypatch): |
| 1223 | """Perf guard for the pagination fix: a warm list with a small limit on |
| 1224 | a large collection must copy only the window's records, not every row |
| 1225 | in the collection. Verified precisely (not by timing, which is |
| 1226 | unreliable on a network share) by shadowing the module's `dict` name |
| 1227 | and counting calls made while servicing the list request.""" |
| 1228 | data_dir = tmp_path / "data" |
| 1229 | path = data_dir / "collections" / "big" / "records.tsv" |
| 1230 | path.parent.mkdir(parents=True, exist_ok=True) |
| 1231 | with path.open("w", newline="") as handle: |
| 1232 | handle.write("id\tname\n") |
| 1233 | for i in range(50_000): |
| 1234 | handle.write(f"r{i}\tName{i}\n") |
| 1235 | |
| 1236 | # Warm the cache with an initial read, outside the counted region. |
| 1237 | warm = object_records.read_collection_records("big", base_dir=data_dir, roots=[]) |
| 1238 | assert len(warm) == 50_000 |
| 1239 | cache_records = object_records._cached_records_ref("big", base_dir=data_dir) |
| 1240 | assert len(cache_records) == 50_000 |
| 1241 | |
| 1242 | copy_calls: list[None] = [] |
| 1243 | real_dict = dict |
| 1244 | |
| 1245 | def counting_dict(*args, **kwargs): |
| 1246 | copy_calls.append(None) |
| 1247 | return real_dict(*args, **kwargs) |
| 1248 | |
| 1249 | monkeypatch.setattr(object_records, "dict", counting_dict, raising=False) |
| 1250 | |
| 1251 | result = object_records.list_collection_records( |
| 1252 | "big", base_dir=data_dir, roots=[], limit=10, offset=0 |
| 1253 | ) |
| 1254 | |
| 1255 | assert result["total"] == 50_000 |
| 1256 | assert result["count"] == 10 |
| 1257 | assert len(result["records"]) == 10 |
| 1258 | for i, record in enumerate(result["records"]): |
| 1259 | assert record == {"id": f"r{i}", "name": f"Name{i}"} |
| 1260 | # Each returned record is its own copy, never the cache's dict. |
| 1261 | assert record is not cache_records[i] |
| 1262 | |
| 1263 | # Exactly one dict() copy per windowed record -- not one per row in |
| 1264 | # the 50k-row collection. |
| 1265 | assert len(copy_calls) == 10 |
| 1266 | |
| 1267 | |
| 1268 | # --- Append-only storage (docs/append-only-storage-design.md) --- |
| 1269 | |
| 1270 | |
| 1271 | def write_append_schema(data_dir: Path, collection: str, fields: list[dict] | None = None) -> Path: |
| 1272 | """Like write_schema, but opts the collection into append storage.""" |
| 1273 | path = data_dir / "schemas" / f"{collection}.json" |
| 1274 | path.parent.mkdir(parents=True, exist_ok=True) |
| 1275 | path.write_text( |
| 1276 | json.dumps({"fields": fields or [{"name": "id"}], "storage": "append"}) |
| 1277 | ) |
| 1278 | return path |
| 1279 | |
| 1280 | |
| 1281 | def append_physical_rows( |
| 1282 | path: Path, physical_fields: list[str], rows: list[dict[str, str]] |
| 1283 | ) -> None: |
| 1284 | """Simulate an external writer (another process, the CLI, a sibling |
| 1285 | pool worker) appending physical rows directly to an already |
| 1286 | append-physical records.tsv -- a plain file write that goes through |
| 1287 | none of object_records' own API or cache-refresh machinery, matching |
| 1288 | exactly the physical format object_records._append_records_rows |
| 1289 | itself writes: tab-delimited, "\\n" line terminator, QUOTE_MINIMAL, |
| 1290 | one row per dict in `physical_fields` order (including the `_op` |
| 1291 | column -- pass object_records.OP_FIELD explicitly in each row).""" |
| 1292 | with path.open("a", newline="") as handle: |
| 1293 | writer = csv.writer( |
| 1294 | handle, delimiter="\t", lineterminator="\n", quoting=csv.QUOTE_MINIMAL |
| 1295 | ) |
| 1296 | for row in rows: |
| 1297 | writer.writerow([row.get(field, "") for field in physical_fields]) |
| 1298 | |
| 1299 | |
| 1300 | def test_op_field_is_rejected_in_user_payloads(tmp_path): |
| 1301 | data_dir = tmp_path / "data" |
| 1302 | write_records(data_dir, "widgets", "id\tname\n") |
| 1303 | |
| 1304 | with pytest.raises(object_records.InvalidRecordPayloadError, match="reserved"): |
| 1305 | object_records.create_collection_record( |
| 1306 | "widgets", {"id": "w1", "name": "Gadget", "_op": "del"}, base_dir=data_dir, roots=[] |
| 1307 | ) |
| 1308 | |
| 1309 | |
| 1310 | def test_schema_field_named_op_is_rejected(tmp_path): |
| 1311 | with pytest.raises(ValueError, match="reserved"): |
| 1312 | object_schemas.normalize_schema("widgets", {"fields": [{"name": "_op"}]}) |
| 1313 | |
| 1314 | |
| 1315 | def test_schema_storage_key_rejects_invalid_values(tmp_path): |
| 1316 | with pytest.raises(ValueError, match="storage"): |
| 1317 | object_schemas.normalize_schema("widgets", {"fields": [], "storage": "bogus"}) |
| 1318 | |
| 1319 | |
| 1320 | def test_schema_storage_key_survives_normalization(tmp_path): |
| 1321 | normalized = object_schemas.normalize_schema( |
| 1322 | "widgets", {"fields": [{"name": "id"}], "storage": "append"} |
| 1323 | ) |
| 1324 | assert normalized["storage"] == "append" |
| 1325 | |
| 1326 | # Absent by default -- byte-identical to a schema that never mentions |
| 1327 | # storage (see the classic-mode-untouched test below). |
| 1328 | classic = object_schemas.normalize_schema("widgets", {"fields": [{"name": "id"}]}) |
| 1329 | assert "storage" not in classic |
| 1330 | |
| 1331 | |
| 1332 | def test_classic_mode_is_byte_identical_to_before_append_storage_existed(tmp_path): |
| 1333 | """A collection with no `storage` key must write records.tsv exactly |
| 1334 | as it always has -- no `_op` column, no folding, no torn-tail |
| 1335 | tolerance. This is the opt-in guarantee: default/absent means |
| 1336 | unchanged behavior.""" |
| 1337 | data_dir = tmp_path / "data" |
| 1338 | write_schema(data_dir, "contacts", [{"name": "id"}, {"name": "name"}]) |
| 1339 | |
| 1340 | object_records.create_collection_record( |
| 1341 | "contacts", {"id": "c1", "name": "Ada"}, base_dir=data_dir, roots=[] |
| 1342 | ) |
| 1343 | object_records.create_collection_record( |
| 1344 | "contacts", {"id": "c2", "name": "Grace"}, base_dir=data_dir, roots=[] |
| 1345 | ) |
| 1346 | object_records.update_collection_record( |
| 1347 | "contacts", "c1", {"name": "Ada Lovelace"}, base_dir=data_dir, roots=[] |
| 1348 | ) |
| 1349 | object_records.delete_collection_record("contacts", "c2", base_dir=data_dir, roots=[]) |
| 1350 | |
| 1351 | path = data_dir / "collections" / "contacts" / "records.tsv" |
| 1352 | assert path.read_text() == "id\tname\nc1\tAda Lovelace\n" |
| 1353 | assert "_op" not in path.read_text().splitlines()[0].split("\t") |
| 1354 | |
| 1355 | |
| 1356 | def test_append_mode_create_update_delete_use_op_column(tmp_path): |
| 1357 | data_dir = tmp_path / "data" |
| 1358 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1359 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1360 | |
| 1361 | object_records.create_collection_record( |
| 1362 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1363 | ) |
| 1364 | object_records.create_collection_record( |
| 1365 | "widgets", {"id": "w2", "name": "Beta"}, base_dir=data_dir, roots=[] |
| 1366 | ) |
| 1367 | with path.open(newline="") as handle: |
| 1368 | rows = list(csv.reader(handle, delimiter="\t")) |
| 1369 | assert rows[0] == ["_op", "id", "name"] |
| 1370 | assert rows[1] == ["", "w1", "Alpha"] |
| 1371 | assert rows[2] == ["", "w2", "Beta"] |
| 1372 | |
| 1373 | object_records.update_collection_record( |
| 1374 | "widgets", "w1", {"name": "Alpha2"}, base_dir=data_dir, roots=[] |
| 1375 | ) |
| 1376 | with path.open(newline="") as handle: |
| 1377 | rows = list(csv.reader(handle, delimiter="\t")) |
| 1378 | # update appends a superseding row rather than rewriting in place. |
| 1379 | assert len(rows) == 4 |
| 1380 | assert rows[3] == ["", "w1", "Alpha2"] |
| 1381 | |
| 1382 | removed = object_records.delete_collection_record("widgets", "w2", base_dir=data_dir, roots=[]) |
| 1383 | assert removed == {"id": "w2", "name": "Beta"} |
| 1384 | with path.open(newline="") as handle: |
| 1385 | rows = list(csv.reader(handle, delimiter="\t")) |
| 1386 | # Tombstone visible in the raw file, pre-compaction: obvious in a cat. |
| 1387 | assert rows[4] == ["del", "w2", ""] |
| 1388 | |
| 1389 | with pytest.raises(object_records.RecordNotFoundError): |
| 1390 | object_records.delete_collection_record("widgets", "w2", base_dir=data_dir, roots=[]) |
| 1391 | |
| 1392 | listing = object_records.list_collection_records("widgets", base_dir=data_dir, roots=[])["records"] |
| 1393 | assert listing == [{"id": "w1", "name": "Alpha2"}] |
| 1394 | |
| 1395 | |
| 1396 | def test_append_mode_resurrection_after_delete(tmp_path): |
| 1397 | data_dir = tmp_path / "data" |
| 1398 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1399 | |
| 1400 | object_records.create_collection_record( |
| 1401 | "widgets", {"id": "w1", "name": "one"}, base_dir=data_dir, roots=[] |
| 1402 | ) |
| 1403 | object_records.delete_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1404 | |
| 1405 | with pytest.raises(object_records.RecordNotFoundError): |
| 1406 | object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1407 | |
| 1408 | resurrected = object_records.create_collection_record( |
| 1409 | "widgets", {"id": "w1", "name": "reborn"}, base_dir=data_dir, roots=[] |
| 1410 | ) |
| 1411 | assert resurrected == {"id": "w1", "name": "reborn"} |
| 1412 | assert object_records.get_collection_record( |
| 1413 | "widgets", "w1", base_dir=data_dir, roots=[] |
| 1414 | ) == {"id": "w1", "name": "reborn"} |
| 1415 | |
| 1416 | |
| 1417 | def test_append_mode_transition_on_next_write_adds_op_column(tmp_path): |
| 1418 | """Switching a classic collection to storage:append doesn't touch the |
| 1419 | file until the collection's next write, which performs one final |
| 1420 | classic-shaped rewrite that adds the `_op` header column (existing |
| 1421 | rows get "").""" |
| 1422 | data_dir = tmp_path / "data" |
| 1423 | schema_path = write_schema(data_dir, "notes", [{"name": "id"}, {"name": "title"}]) |
| 1424 | object_records.create_collection_record( |
| 1425 | "notes", {"id": "n1", "title": "one"}, base_dir=data_dir, roots=[] |
| 1426 | ) |
| 1427 | object_records.create_collection_record( |
| 1428 | "notes", {"id": "n2", "title": "two"}, base_dir=data_dir, roots=[] |
| 1429 | ) |
| 1430 | path = data_dir / "collections" / "notes" / "records.tsv" |
| 1431 | assert path.read_text() == "id\ttitle\nn1\tone\nn2\ttwo\n" |
| 1432 | |
| 1433 | # Opt in -- the file on disk does not change yet, only the schema does. |
| 1434 | schema_path.write_text( |
| 1435 | json.dumps({"fields": [{"name": "id"}, {"name": "title"}], "storage": "append"}) |
| 1436 | ) |
| 1437 | assert path.read_text() == "id\ttitle\nn1\tone\nn2\ttwo\n" |
| 1438 | |
| 1439 | updated = object_records.update_collection_record( |
| 1440 | "notes", "n1", {"title": "one-updated"}, base_dir=data_dir, roots=[] |
| 1441 | ) |
| 1442 | assert updated == {"id": "n1", "title": "one-updated"} |
| 1443 | |
| 1444 | with path.open(newline="") as handle: |
| 1445 | rows = list(csv.reader(handle, delimiter="\t")) |
| 1446 | assert rows[0] == ["_op", "id", "title"] |
| 1447 | assert rows[1] == ["", "n1", "one-updated"] |
| 1448 | assert rows[2] == ["", "n2", "two"] |
| 1449 | assert len(rows) == 3 |
| 1450 | |
| 1451 | |
| 1452 | def test_append_mode_switch_back_compacts_to_classic_form(tmp_path): |
| 1453 | """Removing storage:append causes the collection's next write to fold |
| 1454 | current content and drop the `_op` column -- byte-compatible with a |
| 1455 | classic file (docs/append-only-storage-design.md Migration and |
| 1456 | Compatibility: "A compacted append-only file is byte-compatible with |
| 1457 | a classic one").""" |
| 1458 | data_dir = tmp_path / "data" |
| 1459 | schema_path = write_append_schema(data_dir, "notes", [{"name": "id"}, {"name": "title"}]) |
| 1460 | object_records.create_collection_record( |
| 1461 | "notes", {"id": "n1", "title": "one"}, base_dir=data_dir, roots=[] |
| 1462 | ) |
| 1463 | object_records.update_collection_record( |
| 1464 | "notes", "n1", {"title": "one-updated"}, base_dir=data_dir, roots=[] |
| 1465 | ) |
| 1466 | object_records.create_collection_record( |
| 1467 | "notes", {"id": "n2", "title": "two"}, base_dir=data_dir, roots=[] |
| 1468 | ) |
| 1469 | path = data_dir / "collections" / "notes" / "records.tsv" |
| 1470 | with path.open(newline="") as handle: |
| 1471 | rows_before = list(csv.reader(handle, delimiter="\t")) |
| 1472 | assert rows_before[0][0] == "_op" |
| 1473 | assert len(rows_before) == 4 # header + 3 physical rows (1 superseded) |
| 1474 | |
| 1475 | # Opt back out. |
| 1476 | schema_path.write_text( |
| 1477 | json.dumps({"fields": [{"name": "id"}, {"name": "title"}]}) |
| 1478 | ) |
| 1479 | object_records.update_collection_record( |
| 1480 | "notes", "n2", {"title": "two-updated"}, base_dir=data_dir, roots=[] |
| 1481 | ) |
| 1482 | |
| 1483 | assert path.read_text() == "id\ttitle\nn1\tone-updated\nn2\ttwo-updated\n" |
| 1484 | |
| 1485 | |
| 1486 | def test_append_mode_new_field_falls_back_to_full_rewrite_and_extends_header(tmp_path): |
| 1487 | """Append cannot extend an existing physical header: a write that |
| 1488 | introduces a field the current header doesn't have must fall back to |
| 1489 | a full rewrite that folds current content and adds the column -- the |
| 1490 | new value must not be lost.""" |
| 1491 | data_dir = tmp_path / "data" |
| 1492 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1493 | object_records.create_collection_record( |
| 1494 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1495 | ) |
| 1496 | object_records.create_collection_record( |
| 1497 | "widgets", {"id": "w2", "name": "Beta"}, base_dir=data_dir, roots=[] |
| 1498 | ) |
| 1499 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1500 | with path.open(newline="") as handle: |
| 1501 | header_before = next(csv.reader(handle, delimiter="\t")) |
| 1502 | assert "color" not in header_before |
| 1503 | |
| 1504 | updated = object_records.update_collection_record( |
| 1505 | "widgets", "w1", {"color": "red"}, base_dir=data_dir, roots=[] |
| 1506 | ) |
| 1507 | assert updated == {"id": "w1", "name": "Alpha", "color": "red"} |
| 1508 | |
| 1509 | with path.open(newline="") as handle: |
| 1510 | rows = list(csv.reader(handle, delimiter="\t")) |
| 1511 | assert rows[0] == ["_op", "id", "name", "color"] |
| 1512 | assert "color" in rows[0] |
| 1513 | |
| 1514 | fetched = object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1515 | assert fetched["color"] == "red" |
| 1516 | listing = object_records.list_collection_records("widgets", base_dir=data_dir, roots=[])["records"] |
| 1517 | assert {r["id"]: r["color"] for r in listing} == {"w1": "red", "w2": ""} |
| 1518 | |
| 1519 | |
| 1520 | def test_compact_collection_removes_superseded_and_tombstoned_rows(tmp_path): |
| 1521 | data_dir = tmp_path / "data" |
| 1522 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1523 | object_records.create_collection_record( |
| 1524 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1525 | ) |
| 1526 | object_records.create_collection_record( |
| 1527 | "widgets", {"id": "w2", "name": "Beta"}, base_dir=data_dir, roots=[] |
| 1528 | ) |
| 1529 | object_records.update_collection_record( |
| 1530 | "widgets", "w1", {"name": "Alpha2"}, base_dir=data_dir, roots=[] |
| 1531 | ) |
| 1532 | object_records.update_collection_record( |
| 1533 | "widgets", "w1", {"name": "Alpha3"}, base_dir=data_dir, roots=[] |
| 1534 | ) |
| 1535 | object_records.delete_collection_record("widgets", "w2", base_dir=data_dir, roots=[]) |
| 1536 | |
| 1537 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1538 | raw_before = path.read_text() |
| 1539 | assert "del" in raw_before # tombstone present pre-compaction |
| 1540 | |
| 1541 | result = object_records.compact_collection("widgets", base_dir=data_dir, roots=[]) |
| 1542 | assert result["rows_before"] == 5 |
| 1543 | assert result["rows_after"] == 1 |
| 1544 | assert result["bytes_after"] < result["bytes_before"] |
| 1545 | |
| 1546 | raw_after = path.read_text() |
| 1547 | assert "del" not in raw_after |
| 1548 | assert raw_after == "_op\tid\tname\n\tw1\tAlpha3\n" |
| 1549 | |
| 1550 | # Data is unaffected by compaction. |
| 1551 | listing = object_records.list_collection_records("widgets", base_dir=data_dir, roots=[])["records"] |
| 1552 | assert listing == [{"id": "w1", "name": "Alpha3"}] |
| 1553 | |
| 1554 | |
| 1555 | def test_compact_collection_is_a_reported_no_op_for_classic_collections(tmp_path): |
| 1556 | data_dir = tmp_path / "data" |
| 1557 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 1558 | |
| 1559 | result = object_records.compact_collection("contacts", base_dir=data_dir, roots=[]) |
| 1560 | assert result == {"rows_before": 1, "rows_after": 1, "bytes_before": result["bytes_before"], "bytes_after": result["bytes_before"]} |
| 1561 | assert (data_dir / "collections" / "contacts" / "records.tsv").read_text() == "id\tname\nc1\tAda\n" |
| 1562 | |
| 1563 | |
| 1564 | def test_append_mode_auto_compact_triggers_on_next_write_past_threshold(tmp_path, monkeypatch): |
| 1565 | """When a cold parse observes physical rows past the (monkeypatched |
| 1566 | small) threshold, with superseded+deleted rows outnumbering live |
| 1567 | rows, it flags the collection -- but performs no write itself (reads |
| 1568 | stay read-only). The collection's NEXT write then compacts instead of |
| 1569 | appending.""" |
| 1570 | monkeypatch.setenv("DBBASIC_APPEND_COMPACT_MIN_ROWS", "5") |
| 1571 | data_dir = tmp_path / "data" |
| 1572 | write_append_schema(data_dir, "churn", [{"name": "id"}, {"name": "value"}]) |
| 1573 | |
| 1574 | object_records.create_collection_record( |
| 1575 | "churn", {"id": "c1", "value": "v0"}, base_dir=data_dir, roots=[] |
| 1576 | ) |
| 1577 | for i in range(10): |
| 1578 | object_records.update_collection_record( |
| 1579 | "churn", "c1", {"value": f"v{i}"}, base_dir=data_dir, roots=[] |
| 1580 | ) |
| 1581 | |
| 1582 | path = data_dir / "collections" / "churn" / "records.tsv" |
| 1583 | physical_rows_before = len(path.read_text().splitlines()) - 1 |
| 1584 | assert physical_rows_before > 5 |
| 1585 | |
| 1586 | cache_key = str(object_records.collection_records_file("churn", base_dir=data_dir).resolve()) |
| 1587 | |
| 1588 | # Force a cold parse (simulating cache eviction / a fresh process) -- |
| 1589 | # a pure read. |
| 1590 | object_records._RECORDS_CACHE.clear() |
| 1591 | before_stat = path.stat() |
| 1592 | object_records.get_collection_record("churn", "c1", base_dir=data_dir, roots=[]) |
| 1593 | after_stat = path.stat() |
| 1594 | assert before_stat.st_mtime_ns == after_stat.st_mtime_ns |
| 1595 | assert before_stat.st_size == after_stat.st_size |
| 1596 | assert cache_key in object_records._PENDING_COMPACTION |
| 1597 | |
| 1598 | object_records.update_collection_record( |
| 1599 | "churn", "c1", {"value": "final"}, base_dir=data_dir, roots=[] |
| 1600 | ) |
| 1601 | physical_rows_after = len(path.read_text().splitlines()) - 1 |
| 1602 | assert physical_rows_after == 1 |
| 1603 | assert cache_key not in object_records._PENDING_COMPACTION |
| 1604 | |
| 1605 | listing = object_records.list_collection_records("churn", base_dir=data_dir, roots=[])["records"] |
| 1606 | assert listing == [{"id": "c1", "value": "final"}] |
| 1607 | |
| 1608 | |
| 1609 | def test_append_mode_torn_tail_is_ignored_and_self_heals(tmp_path): |
| 1610 | data_dir = tmp_path / "data" |
| 1611 | write_append_schema(data_dir, "logs", [{"name": "id"}, {"name": "value"}]) |
| 1612 | object_records.create_collection_record( |
| 1613 | "logs", {"id": "l1", "value": "one"}, base_dir=data_dir, roots=[] |
| 1614 | ) |
| 1615 | object_records.create_collection_record( |
| 1616 | "logs", {"id": "l2", "value": "two"}, base_dir=data_dir, roots=[] |
| 1617 | ) |
| 1618 | path = data_dir / "collections" / "logs" / "records.tsv" |
| 1619 | full_text = path.read_text() |
| 1620 | |
| 1621 | # Simulate a crash mid-write: chop off the last few bytes of the file, |
| 1622 | # tearing the final row (no trailing newline). |
| 1623 | path.write_text(full_text[:-4]) |
| 1624 | assert not path.read_text().endswith("\n") |
| 1625 | |
| 1626 | object_records._RECORDS_CACHE.clear() |
| 1627 | before_stat = path.stat() |
| 1628 | listing = object_records.list_collection_records("logs", base_dir=data_dir, roots=[])["records"] |
| 1629 | after_stat = path.stat() |
| 1630 | |
| 1631 | # Read succeeds, ignoring the torn fragment -- and never writes. |
| 1632 | assert listing == [{"id": "l1", "value": "one"}] |
| 1633 | assert before_stat.st_mtime_ns == after_stat.st_mtime_ns |
| 1634 | assert before_stat.st_size == after_stat.st_size |
| 1635 | |
| 1636 | # The next write self-heals: the fragment is truncated away (not |
| 1637 | # resurrected), and a fresh, well-formed row is appended after it. |
| 1638 | object_records.create_collection_record( |
| 1639 | "logs", {"id": "l3", "value": "three"}, base_dir=data_dir, roots=[] |
| 1640 | ) |
| 1641 | assert path.read_text().endswith("\n") |
| 1642 | |
| 1643 | # Subsequent reads are clean -- the fragment never reappears. |
| 1644 | object_records._RECORDS_CACHE.clear() |
| 1645 | listing_after_heal = object_records.list_collection_records( |
| 1646 | "logs", base_dir=data_dir, roots=[] |
| 1647 | )["records"] |
| 1648 | assert listing_after_heal == [ |
| 1649 | {"id": "l1", "value": "one"}, |
| 1650 | {"id": "l3", "value": "three"}, |
| 1651 | ] |
| 1652 | |
| 1653 | |
| 1654 | def test_append_mode_cache_refreshes_on_same_inode_append(tmp_path): |
| 1655 | """Unlike a classic full rewrite (always a fresh inode via atomic |
| 1656 | replace), a fast append mutates the SAME inode in place. The cache |
| 1657 | must still detect and correctly refresh on this change (only size/ |
| 1658 | mtime move) -- verified here by asserting the inode is unchanged |
| 1659 | across the write while a subsequent cache-hit read returns fresh |
| 1660 | data, not stale data from the pre-append signature.""" |
| 1661 | data_dir = tmp_path / "data" |
| 1662 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1663 | object_records.create_collection_record( |
| 1664 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1665 | ) |
| 1666 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1667 | inode_before = path.stat().st_ino |
| 1668 | |
| 1669 | object_records.update_collection_record( |
| 1670 | "widgets", "w1", {"name": "Alpha2"}, base_dir=data_dir, roots=[] |
| 1671 | ) |
| 1672 | inode_after = path.stat().st_ino |
| 1673 | assert inode_after == inode_before # same inode: a fast append, not a rewrite |
| 1674 | |
| 1675 | # A cache HIT (no reparse) must still return the fresh value. |
| 1676 | cache_key = str(path.resolve()) |
| 1677 | signature, _, cached_records, _, _ = object_records._RECORDS_CACHE[cache_key] |
| 1678 | assert signature == object_records._stat_signature(path) |
| 1679 | assert cached_records == [{"id": "w1", "name": "Alpha2"}] |
| 1680 | |
| 1681 | fetched = object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1682 | assert fetched == {"id": "w1", "name": "Alpha2"} |
| 1683 | |
| 1684 | |
| 1685 | def test_append_mode_equivalent_to_classic_mode_across_random_operations(tmp_path): |
| 1686 | """THE EQUIVALENCE PROPERTY. An identical, seeded, randomized sequence |
| 1687 | of create/update/delete/get/list operations run against a classic |
| 1688 | collection and an append-mode collection (same schema otherwise) must |
| 1689 | produce API-visible-identical results after EVERY single operation: |
| 1690 | same returned records, same exception types, same list contents (see |
| 1691 | _fold_append_rows for why order matches classic exactly), same |
| 1692 | totals. Deterministic seed, no wall-clock dependence -- every value |
| 1693 | written is derived from the loop counter, never wall-clock time. |
| 1694 | """ |
| 1695 | classic_dir = tmp_path / "classic" |
| 1696 | append_dir = tmp_path / "append" |
| 1697 | write_schema(classic_dir, "items", [{"name": "id"}, {"name": "value"}]) |
| 1698 | write_append_schema(append_dir, "items", [{"name": "id"}, {"name": "value"}]) |
| 1699 | |
| 1700 | def apply(data_dir, kind, record_id, payload): |
| 1701 | try: |
| 1702 | if kind == "create": |
| 1703 | return ("ok", object_records.create_collection_record( |
| 1704 | "items", {"id": record_id, "value": payload}, base_dir=data_dir, roots=[] |
| 1705 | )) |
| 1706 | if kind == "update": |
| 1707 | return ("ok", object_records.update_collection_record( |
| 1708 | "items", record_id, {"value": payload}, base_dir=data_dir, roots=[] |
| 1709 | )) |
| 1710 | if kind == "delete": |
| 1711 | return ("ok", object_records.delete_collection_record( |
| 1712 | "items", record_id, base_dir=data_dir, roots=[] |
| 1713 | )) |
| 1714 | if kind == "get": |
| 1715 | return ("ok", object_records.get_collection_record( |
| 1716 | "items", record_id, base_dir=data_dir, roots=[] |
| 1717 | )) |
| 1718 | return ("ok", object_records.list_collection_records( |
| 1719 | "items", base_dir=data_dir, roots=[] |
| 1720 | )) |
| 1721 | except Exception as exc: # noqa: BLE001 -- comparing exception SHAPES across both runs |
| 1722 | return ("error", type(exc), str(exc)) |
| 1723 | |
| 1724 | rng = random.Random(20260718) |
| 1725 | id_pool = [f"e{i}" for i in range(12)] |
| 1726 | |
| 1727 | for step in range(200): |
| 1728 | kind = rng.choice(["create", "update", "delete", "get", "list"]) |
| 1729 | record_id = rng.choice(id_pool) |
| 1730 | payload = f"v{step}" |
| 1731 | |
| 1732 | classic_outcome = apply(classic_dir, kind, record_id, payload) |
| 1733 | append_outcome = apply(append_dir, kind, record_id, payload) |
| 1734 | |
| 1735 | assert classic_outcome == append_outcome, ( |
| 1736 | f"step {step} op={kind!r} id={record_id!r}: " |
| 1737 | f"classic={classic_outcome!r} append={append_outcome!r}" |
| 1738 | ) |
| 1739 | |
| 1740 | # Independent final-state cross-check, beyond the per-step comparisons. |
| 1741 | classic_listing = object_records.list_collection_records("items", base_dir=classic_dir, roots=[]) |
| 1742 | append_listing = object_records.list_collection_records("items", base_dir=append_dir, roots=[]) |
| 1743 | assert classic_listing == append_listing |
| 1744 | |
| 1745 | # And a full compaction on the append side must not change what it |
| 1746 | # reports afterward, either. |
| 1747 | object_records.compact_collection("items", base_dir=append_dir, roots=[]) |
| 1748 | append_listing_after_compact = object_records.list_collection_records( |
| 1749 | "items", base_dir=append_dir, roots=[] |
| 1750 | ) |
| 1751 | assert append_listing_after_compact == classic_listing |
| 1752 | |
| 1753 | |
| 1754 | # --------------------------------------------------------------------------- |
| 1755 | # id -> byte-offset sidecar (docs/append-only-storage-design.md item 4). |
| 1756 | # |
| 1757 | # Every test below forces DBBASIC_RECORDS_CACHE_MAX_ROWS to 0 so that |
| 1758 | # _RECORDS_CACHE never stores an entry for these (tiny, real-world-normal- |
| 1759 | # sized) test collections -- i.e. every point op is a cache MISS, which is |
| 1760 | # exactly the condition _fast_record_lookup uses to try the sidecar. This |
| 1761 | # is what lets a handful of records exercise the same code path a real |
| 1762 | # over-threshold, million-row collection would use, without the tests |
| 1763 | # needing to actually write a million rows. |
| 1764 | # --------------------------------------------------------------------------- |
| 1765 | |
| 1766 | |
| 1767 | def test_oidx_sidecar_builds_lazily_and_serves_point_ops(tmp_path, monkeypatch): |
| 1768 | """No sidecar file exists until the first point op against an already |
| 1769 | append-physical file with a cold cache; from then on, get/create's |
| 1770 | duplicate check/update/delete all resolve correctly through it.""" |
| 1771 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 1772 | data_dir = tmp_path / "data" |
| 1773 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1774 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1775 | oidx_path = object_records._oidx_path(path) |
| 1776 | |
| 1777 | assert not oidx_path.exists() |
| 1778 | object_records.create_collection_record( |
| 1779 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1780 | ) |
| 1781 | # This first write is a TRANSITION-IN full rewrite (the file only |
| 1782 | # just became append-physical) -- nothing has read through the |
| 1783 | # sidecar yet, so it still doesn't exist. |
| 1784 | assert not oidx_path.exists() |
| 1785 | |
| 1786 | object_records.create_collection_record( |
| 1787 | "widgets", {"id": "w2", "name": "Beta"}, base_dir=data_dir, roots=[] |
| 1788 | ) |
| 1789 | # This create's duplicate check is the first point op against an |
| 1790 | # already append-physical file with a cold cache -- it must have |
| 1791 | # built the sidecar. |
| 1792 | assert oidx_path.exists() |
| 1793 | |
| 1794 | with pytest.raises(object_records.DuplicateRecordIdError): |
| 1795 | object_records.create_collection_record( |
| 1796 | "widgets", {"id": "w1", "name": "Gamma"}, base_dir=data_dir, roots=[] |
| 1797 | ) |
| 1798 | |
| 1799 | assert object_records.get_collection_record( |
| 1800 | "widgets", "w2", base_dir=data_dir, roots=[] |
| 1801 | ) == {"id": "w2", "name": "Beta"} |
| 1802 | |
| 1803 | updated = object_records.update_collection_record( |
| 1804 | "widgets", "w1", {"name": "Alpha2"}, base_dir=data_dir, roots=[] |
| 1805 | ) |
| 1806 | assert updated == {"id": "w1", "name": "Alpha2"} |
| 1807 | assert object_records.get_collection_record( |
| 1808 | "widgets", "w1", base_dir=data_dir, roots=[] |
| 1809 | ) == {"id": "w1", "name": "Alpha2"} |
| 1810 | |
| 1811 | removed = object_records.delete_collection_record( |
| 1812 | "widgets", "w2", base_dir=data_dir, roots=[] |
| 1813 | ) |
| 1814 | assert removed == {"id": "w2", "name": "Beta"} |
| 1815 | with pytest.raises(object_records.RecordNotFoundError): |
| 1816 | object_records.get_collection_record("widgets", "w2", base_dir=data_dir, roots=[]) |
| 1817 | |
| 1818 | listing = object_records.list_collection_records( |
| 1819 | "widgets", base_dir=data_dir, roots=[] |
| 1820 | )["records"] |
| 1821 | assert listing == [{"id": "w1", "name": "Alpha2"}] |
| 1822 | |
| 1823 | |
| 1824 | def test_oidx_sidecar_catches_up_when_behind(tmp_path, monkeypatch): |
| 1825 | """A sidecar whose body is missing its last few lines (simulating a |
| 1826 | crash between a data append and its matching idx append, or another |
| 1827 | writer that didn't maintain the sidecar at all) is caught up by |
| 1828 | scanning only the uncovered tail of records.tsv, not rebuilt from |
| 1829 | scratch.""" |
| 1830 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 1831 | data_dir = tmp_path / "data" |
| 1832 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1833 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1834 | oidx_path = object_records._oidx_path(path) |
| 1835 | |
| 1836 | for i in range(5): |
| 1837 | object_records.create_collection_record( |
| 1838 | "widgets", {"id": f"w{i}", "name": f"n{i}"}, base_dir=data_dir, roots=[] |
| 1839 | ) |
| 1840 | object_records._OIDX_CACHE.clear() |
| 1841 | object_records.get_collection_record("widgets", "w0", base_dir=data_dir, roots=[]) |
| 1842 | assert oidx_path.exists() |
| 1843 | |
| 1844 | lines = oidx_path.read_text().splitlines(keepends=True) |
| 1845 | assert len(lines) == 6 # header + 5 data lines |
| 1846 | oidx_path.write_text("".join(lines[:-2])) |
| 1847 | object_records._OIDX_CACHE.clear() |
| 1848 | |
| 1849 | fetched4 = object_records.get_collection_record("widgets", "w4", base_dir=data_dir, roots=[]) |
| 1850 | assert fetched4 == {"id": "w4", "name": "n4"} |
| 1851 | fetched0 = object_records.get_collection_record("widgets", "w0", base_dir=data_dir, roots=[]) |
| 1852 | assert fetched0 == {"id": "w0", "name": "n0"} |
| 1853 | |
| 1854 | id_offsets, coherent = object_records._load_oidx(path) |
| 1855 | assert coherent |
| 1856 | assert set(id_offsets) == {"w0", "w1", "w2", "w3", "w4"} |
| 1857 | assert len(oidx_path.read_text().splitlines()) == 6 |
| 1858 | |
| 1859 | |
| 1860 | def test_oidx_sidecar_rebuilds_on_stale_ino_header(tmp_path, monkeypatch): |
| 1861 | """A sidecar whose header ino doesn't match the data file's current |
| 1862 | inode (e.g. left over from before a compaction) must never be |
| 1863 | trusted -- _load_oidx rebuilds from scratch instead.""" |
| 1864 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 1865 | data_dir = tmp_path / "data" |
| 1866 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1867 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1868 | oidx_path = object_records._oidx_path(path) |
| 1869 | |
| 1870 | object_records.create_collection_record( |
| 1871 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1872 | ) |
| 1873 | |
| 1874 | # Hand-write a sidecar with a bogus header ino and a body entry for |
| 1875 | # an id that was never actually written -- if the mismatch weren't |
| 1876 | # detected, this would leak straight into the answer. |
| 1877 | oidx_path.write_text("oidx1\t999999999\n0\t20\t\tbogus\n") |
| 1878 | object_records._OIDX_CACHE.clear() |
| 1879 | |
| 1880 | id_offsets, coherent = object_records._load_oidx(path) |
| 1881 | assert coherent |
| 1882 | assert "bogus" not in id_offsets |
| 1883 | assert "w1" in id_offsets |
| 1884 | assert object_records.get_collection_record( |
| 1885 | "widgets", "w1", base_dir=data_dir, roots=[] |
| 1886 | ) == {"id": "w1", "name": "Alpha"} |
| 1887 | |
| 1888 | |
| 1889 | def test_oidx_sidecar_rebuilds_on_corrupt_sidecar_file(tmp_path, monkeypatch): |
| 1890 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 1891 | data_dir = tmp_path / "data" |
| 1892 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1893 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1894 | oidx_path = object_records._oidx_path(path) |
| 1895 | |
| 1896 | object_records.create_collection_record( |
| 1897 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1898 | ) |
| 1899 | object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1900 | assert oidx_path.exists() |
| 1901 | |
| 1902 | oidx_path.write_text("not even close to a valid sidecar\n\x00garbage\tmore\tfields\there\n") |
| 1903 | object_records._OIDX_CACHE.clear() |
| 1904 | |
| 1905 | # Never raises -- self-heals via a rebuild, and the API call still |
| 1906 | # returns the correct record. |
| 1907 | fetched = object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1908 | assert fetched == {"id": "w1", "name": "Alpha"} |
| 1909 | id_offsets, coherent = object_records._load_oidx(path) |
| 1910 | assert coherent |
| 1911 | assert set(id_offsets) == {"w1"} |
| 1912 | |
| 1913 | |
| 1914 | def test_oidx_sidecar_rebuilds_when_file_missing(tmp_path, monkeypatch): |
| 1915 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 1916 | data_dir = tmp_path / "data" |
| 1917 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1918 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1919 | oidx_path = object_records._oidx_path(path) |
| 1920 | |
| 1921 | object_records.create_collection_record( |
| 1922 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1923 | ) |
| 1924 | object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1925 | assert oidx_path.exists() |
| 1926 | |
| 1927 | oidx_path.unlink() |
| 1928 | object_records._OIDX_CACHE.clear() |
| 1929 | |
| 1930 | fetched = object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1931 | assert fetched == {"id": "w1", "name": "Alpha"} |
| 1932 | assert oidx_path.exists() |
| 1933 | |
| 1934 | |
| 1935 | def test_oidx_sidecar_rebuilds_after_compaction_changes_inode(tmp_path, monkeypatch): |
| 1936 | """Compaction rewrites records.tsv atomically (new inode) and |
| 1937 | explicitly discards the sidecar (docs/append-only-storage-design.md: |
| 1938 | "delete or rebuild sidecar after the rewrite") -- a subsequent point |
| 1939 | op rebuilds correctly against the compacted content.""" |
| 1940 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 1941 | data_dir = tmp_path / "data" |
| 1942 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1943 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1944 | oidx_path = object_records._oidx_path(path) |
| 1945 | |
| 1946 | object_records.create_collection_record( |
| 1947 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1948 | ) |
| 1949 | object_records.update_collection_record( |
| 1950 | "widgets", "w1", {"name": "Alpha2"}, base_dir=data_dir, roots=[] |
| 1951 | ) |
| 1952 | object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1953 | assert oidx_path.exists() |
| 1954 | ino_before = path.stat().st_ino |
| 1955 | |
| 1956 | object_records.compact_collection("widgets", base_dir=data_dir, roots=[]) |
| 1957 | assert not oidx_path.exists() |
| 1958 | assert path.stat().st_ino != ino_before |
| 1959 | |
| 1960 | fetched = object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1961 | assert fetched == {"id": "w1", "name": "Alpha2"} |
| 1962 | assert oidx_path.exists() |
| 1963 | |
| 1964 | |
| 1965 | def test_oidx_sidecar_torn_idx_tail_is_dropped(tmp_path, monkeypatch): |
| 1966 | """A torn sidecar line (no trailing newline on its own final physical |
| 1967 | line, from a crash mid idx-write) is dropped, not resurrected as a |
| 1968 | bogus entry -- the catch-up scan of records.tsv resupplies the |
| 1969 | correct data for whatever the torn line was trying to record.""" |
| 1970 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 1971 | data_dir = tmp_path / "data" |
| 1972 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 1973 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 1974 | oidx_path = object_records._oidx_path(path) |
| 1975 | |
| 1976 | object_records.create_collection_record( |
| 1977 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 1978 | ) |
| 1979 | object_records.create_collection_record( |
| 1980 | "widgets", {"id": "w2", "name": "Beta"}, base_dir=data_dir, roots=[] |
| 1981 | ) |
| 1982 | object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 1983 | assert oidx_path.exists() |
| 1984 | |
| 1985 | text = oidx_path.read_text() |
| 1986 | assert text.endswith("\n") |
| 1987 | oidx_path.write_text(text[:-4]) |
| 1988 | assert not oidx_path.read_text().endswith("\n") |
| 1989 | |
| 1990 | object_records._OIDX_CACHE.clear() |
| 1991 | id_offsets, coherent = object_records._load_oidx(path) |
| 1992 | assert coherent |
| 1993 | assert set(id_offsets) == {"w1", "w2"} |
| 1994 | assert object_records.get_collection_record( |
| 1995 | "widgets", "w2", base_dir=data_dir, roots=[] |
| 1996 | ) == {"id": "w2", "name": "Beta"} |
| 1997 | |
| 1998 | |
| 1999 | def test_oidx_sidecar_survives_torn_data_tail_repair(tmp_path, monkeypatch): |
| 2000 | """A crash mid-write can leave records.tsv itself with a torn final |
| 2001 | row. The next append self-heals it (_repair_torn_tail truncates the |
| 2002 | fragment away before appending), which can leave a previously-built |
| 2003 | sidecar's covered_bytes pointing past the (now shorter, then |
| 2004 | re-grown) file -- _load_oidx must detect and rebuild rather than |
| 2005 | trust stale offsets, at every step of that sequence.""" |
| 2006 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 2007 | data_dir = tmp_path / "data" |
| 2008 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2009 | object_records.create_collection_record( |
| 2010 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 2011 | ) |
| 2012 | object_records.create_collection_record( |
| 2013 | "widgets", {"id": "w2", "name": "Beta"}, base_dir=data_dir, roots=[] |
| 2014 | ) |
| 2015 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 2016 | oidx_path = object_records._oidx_path(path) |
| 2017 | object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 2018 | assert oidx_path.exists() |
| 2019 | |
| 2020 | full_text = path.read_text() |
| 2021 | path.write_text(full_text[:-3]) |
| 2022 | assert not path.read_text().endswith("\n") |
| 2023 | |
| 2024 | object_records._OIDX_CACHE.clear() |
| 2025 | listing = object_records.list_collection_records("widgets", base_dir=data_dir, roots=[])["records"] |
| 2026 | assert listing == [{"id": "w1", "name": "Alpha"}] |
| 2027 | |
| 2028 | object_records._OIDX_CACHE.clear() |
| 2029 | created = object_records.create_collection_record( |
| 2030 | "widgets", {"id": "w3", "name": "Gamma"}, base_dir=data_dir, roots=[] |
| 2031 | ) |
| 2032 | assert created == {"id": "w3", "name": "Gamma"} |
| 2033 | assert path.read_text().endswith("\n") |
| 2034 | |
| 2035 | object_records._OIDX_CACHE.clear() |
| 2036 | assert object_records.get_collection_record( |
| 2037 | "widgets", "w3", base_dir=data_dir, roots=[] |
| 2038 | ) == {"id": "w3", "name": "Gamma"} |
| 2039 | listing_after = object_records.list_collection_records( |
| 2040 | "widgets", base_dir=data_dir, roots=[] |
| 2041 | )["records"] |
| 2042 | assert listing_after == [{"id": "w1", "name": "Alpha"}, {"id": "w3", "name": "Gamma"}] |
| 2043 | |
| 2044 | |
| 2045 | def test_oidx_sidecar_resurrection_after_delete(tmp_path, monkeypatch): |
| 2046 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 2047 | data_dir = tmp_path / "data" |
| 2048 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2049 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 2050 | oidx_path = object_records._oidx_path(path) |
| 2051 | |
| 2052 | object_records.create_collection_record( |
| 2053 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 2054 | ) |
| 2055 | object_records.create_collection_record( |
| 2056 | "widgets", {"id": "w2", "name": "Beta"}, base_dir=data_dir, roots=[] |
| 2057 | ) |
| 2058 | assert oidx_path.exists() # built by w2's duplicate check |
| 2059 | |
| 2060 | object_records.delete_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 2061 | with pytest.raises(object_records.RecordNotFoundError): |
| 2062 | object_records.get_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 2063 | |
| 2064 | # Resurrect: creating "w1" again must succeed (not raise Duplicate), |
| 2065 | # resolved via the sidecar's absence-is-authoritative answer. |
| 2066 | resurrected = object_records.create_collection_record( |
| 2067 | "widgets", {"id": "w1", "name": "Gamma"}, base_dir=data_dir, roots=[] |
| 2068 | ) |
| 2069 | assert resurrected == {"id": "w1", "name": "Gamma"} |
| 2070 | assert object_records.get_collection_record( |
| 2071 | "widgets", "w1", base_dir=data_dir, roots=[] |
| 2072 | ) == {"id": "w1", "name": "Gamma"} |
| 2073 | |
| 2074 | |
| 2075 | def test_classic_mode_never_creates_oidx_sidecar(tmp_path, monkeypatch): |
| 2076 | """A classic-mode collection's records.tsv is never `_op`-columned, |
| 2077 | so _fast_record_lookup's sidecar branch is never reached -- confirm |
| 2078 | no `.records.oidx` ever appears, across create/update/delete/get/list, |
| 2079 | even with the cache forced cold on every single op.""" |
| 2080 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 2081 | data_dir = tmp_path / "data" |
| 2082 | write_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2083 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 2084 | oidx_path = object_records._oidx_path(path) |
| 2085 | |
| 2086 | for i in range(5): |
| 2087 | object_records.create_collection_record( |
| 2088 | "widgets", {"id": f"w{i}", "name": f"n{i}"}, base_dir=data_dir, roots=[] |
| 2089 | ) |
| 2090 | for i in range(5): |
| 2091 | object_records.update_collection_record( |
| 2092 | "widgets", f"w{i}", {"name": f"n{i}-updated"}, base_dir=data_dir, roots=[] |
| 2093 | ) |
| 2094 | object_records.delete_collection_record("widgets", "w0", base_dir=data_dir, roots=[]) |
| 2095 | for i in range(1, 5): |
| 2096 | object_records.get_collection_record("widgets", f"w{i}", base_dir=data_dir, roots=[]) |
| 2097 | object_records.list_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2098 | |
| 2099 | assert not oidx_path.exists() |
| 2100 | assert str(path.resolve()) not in object_records._OIDX_CACHE |
| 2101 | |
| 2102 | |
| 2103 | def test_append_mode_equivalent_to_classic_mode_with_sidecar_forced(tmp_path, monkeypatch): |
| 2104 | """Same equivalence property as |
| 2105 | test_append_mode_equivalent_to_classic_mode_across_random_operations, |
| 2106 | but with DBBASIC_RECORDS_CACHE_MAX_ROWS forced to 0 so every single |
| 2107 | op on both sides is a _RECORDS_CACHE miss -- meaning every append-side |
| 2108 | create/update/delete/get resolves through the id->offset sidecar |
| 2109 | (build, catch-up, or its full-rewrite fallback) for the entire 200-op |
| 2110 | sequence, never the warm-cache fast path.""" |
| 2111 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 2112 | classic_dir = tmp_path / "classic" |
| 2113 | append_dir = tmp_path / "append" |
| 2114 | write_schema(classic_dir, "items", [{"name": "id"}, {"name": "value"}]) |
| 2115 | write_append_schema(append_dir, "items", [{"name": "id"}, {"name": "value"}]) |
| 2116 | |
| 2117 | def apply(data_dir, kind, record_id, payload): |
| 2118 | try: |
| 2119 | if kind == "create": |
| 2120 | return ("ok", object_records.create_collection_record( |
| 2121 | "items", {"id": record_id, "value": payload}, base_dir=data_dir, roots=[] |
| 2122 | )) |
| 2123 | if kind == "update": |
| 2124 | return ("ok", object_records.update_collection_record( |
| 2125 | "items", record_id, {"value": payload}, base_dir=data_dir, roots=[] |
| 2126 | )) |
| 2127 | if kind == "delete": |
| 2128 | return ("ok", object_records.delete_collection_record( |
| 2129 | "items", record_id, base_dir=data_dir, roots=[] |
| 2130 | )) |
| 2131 | if kind == "get": |
| 2132 | return ("ok", object_records.get_collection_record( |
| 2133 | "items", record_id, base_dir=data_dir, roots=[] |
| 2134 | )) |
| 2135 | return ("ok", object_records.list_collection_records( |
| 2136 | "items", base_dir=data_dir, roots=[] |
| 2137 | )) |
| 2138 | except Exception as exc: # noqa: BLE001 -- comparing exception SHAPES across both runs |
| 2139 | return ("error", type(exc), str(exc)) |
| 2140 | |
| 2141 | rng = random.Random(20260718) |
| 2142 | id_pool = [f"e{i}" for i in range(12)] |
| 2143 | |
| 2144 | for step in range(200): |
| 2145 | kind = rng.choice(["create", "update", "delete", "get", "list"]) |
| 2146 | record_id = rng.choice(id_pool) |
| 2147 | payload = f"v{step}" |
| 2148 | |
| 2149 | classic_outcome = apply(classic_dir, kind, record_id, payload) |
| 2150 | append_outcome = apply(append_dir, kind, record_id, payload) |
| 2151 | |
| 2152 | assert classic_outcome == append_outcome, ( |
| 2153 | f"step {step} op={kind!r} id={record_id!r}: " |
| 2154 | f"classic={classic_outcome!r} append={append_outcome!r}" |
| 2155 | ) |
| 2156 | |
| 2157 | classic_listing = object_records.list_collection_records("items", base_dir=classic_dir, roots=[]) |
| 2158 | append_listing = object_records.list_collection_records("items", base_dir=append_dir, roots=[]) |
| 2159 | assert classic_listing == append_listing |
| 2160 | |
| 2161 | |
| 2162 | # --------------------------------------------------------------------------- |
| 2163 | # Records-cache tail-delta incremental read (docs/append-only-storage- |
| 2164 | # design.md Sidecars, bullet 2). Every test below writes external append |
| 2165 | # rows directly with append_physical_rows -- bypassing object_records' |
| 2166 | # API and its own cache-refresh calls entirely -- to simulate a SEPARATE |
| 2167 | # writer (another process, the CLI, a sibling pool worker) growing an |
| 2168 | # append-mode file underneath a cache entry this process already has warm. |
| 2169 | # --------------------------------------------------------------------------- |
| 2170 | |
| 2171 | |
| 2172 | def test_incremental_cache_folds_external_append_without_full_reparse(tmp_path, monkeypatch): |
| 2173 | """A warm cache entry for an append-mode file, grown in place by an |
| 2174 | external writer (same inode, pure growth), must be caught up via the |
| 2175 | tail-delta incremental fold -- not a full re-parse of the whole file. |
| 2176 | Proven two ways: the module's incremental-fold counter advances by |
| 2177 | exactly one, and _parse_records_file is monkeypatched to blow up if |
| 2178 | it's ever called from this point on.""" |
| 2179 | data_dir = tmp_path / "data" |
| 2180 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2181 | object_records.create_collection_record( |
| 2182 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 2183 | ) |
| 2184 | warm = object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2185 | assert warm == [{"id": "w1", "name": "Alpha"}] |
| 2186 | |
| 2187 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 2188 | inode_before = path.stat().st_ino |
| 2189 | |
| 2190 | def _boom(*_args, **_kwargs): |
| 2191 | raise AssertionError( |
| 2192 | "_parse_records_file ran: the incremental tail-delta path did not take effect" |
| 2193 | ) |
| 2194 | |
| 2195 | monkeypatch.setattr(object_records, "_parse_records_file", _boom) |
| 2196 | |
| 2197 | append_physical_rows( |
| 2198 | path, |
| 2199 | [object_records.OP_FIELD, "id", "name"], |
| 2200 | [{"id": "w2", "name": "Beta", object_records.OP_FIELD: ""}], |
| 2201 | ) |
| 2202 | assert path.stat().st_ino == inode_before # confirms this was growth-in-place, not a rewrite |
| 2203 | |
| 2204 | before_folds = object_records._INCREMENTAL_CACHE_FOLDS |
| 2205 | result = object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2206 | assert result == [{"id": "w1", "name": "Alpha"}, {"id": "w2", "name": "Beta"}] |
| 2207 | assert object_records._INCREMENTAL_CACHE_FOLDS == before_folds + 1 |
| 2208 | |
| 2209 | |
| 2210 | def test_incremental_cache_folds_external_update_at_same_position(tmp_path): |
| 2211 | """An externally appended upsert for an id already present must |
| 2212 | replace it AT ITS EXISTING POSITION (matching a full fold's |
| 2213 | last-wins-in-place semantics -- see _fold_append_rows), not move it |
| 2214 | to the end. Cross-checked against a forced cold full parse of the |
| 2215 | identical on-disk content.""" |
| 2216 | data_dir = tmp_path / "data" |
| 2217 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2218 | for i in range(3): |
| 2219 | object_records.create_collection_record( |
| 2220 | "widgets", {"id": f"w{i}", "name": f"n{i}"}, base_dir=data_dir, roots=[] |
| 2221 | ) |
| 2222 | warm = object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2223 | assert [record["id"] for record in warm] == ["w0", "w1", "w2"] |
| 2224 | |
| 2225 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 2226 | before_folds = object_records._INCREMENTAL_CACHE_FOLDS |
| 2227 | append_physical_rows( |
| 2228 | path, |
| 2229 | [object_records.OP_FIELD, "id", "name"], |
| 2230 | [{"id": "w1", "name": "n1-updated", object_records.OP_FIELD: ""}], |
| 2231 | ) |
| 2232 | |
| 2233 | result = object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2234 | assert object_records._INCREMENTAL_CACHE_FOLDS == before_folds + 1 |
| 2235 | assert result == [ |
| 2236 | {"id": "w0", "name": "n0"}, |
| 2237 | {"id": "w1", "name": "n1-updated"}, |
| 2238 | {"id": "w2", "name": "n2"}, |
| 2239 | ] |
| 2240 | |
| 2241 | # Cross-check against an independent cold full parse of the exact same |
| 2242 | # on-disk bytes -- the cache is untouched by this call. |
| 2243 | _, cold_records = object_records._parse_records_file("widgets", path) |
| 2244 | assert cold_records == result |
| 2245 | |
| 2246 | |
| 2247 | def test_incremental_cache_falls_back_to_full_fold_on_tombstone(tmp_path): |
| 2248 | """A delta containing ANY tombstone must bail out of the incremental |
| 2249 | merge entirely and fall back to a full fold (deletes shift every |
| 2250 | later position, which the cheap replace-or-append merge does not |
| 2251 | model) -- but the RESULT must still be correct, and the fold counter |
| 2252 | must show the incremental path did NOT run.""" |
| 2253 | data_dir = tmp_path / "data" |
| 2254 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2255 | for i in range(3): |
| 2256 | object_records.create_collection_record( |
| 2257 | "widgets", {"id": f"w{i}", "name": f"n{i}"}, base_dir=data_dir, roots=[] |
| 2258 | ) |
| 2259 | object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) # warm it |
| 2260 | |
| 2261 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 2262 | before_folds = object_records._INCREMENTAL_CACHE_FOLDS |
| 2263 | append_physical_rows( |
| 2264 | path, |
| 2265 | [object_records.OP_FIELD, "id", "name"], |
| 2266 | [{"id": "w1", "name": "", object_records.OP_FIELD: object_records.OP_DELETE}], |
| 2267 | ) |
| 2268 | |
| 2269 | result = object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2270 | assert object_records._INCREMENTAL_CACHE_FOLDS == before_folds # did NOT take the fast path |
| 2271 | assert result == [{"id": "w0", "name": "n0"}, {"id": "w2", "name": "n2"}] |
| 2272 | |
| 2273 | |
| 2274 | def test_incremental_cache_ignores_torn_tail_fragment_in_delta(tmp_path): |
| 2275 | """A torn (unterminated) fragment at EOF -- the classic log torn-tail |
| 2276 | case, from a writer that crashed mid-row -- must be excluded from the |
| 2277 | incremental fold's delta, exactly as a full fold would exclude it, |
| 2278 | and must never resurface once a later, legitimate write repairs and |
| 2279 | extends the file.""" |
| 2280 | data_dir = tmp_path / "data" |
| 2281 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2282 | object_records.create_collection_record( |
| 2283 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[] |
| 2284 | ) |
| 2285 | object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) # warm it |
| 2286 | |
| 2287 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 2288 | append_physical_rows( |
| 2289 | path, |
| 2290 | [object_records.OP_FIELD, "id", "name"], |
| 2291 | [{"id": "w2", "name": "Beta", object_records.OP_FIELD: ""}], |
| 2292 | ) |
| 2293 | # A crash mid-write: a fragment with no trailing newline. |
| 2294 | with path.open("a", newline="") as handle: |
| 2295 | handle.write("\tw3\tGam") |
| 2296 | assert not path.read_text().endswith("\n") |
| 2297 | |
| 2298 | result = object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2299 | assert result == [{"id": "w1", "name": "Alpha"}, {"id": "w2", "name": "Beta"}] |
| 2300 | |
| 2301 | # A subsequent legitimate write must self-heal (truncate the fragment |
| 2302 | # away) and append cleanly -- the fragment must never come back. |
| 2303 | created = object_records.create_collection_record( |
| 2304 | "widgets", {"id": "w3", "name": "Gamma"}, base_dir=data_dir, roots=[] |
| 2305 | ) |
| 2306 | assert created == {"id": "w3", "name": "Gamma"} |
| 2307 | assert path.read_text().endswith("\n") |
| 2308 | listing = object_records.list_collection_records( |
| 2309 | "widgets", base_dir=data_dir, roots=[] |
| 2310 | )["records"] |
| 2311 | assert listing == [ |
| 2312 | {"id": "w1", "name": "Alpha"}, |
| 2313 | {"id": "w2", "name": "Beta"}, |
| 2314 | {"id": "w3", "name": "Gamma"}, |
| 2315 | ] |
| 2316 | |
| 2317 | |
| 2318 | def test_incremental_cache_evicts_when_growth_crosses_row_threshold(tmp_path, monkeypatch): |
| 2319 | """A successful incremental fold that pushes a collection's row count |
| 2320 | past DBBASIC_RECORDS_CACHE_MAX_ROWS must still return the correct |
| 2321 | (larger) result, but must NOT leave an entry cached -- same rule the |
| 2322 | full-parse path already applies (test_records_cache_row_threshold_ |
| 2323 | excludes_large_collections).""" |
| 2324 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "3") |
| 2325 | data_dir = tmp_path / "data" |
| 2326 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2327 | for i in range(3): |
| 2328 | object_records.create_collection_record( |
| 2329 | "widgets", {"id": f"w{i}", "name": f"n{i}"}, base_dir=data_dir, roots=[] |
| 2330 | ) |
| 2331 | object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2332 | cache_key = _cache_key("widgets", data_dir) |
| 2333 | assert cache_key in object_records._RECORDS_CACHE |
| 2334 | |
| 2335 | path = data_dir / "collections" / "widgets" / "records.tsv" |
| 2336 | append_physical_rows( |
| 2337 | path, |
| 2338 | [object_records.OP_FIELD, "id", "name"], |
| 2339 | [ |
| 2340 | {"id": "w3", "name": "n3", object_records.OP_FIELD: ""}, |
| 2341 | {"id": "w4", "name": "n4", object_records.OP_FIELD: ""}, |
| 2342 | ], |
| 2343 | ) |
| 2344 | |
| 2345 | result = object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2346 | assert len(result) == 5 |
| 2347 | assert cache_key not in object_records._RECORDS_CACHE |
| 2348 | |
| 2349 | # Must still reparse correctly on the very next read (uncached, but |
| 2350 | # correct every time). |
| 2351 | result_again = object_records.read_collection_records("widgets", base_dir=data_dir, roots=[]) |
| 2352 | assert result_again == result |
| 2353 | |
| 2354 | |
| 2355 | def test_incremental_cache_equivalence_across_random_external_append_batches(tmp_path): |
| 2356 | """PROPERTY: for a long, seeded, randomized sequence of external |
| 2357 | append batches (each batch a handful of upserts and occasional |
| 2358 | deletes, written directly to disk exactly as a separate process |
| 2359 | would), the WARM read path -- whichever of incremental-fold or |
| 2360 | fallback-full-fold it takes -- must always agree with an independent |
| 2361 | COLD full parse of the identical on-disk bytes. ~50 iterations.""" |
| 2362 | data_dir = tmp_path / "data" |
| 2363 | write_append_schema(data_dir, "items", [{"name": "id"}, {"name": "value"}]) |
| 2364 | path = data_dir / "collections" / "items" / "records.tsv" |
| 2365 | physical_fields = [object_records.OP_FIELD, "id", "value"] |
| 2366 | |
| 2367 | object_records.create_collection_record( |
| 2368 | "items", {"id": "seed", "value": "0"}, base_dir=data_dir, roots=[] |
| 2369 | ) |
| 2370 | object_records.read_collection_records("items", base_dir=data_dir, roots=[]) # warm once |
| 2371 | |
| 2372 | rng = random.Random(20260718) |
| 2373 | id_pool = [f"e{i}" for i in range(8)] |
| 2374 | |
| 2375 | for iteration in range(50): |
| 2376 | batch = [] |
| 2377 | for _ in range(rng.randint(1, 4)): |
| 2378 | record_id = rng.choice(id_pool) |
| 2379 | if rng.random() < 0.15: |
| 2380 | batch.append({ |
| 2381 | "id": record_id, |
| 2382 | "value": "", |
| 2383 | object_records.OP_FIELD: object_records.OP_DELETE, |
| 2384 | }) |
| 2385 | else: |
| 2386 | batch.append({ |
| 2387 | "id": record_id, |
| 2388 | "value": f"v{iteration}", |
| 2389 | object_records.OP_FIELD: "", |
| 2390 | }) |
| 2391 | append_physical_rows(path, physical_fields, batch) |
| 2392 | |
| 2393 | warm_result = object_records.read_collection_records("items", base_dir=data_dir, roots=[]) |
| 2394 | _, cold_records = object_records._parse_records_file("items", path) |
| 2395 | assert warm_result == cold_records, ( |
| 2396 | f"iteration {iteration}: warm={warm_result!r} cold={cold_records!r}" |
| 2397 | ) |
| 2398 | |
| 2399 | |
| 2400 | # --- Compaction observability (docs/append-only-storage-design.md |
| 2401 | # "Compaction", docs/storage-modes.md) --------------------------------------- |
| 2402 | |
| 2403 | |
| 2404 | def _seed_widgets(data_dir: Path) -> None: |
| 2405 | """5 creates, 1 update (supersedes), 1 delete (tombstone): 7 physical |
| 2406 | rows, 4 live (w0 updated, w1 deleted -> 3 remain plus updated w0), 3 |
| 2407 | dead. Exact numbers every stats test below relies on.""" |
| 2408 | for i in range(5): |
| 2409 | object_records.create_collection_record( |
| 2410 | "widgets", {"id": f"w{i}", "name": f"n{i}"}, base_dir=data_dir, roots=[] |
| 2411 | ) |
| 2412 | object_records.update_collection_record( |
| 2413 | "widgets", "w0", {"name": "n0-updated"}, base_dir=data_dir, roots=[] |
| 2414 | ) |
| 2415 | object_records.delete_collection_record("widgets", "w1", base_dir=data_dir, roots=[]) |
| 2416 | |
| 2417 | |
| 2418 | def test_append_collection_stats_is_none_for_classic_collections(tmp_path): |
| 2419 | data_dir = tmp_path / "data" |
| 2420 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 2421 | assert object_records.append_collection_stats("contacts", base_dir=data_dir, roots=[]) is None |
| 2422 | |
| 2423 | |
| 2424 | def test_append_collection_stats_is_none_for_schema_with_no_storage_key(tmp_path): |
| 2425 | data_dir = tmp_path / "data" |
| 2426 | write_schema(data_dir, "contacts", [{"name": "id"}, {"name": "name"}]) |
| 2427 | object_records.create_collection_record( |
| 2428 | "contacts", {"id": "c1", "name": "Ada"}, base_dir=data_dir, roots=[] |
| 2429 | ) |
| 2430 | assert object_records.append_collection_stats("contacts", base_dir=data_dir, roots=[]) is None |
| 2431 | |
| 2432 | |
| 2433 | def test_append_collection_stats_for_never_written_collection_is_honest_zeros(tmp_path): |
| 2434 | data_dir = tmp_path / "data" |
| 2435 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2436 | assert object_records.append_collection_stats("widgets", base_dir=data_dir, roots=[]) == { |
| 2437 | "collection": "widgets", |
| 2438 | "storage": "append", |
| 2439 | "physical_rows": 0, |
| 2440 | "live_rows": 0, |
| 2441 | "dead_rows": 0, |
| 2442 | "bloat_ratio": 0.0, |
| 2443 | "file_bytes": 0, |
| 2444 | "sidecar_present": False, |
| 2445 | "compaction_flagged": False, |
| 2446 | } |
| 2447 | |
| 2448 | |
| 2449 | def test_append_collection_stats_via_warm_records_cache(tmp_path): |
| 2450 | """Default cache threshold: the seeded collection stays warm in |
| 2451 | _RECORDS_CACHE, so stats must be served from it (no sidecar |
| 2452 | involved).""" |
| 2453 | data_dir = tmp_path / "data" |
| 2454 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2455 | _seed_widgets(data_dir) |
| 2456 | |
| 2457 | path = object_records.collection_records_file("widgets", base_dir=data_dir) |
| 2458 | cache_key = str(path.resolve()) |
| 2459 | assert cache_key in object_records._RECORDS_CACHE # warm precondition |
| 2460 | |
| 2461 | stats = object_records.append_collection_stats("widgets", base_dir=data_dir, roots=[]) |
| 2462 | assert stats == { |
| 2463 | "collection": "widgets", |
| 2464 | "storage": "append", |
| 2465 | "physical_rows": 7, |
| 2466 | "live_rows": 4, |
| 2467 | "dead_rows": 3, |
| 2468 | "bloat_ratio": 0.75, |
| 2469 | "file_bytes": path.stat().st_size, |
| 2470 | "sidecar_present": False, |
| 2471 | "compaction_flagged": False, |
| 2472 | } |
| 2473 | |
| 2474 | |
| 2475 | def test_append_collection_stats_via_sidecar_with_cold_cache(tmp_path, monkeypatch): |
| 2476 | """DBBASIC_RECORDS_CACHE_MAX_ROWS=0 forces every point op through the |
| 2477 | id->offset sidecar instead of _RECORDS_CACHE (same technique the oidx |
| 2478 | test block above this one uses) -- stats must then be answered by a |
| 2479 | cheap sidecar line-count read, not a full fold.""" |
| 2480 | monkeypatch.setenv("DBBASIC_RECORDS_CACHE_MAX_ROWS", "0") |
| 2481 | data_dir = tmp_path / "data" |
| 2482 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2483 | _seed_widgets(data_dir) |
| 2484 | |
| 2485 | path = object_records.collection_records_file("widgets", base_dir=data_dir) |
| 2486 | cache_key = str(path.resolve()) |
| 2487 | assert cache_key not in object_records._RECORDS_CACHE # cold precondition |
| 2488 | oidx_path = object_records._oidx_path(path) |
| 2489 | assert oidx_path.exists() # built lazily by the point ops above |
| 2490 | |
| 2491 | # Belt and suspenders: an actual full fold must not run for this case. |
| 2492 | def _boom(*args, **kwargs): |
| 2493 | raise AssertionError("full fold ran when the sidecar should have answered") |
| 2494 | |
| 2495 | monkeypatch.setattr(object_records, "_parse_append_body", _boom) |
| 2496 | |
| 2497 | stats = object_records.append_collection_stats("widgets", base_dir=data_dir, roots=[]) |
| 2498 | assert stats == { |
| 2499 | "collection": "widgets", |
| 2500 | "storage": "append", |
| 2501 | "physical_rows": 7, |
| 2502 | "live_rows": 4, |
| 2503 | "dead_rows": 3, |
| 2504 | "bloat_ratio": 0.75, |
| 2505 | "file_bytes": path.stat().st_size, |
| 2506 | "sidecar_present": True, |
| 2507 | "compaction_flagged": False, |
| 2508 | } |
| 2509 | |
| 2510 | |
| 2511 | def test_append_collection_stats_falls_back_to_full_fold_when_allowed(tmp_path, monkeypatch): |
| 2512 | """Neither cache nor sidecar available (sidecar never built: default |
| 2513 | cache threshold means no point op ever missed _RECORDS_CACHE) -- |
| 2514 | allow_fold=True (the default) must still produce the correct numbers |
| 2515 | via a full parse.""" |
| 2516 | data_dir = tmp_path / "data" |
| 2517 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2518 | _seed_widgets(data_dir) |
| 2519 | |
| 2520 | path = object_records.collection_records_file("widgets", base_dir=data_dir) |
| 2521 | oidx_path = object_records._oidx_path(path) |
| 2522 | assert not oidx_path.exists() # never built: cache always answered point ops |
| 2523 | object_records._RECORDS_CACHE.clear() |
| 2524 | |
| 2525 | stats = object_records.append_collection_stats("widgets", base_dir=data_dir, roots=[]) |
| 2526 | assert stats == { |
| 2527 | "collection": "widgets", |
| 2528 | "storage": "append", |
| 2529 | "physical_rows": 7, |
| 2530 | "live_rows": 4, |
| 2531 | "dead_rows": 3, |
| 2532 | "bloat_ratio": 0.75, |
| 2533 | "file_bytes": path.stat().st_size, |
| 2534 | "sidecar_present": False, |
| 2535 | "compaction_flagged": False, |
| 2536 | } |
| 2537 | |
| 2538 | |
| 2539 | def test_append_collection_stats_allow_fold_false_returns_estimated_without_folding(tmp_path): |
| 2540 | """allow_fold=False must never pay for a full parse: with the warm |
| 2541 | cache dropped and no sidecar built, it returns an "estimated" entry |
| 2542 | (file_bytes only, physical/live/dead/bloat null) instead of folding.""" |
| 2543 | data_dir = tmp_path / "data" |
| 2544 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2545 | _seed_widgets(data_dir) |
| 2546 | object_records._RECORDS_CACHE.clear() |
| 2547 | |
| 2548 | path = object_records.collection_records_file("widgets", base_dir=data_dir) |
| 2549 | assert not object_records._oidx_path(path).exists() |
| 2550 | |
| 2551 | calls = [] |
| 2552 | original = object_records._parse_append_body |
| 2553 | |
| 2554 | def _tracking(*args, **kwargs): |
| 2555 | calls.append(True) |
| 2556 | return original(*args, **kwargs) |
| 2557 | |
| 2558 | with unittest.mock.patch.object(object_records, "_parse_append_body", _tracking): |
| 2559 | stats = object_records.append_collection_stats( |
| 2560 | "widgets", base_dir=data_dir, roots=[], allow_fold=False |
| 2561 | ) |
| 2562 | |
| 2563 | assert calls == [] # sentinel: no full parse ran |
| 2564 | assert stats == { |
| 2565 | "collection": "widgets", |
| 2566 | "storage": "append", |
| 2567 | "physical_rows": None, |
| 2568 | "live_rows": None, |
| 2569 | "dead_rows": None, |
| 2570 | "bloat_ratio": None, |
| 2571 | "file_bytes": path.stat().st_size, |
| 2572 | "sidecar_present": False, |
| 2573 | "compaction_flagged": False, |
| 2574 | "estimated": True, |
| 2575 | } |
| 2576 | |
| 2577 | |
| 2578 | def test_append_collection_stats_reflects_compaction_flag(tmp_path, monkeypatch): |
| 2579 | monkeypatch.setenv("DBBASIC_APPEND_COMPACT_MIN_ROWS", "5") |
| 2580 | data_dir = tmp_path / "data" |
| 2581 | write_append_schema(data_dir, "churn", [{"name": "id"}, {"name": "value"}]) |
| 2582 | |
| 2583 | object_records.create_collection_record( |
| 2584 | "churn", {"id": "c1", "value": "v0"}, base_dir=data_dir, roots=[] |
| 2585 | ) |
| 2586 | for i in range(10): |
| 2587 | object_records.update_collection_record( |
| 2588 | "churn", "c1", {"value": f"v{i}"}, base_dir=data_dir, roots=[] |
| 2589 | ) |
| 2590 | |
| 2591 | path = object_records.collection_records_file("churn", base_dir=data_dir) |
| 2592 | cache_key = str(path.resolve()) |
| 2593 | object_records._RECORDS_CACHE.pop(cache_key, None) # force a cold parse to flag |
| 2594 | object_records._read_collection_records("churn", base_dir=data_dir) |
| 2595 | assert cache_key in object_records._PENDING_COMPACTION |
| 2596 | |
| 2597 | stats = object_records.append_collection_stats("churn", base_dir=data_dir, roots=[]) |
| 2598 | assert stats["compaction_flagged"] is True |
| 2599 | |
| 2600 | |
| 2601 | def test_append_collection_stats_after_compaction_is_bloat_free(tmp_path): |
| 2602 | data_dir = tmp_path / "data" |
| 2603 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2604 | _seed_widgets(data_dir) |
| 2605 | |
| 2606 | object_records.compact_collection("widgets", base_dir=data_dir, roots=[]) |
| 2607 | |
| 2608 | stats = object_records.append_collection_stats("widgets", base_dir=data_dir, roots=[]) |
| 2609 | assert stats["physical_rows"] == 4 |
| 2610 | assert stats["live_rows"] == 4 |
| 2611 | assert stats["dead_rows"] == 0 |
| 2612 | assert stats["bloat_ratio"] == 0.0 |
| 2613 | |
| 2614 | |
| 2615 | def test_list_append_collection_stats_only_includes_append_mode_collections(tmp_path): |
| 2616 | data_dir = tmp_path / "data" |
| 2617 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2618 | _seed_widgets(data_dir) |
| 2619 | write_schema(data_dir, "contacts", [{"name": "id"}, {"name": "name"}]) |
| 2620 | object_records.create_collection_record( |
| 2621 | "contacts", {"id": "c1", "name": "Ada"}, base_dir=data_dir, roots=[] |
| 2622 | ) |
| 2623 | |
| 2624 | stats = object_records.list_append_collection_stats(base_dir=data_dir, roots=[]) |
| 2625 | assert [entry["collection"] for entry in stats] == ["widgets"] |
| 2626 | assert stats[0]["physical_rows"] == 7 |
| 2627 | |
| 2628 | |
| 2629 | def test_list_append_collection_stats_skips_missing_collection_records_file(tmp_path): |
| 2630 | """A schema declaring storage:append with no records.tsv yet must show |
| 2631 | up with zeroed stats, not be skipped or raise.""" |
| 2632 | data_dir = tmp_path / "data" |
| 2633 | write_append_schema(data_dir, "empties", [{"name": "id"}]) |
| 2634 | |
| 2635 | stats = object_records.list_append_collection_stats(base_dir=data_dir, roots=[]) |
| 2636 | assert len(stats) == 1 |
| 2637 | assert stats[0]["collection"] == "empties" |
| 2638 | assert stats[0]["physical_rows"] == 0 |
| 2639 | |
| 2640 | |
| 2641 | def test_list_append_collection_stats_empty_when_no_append_collections(tmp_path): |
| 2642 | data_dir = tmp_path / "data" |
| 2643 | write_schema(data_dir, "contacts", [{"name": "id"}]) |
| 2644 | object_records.create_collection_record( |
| 2645 | "contacts", {"id": "c1"}, base_dir=data_dir, roots=[] |
| 2646 | ) |
| 2647 | assert object_records.list_append_collection_stats(base_dir=data_dir, roots=[]) == [] |
| 2648 | |
| 2649 | |
| 2650 | # --- Universal attribution: every write emits exactly one attributed change --- |
| 2651 | # (create/update/delete_collection_record now emit via object_record_changes |
| 2652 | # themselves, regardless of caller -- HTTP API, CLI, package installs, etc.) |
| 2653 | |
| 2654 | |
| 2655 | def test_create_collection_record_with_no_actor_emits_unattributed_change(tmp_path): |
| 2656 | data_dir = tmp_path / "data" |
| 2657 | write_records(data_dir, "contacts", "id\tname\n") |
| 2658 | |
| 2659 | object_records.create_collection_record( |
| 2660 | "contacts", {"id": "c1", "name": "Ada"}, base_dir=data_dir, roots=[] |
| 2661 | ) |
| 2662 | |
| 2663 | changes = object_record_changes.list_record_changes("contacts", base_dir=data_dir)["changes"] |
| 2664 | assert len(changes) == 1 |
| 2665 | assert changes[0]["actor"] == "unattributed" |
| 2666 | assert changes[0]["action"] == "create" |
| 2667 | assert changes[0]["record_id"] == "c1" |
| 2668 | assert changes[0]["before"] is None |
| 2669 | assert changes[0]["after"] == {"id": "c1", "name": "Ada"} |
| 2670 | |
| 2671 | |
| 2672 | def test_create_collection_record_with_actor_records_it(tmp_path): |
| 2673 | data_dir = tmp_path / "data" |
| 2674 | write_records(data_dir, "contacts", "id\tname\n") |
| 2675 | |
| 2676 | object_records.create_collection_record( |
| 2677 | "contacts", {"id": "c1", "name": "Ada"}, base_dir=data_dir, roots=[], actor="x" |
| 2678 | ) |
| 2679 | |
| 2680 | changes = object_record_changes.list_record_changes("contacts", base_dir=data_dir)["changes"] |
| 2681 | assert len(changes) == 1 |
| 2682 | assert changes[0]["actor"] == "x" |
| 2683 | |
| 2684 | |
| 2685 | def test_update_collection_record_with_no_actor_emits_unattributed_change(tmp_path): |
| 2686 | data_dir = tmp_path / "data" |
| 2687 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 2688 | |
| 2689 | object_records.update_collection_record( |
| 2690 | "contacts", "c1", {"name": "Ada Lovelace"}, base_dir=data_dir, roots=[] |
| 2691 | ) |
| 2692 | |
| 2693 | changes = object_record_changes.list_record_changes("contacts", base_dir=data_dir)["changes"] |
| 2694 | assert len(changes) == 1 |
| 2695 | assert changes[0]["actor"] == "unattributed" |
| 2696 | assert changes[0]["action"] == "update" |
| 2697 | assert changes[0]["before"] == {"id": "c1", "name": "Ada"} |
| 2698 | assert changes[0]["after"] == {"id": "c1", "name": "Ada Lovelace"} |
| 2699 | |
| 2700 | |
| 2701 | def test_update_collection_record_with_actor_records_it(tmp_path): |
| 2702 | data_dir = tmp_path / "data" |
| 2703 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 2704 | |
| 2705 | object_records.update_collection_record( |
| 2706 | "contacts", "c1", {"name": "Ada Lovelace"}, base_dir=data_dir, roots=[], actor="x" |
| 2707 | ) |
| 2708 | |
| 2709 | changes = object_record_changes.list_record_changes("contacts", base_dir=data_dir)["changes"] |
| 2710 | assert len(changes) == 1 |
| 2711 | assert changes[0]["actor"] == "x" |
| 2712 | |
| 2713 | |
| 2714 | def test_delete_collection_record_with_no_actor_emits_unattributed_change(tmp_path): |
| 2715 | data_dir = tmp_path / "data" |
| 2716 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 2717 | |
| 2718 | object_records.delete_collection_record("contacts", "c1", base_dir=data_dir, roots=[]) |
| 2719 | |
| 2720 | changes = object_record_changes.list_record_changes("contacts", base_dir=data_dir)["changes"] |
| 2721 | assert len(changes) == 1 |
| 2722 | assert changes[0]["actor"] == "unattributed" |
| 2723 | assert changes[0]["action"] == "delete" |
| 2724 | assert changes[0]["before"] == {"id": "c1", "name": "Ada"} |
| 2725 | assert changes[0]["after"] is None |
| 2726 | |
| 2727 | |
| 2728 | def test_delete_collection_record_with_actor_records_it(tmp_path): |
| 2729 | data_dir = tmp_path / "data" |
| 2730 | write_records(data_dir, "contacts", "id\tname\nc1\tAda\n") |
| 2731 | |
| 2732 | object_records.delete_collection_record("contacts", "c1", base_dir=data_dir, roots=[], actor="x") |
| 2733 | |
| 2734 | changes = object_record_changes.list_record_changes("contacts", base_dir=data_dir)["changes"] |
| 2735 | assert len(changes) == 1 |
| 2736 | assert changes[0]["actor"] == "x" |
| 2737 | |
| 2738 | |
| 2739 | def test_compaction_does_not_emit_a_change(tmp_path): |
| 2740 | """Compaction is a physical-only rewrite -- it must never emit a change.""" |
| 2741 | data_dir = tmp_path / "data" |
| 2742 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2743 | object_records.create_collection_record( |
| 2744 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[], actor="x" |
| 2745 | ) |
| 2746 | object_records.update_collection_record( |
| 2747 | "widgets", "w1", {"name": "Alpha2"}, base_dir=data_dir, roots=[], actor="x" |
| 2748 | ) |
| 2749 | before = object_record_changes.list_record_changes("widgets", base_dir=data_dir)["total"] |
| 2750 | |
| 2751 | object_records.compact_collection("widgets", base_dir=data_dir, roots=[]) |
| 2752 | |
| 2753 | after = object_record_changes.list_record_changes("widgets", base_dir=data_dir)["total"] |
| 2754 | assert after == before |
| 2755 | |
| 2756 | |
| 2757 | def test_append_mode_create_update_delete_attribution(tmp_path): |
| 2758 | """Universal attribution also applies to append-storage collections.""" |
| 2759 | data_dir = tmp_path / "data" |
| 2760 | write_append_schema(data_dir, "widgets", [{"name": "id"}, {"name": "name"}]) |
| 2761 | |
| 2762 | object_records.create_collection_record( |
| 2763 | "widgets", {"id": "w1", "name": "Alpha"}, base_dir=data_dir, roots=[], actor="seed" |
| 2764 | ) |
| 2765 | object_records.update_collection_record( |
| 2766 | "widgets", "w1", {"name": "Alpha2"}, base_dir=data_dir, roots=[] |
| 2767 | ) |
| 2768 | object_records.delete_collection_record("widgets", "w1", base_dir=data_dir, roots=[], actor="cleanup") |
| 2769 | |
| 2770 | changes = object_record_changes.list_record_changes("widgets", base_dir=data_dir)["changes"] |
| 2771 | assert [change["action"] for change in changes] == ["delete", "update", "create"] |
| 2772 | assert changes[2]["actor"] == "seed" |
| 2773 | assert changes[1]["actor"] == "unattributed" |
| 2774 | assert changes[0]["actor"] == "cleanup" |
| 2775 | |
| 2776 | |
| 2777 | def test_record_write_emits_exactly_one_change_no_recursion(tmp_path, monkeypatch): |
| 2778 | """A record write must emit its own change exactly once -- and must never |
| 2779 | |
| 2780 | trigger a change-about-a-change: object_record_changes.append_record_change |
| 2781 | writes its JSONL entry directly (it never calls back into object_records' |
| 2782 | create/update/delete_collection_record), so there is no path for one write |
| 2783 | to recursively emit a second, unbounded chain of changes. |
| 2784 | """ |
| 2785 | data_dir = tmp_path / "data" |
| 2786 | write_records(data_dir, "contacts", "id\tname\n") |
| 2787 | |
| 2788 | calls: list[str] = [] |
| 2789 | original = object_record_changes.append_record_change |
| 2790 | |
| 2791 | def counting(*args, **kwargs): |
| 2792 | calls.append(kwargs.get("collection")) |
| 2793 | return original(*args, **kwargs) |
| 2794 | |
| 2795 | monkeypatch.setattr(object_record_changes, "append_record_change", counting) |
| 2796 | |
| 2797 | object_records.create_collection_record( |
| 2798 | "contacts", {"id": "c1", "name": "Ada"}, base_dir=data_dir, roots=[], actor="x" |
| 2799 | ) |
| 2800 | |
| 2801 | # Exactly one emission for one write -- not zero, not two, not a chain. |
| 2802 | assert calls == ["contacts"] |
| 2803 | # The change log itself is never stored as a queryable collection record, |
| 2804 | # so there is nothing for a write to recurse into in the first place. |
| 2805 | assert not (data_dir / "collections" / "record_changes").exists() |
| 2806 | |
| 2807 | |
| 2808 | def test_record_change_entry_shape_is_unchanged(tmp_path): |
| 2809 | """Backward compat: the stored change entry keeps its original shape |
| 2810 | whether emission is triggered from the write path or from a direct |
| 2811 | object_record_changes.append_record_change call. |
| 2812 | """ |
| 2813 | data_dir = tmp_path / "data" |
| 2814 | write_records(data_dir, "contacts", "id\tname\n") |
| 2815 | |
| 2816 | object_records.create_collection_record( |
| 2817 | "contacts", {"id": "c1", "name": "Ada"}, base_dir=data_dir, roots=[], actor="x" |
| 2818 | ) |
| 2819 | |
| 2820 | entry = object_record_changes.list_record_changes("contacts", base_dir=data_dir)["changes"][0] |
| 2821 | assert set(entry) == { |
| 2822 | "change_id", |
| 2823 | "timestamp", |
| 2824 | "collection", |
| 2825 | "record_id", |
| 2826 | "action", |
| 2827 | "actor", |
| 2828 | "message", |
| 2829 | "correlation_id", |
| 2830 | "changed_fields", |
| 2831 | "before", |
| 2832 | "after", |
| 2833 | } |
| 2834 | UUID(entry["change_id"]) |
| 2835 | |
| 2836 | |
| 2837 | # --- 58 query/filter: filter_records, list_collection_records(where=...), --- |
| 2838 | # --- normalize_filter_value -------------------------------------------- |
| 2839 | |
| 2840 | |
| 2841 | def cond(op, value): |
| 2842 | return object_permissions.filter_condition(op, value) |
| 2843 | |
| 2844 | |
| 2845 | CONTACTS_TSV = ( |
| 2846 | "id\tname\tstatus\towner_id\ttotal\tcreated_at\n" |
| 2847 | "c1\tAda\thot\t7\t150\t2026-07-01\n" |
| 2848 | "c2\tGrace\tcold\t7\t50\t2026-07-15\n" |
| 2849 | "c3\tKatherine\thot\t8\t300\t2026-07-20\n" |
| 2850 | "c4\tHedy\topen\t8\t20\t2026-06-01\n" |
| 2851 | ) |
| 2852 | |
| 2853 | |
| 2854 | def test_filter_records_eq_default_operator(tmp_path): |
| 2855 | data_dir = tmp_path / "data" |
| 2856 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2857 | records = object_records.read_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 2858 | |
| 2859 | matched = object_records.filter_records(records, {"status": "hot"}) |
| 2860 | |
| 2861 | assert [r["id"] for r in matched] == ["c1", "c3"] |
| 2862 | |
| 2863 | |
| 2864 | def test_filter_records_ne(tmp_path): |
| 2865 | data_dir = tmp_path / "data" |
| 2866 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2867 | records = object_records.read_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 2868 | |
| 2869 | matched = object_records.filter_records(records, {"status": cond("ne", "hot")}) |
| 2870 | |
| 2871 | assert [r["id"] for r in matched] == ["c2", "c4"] |
| 2872 | |
| 2873 | |
| 2874 | def test_filter_records_in(tmp_path): |
| 2875 | data_dir = tmp_path / "data" |
| 2876 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2877 | records = object_records.read_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 2878 | |
| 2879 | matched = object_records.filter_records(records, {"status": cond("in", ("hot", "open"))}) |
| 2880 | |
| 2881 | assert [r["id"] for r in matched] == ["c1", "c3", "c4"] |
| 2882 | |
| 2883 | |
| 2884 | def test_filter_records_ordered_operators(tmp_path): |
| 2885 | data_dir = tmp_path / "data" |
| 2886 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2887 | records = object_records.read_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 2888 | |
| 2889 | assert [r["id"] for r in object_records.filter_records(records, {"total": cond("gte", "150")})] == [ |
| 2890 | "c1", |
| 2891 | "c3", |
| 2892 | ] |
| 2893 | assert [r["id"] for r in object_records.filter_records(records, {"total": cond("lte", "50")})] == [ |
| 2894 | "c2", |
| 2895 | "c4", |
| 2896 | ] |
| 2897 | assert [r["id"] for r in object_records.filter_records(records, {"total": cond("gt", "150")})] == ["c3"] |
| 2898 | assert [r["id"] for r in object_records.filter_records(records, {"total": cond("lt", "50")})] == ["c4"] |
| 2899 | |
| 2900 | |
| 2901 | def test_filter_records_multiple_conditions_are_anded_range_query(tmp_path): |
| 2902 | data_dir = tmp_path / "data" |
| 2903 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2904 | records = object_records.read_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 2905 | |
| 2906 | where = {"created_at": [cond("gte", "2026-07-01"), cond("lte", "2026-07-15")]} |
| 2907 | matched = object_records.filter_records(records, where) |
| 2908 | |
| 2909 | assert [r["id"] for r in matched] == ["c1", "c2"] |
| 2910 | |
| 2911 | |
| 2912 | def test_filter_records_no_where_returns_full_copy_not_alias(tmp_path): |
| 2913 | data_dir = tmp_path / "data" |
| 2914 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2915 | records = object_records.read_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 2916 | |
| 2917 | matched_none = object_records.filter_records(records, None) |
| 2918 | matched_empty = object_records.filter_records(records, {}) |
| 2919 | |
| 2920 | assert matched_none == records |
| 2921 | assert matched_empty == records |
| 2922 | assert matched_none[0] is not records[0] |
| 2923 | matched_none[0]["name"] = "mutated" |
| 2924 | assert records[0]["name"] == "Ada" |
| 2925 | |
| 2926 | |
| 2927 | def test_filter_records_matching_nothing_is_an_empty_list_not_an_error(tmp_path): |
| 2928 | data_dir = tmp_path / "data" |
| 2929 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2930 | records = object_records.read_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 2931 | |
| 2932 | matched = object_records.filter_records(records, {"status": "nonexistent"}) |
| 2933 | |
| 2934 | assert matched == [] |
| 2935 | |
| 2936 | |
| 2937 | def test_filter_records_can_only_narrow_its_own_input_never_widen(tmp_path): |
| 2938 | """The structural narrow-only guarantee (58's Permissions Posture): |
| 2939 | filter_records only ever removes rows from what it is handed. Feed it |
| 2940 | an already-narrowed (e.g. row-permission-filtered) subset and a filter |
| 2941 | that -- if it read the ORIGINAL collection -- would match a row outside |
| 2942 | that subset; the excluded row can never reappear.""" |
| 2943 | data_dir = tmp_path / "data" |
| 2944 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2945 | all_records = object_records.read_collection_records("contacts", base_dir=data_dir, roots=[]) |
| 2946 | # Simulate a permission row filter that already excluded c3/c4 (owner_id 8). |
| 2947 | row_filtered = [r for r in all_records if r["owner_id"] == "7"] |
| 2948 | |
| 2949 | # A field filter matching owner_id=8 rows (c3, c4) against the FULL |
| 2950 | # collection would return them -- but applied to the already-narrowed |
| 2951 | # input, it can only ever be a subset of that input. |
| 2952 | matched = object_records.filter_records(row_filtered, {"owner_id": "8"}) |
| 2953 | |
| 2954 | assert matched == [] |
| 2955 | assert all(record["id"] in {"c1", "c2"} for record in row_filtered) |
| 2956 | |
| 2957 | |
| 2958 | def test_list_collection_records_where_filters_before_pagination(tmp_path): |
| 2959 | data_dir = tmp_path / "data" |
| 2960 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2961 | |
| 2962 | result = object_records.list_collection_records( |
| 2963 | "contacts", |
| 2964 | base_dir=data_dir, |
| 2965 | roots=[], |
| 2966 | where={"status": "hot"}, |
| 2967 | limit=1, |
| 2968 | offset=1, |
| 2969 | ) |
| 2970 | |
| 2971 | assert result["total"] == 2 |
| 2972 | assert result["count"] == 1 |
| 2973 | assert result["records"] == [{"id": "c3", "name": "Katherine", "status": "hot", "owner_id": "8", "total": "300", "created_at": "2026-07-20"}] |
| 2974 | assert result["has_more"] is False |
| 2975 | |
| 2976 | |
| 2977 | def test_list_collection_records_where_none_is_unaffected(tmp_path): |
| 2978 | data_dir = tmp_path / "data" |
| 2979 | write_records(data_dir, "contacts", CONTACTS_TSV) |
| 2980 | |
| 2981 | result = object_records.list_collection_records("contacts", base_dir=data_dir, roots=[], limit=2) |
| 2982 | |
| 2983 | assert result["total"] == 4 |
| 2984 | assert result["count"] == 2 |
| 2985 | |
| 2986 | |
| 2987 | def test_list_collection_records_where_on_append_mode_sees_folded_state(tmp_path): |
| 2988 | """58's Storage/Append-mode note: a filter applies to the folded |
| 2989 | (current) view, never a superseded physical version or a tombstone.""" |
| 2990 | data_dir = tmp_path / "data" |
| 2991 | write_append_schema( |
| 2992 | data_dir, |
| 2993 | "widgets", |
| 2994 | [{"name": "id"}, {"name": "name"}, {"name": "status"}], |
| 2995 | ) |
| 2996 | |
| 2997 | object_records.create_collection_record( |
| 2998 | "widgets", {"id": "w1", "name": "Alpha", "status": "draft"}, base_dir=data_dir, roots=[] |
| 2999 | ) |
| 3000 | object_records.create_collection_record( |
| 3001 | "widgets", {"id": "w2", "name": "Beta", "status": "live"}, base_dir=data_dir, roots=[] |
| 3002 | ) |
| 3003 | object_records.create_collection_record( |
| 3004 | "widgets", {"id": "w3", "name": "Gamma", "status": "live"}, base_dir=data_dir, roots=[] |
| 3005 | ) |
| 3006 | # w1 supersedes to "live" -- an unfiltered pre-update read would have |
| 3007 | # missed it under status=live; the fold must happen before filtering. |
| 3008 | object_records.update_collection_record( |
| 3009 | "widgets", "w1", {"status": "live"}, base_dir=data_dir, roots=[] |
| 3010 | ) |
| 3011 | # w3 is tombstoned -- must never show up in a filtered result either. |
| 3012 | object_records.delete_collection_record("widgets", "w3", base_dir=data_dir, roots=[]) |
| 3013 | |
| 3014 | result = object_records.list_collection_records( |
| 3015 | "widgets", base_dir=data_dir, roots=[], where={"status": "live"} |
| 3016 | ) |
| 3017 | |
| 3018 | assert sorted(r["id"] for r in result["records"]) == ["w1", "w2"] |
| 3019 | assert result["total"] == 2 |
| 3020 | |
| 3021 | |
| 3022 | def test_normalize_filter_value_validates_and_normalizes_boolean(tmp_path): |
| 3023 | field = {"name": "is_public", "type": "boolean"} |
| 3024 | |
| 3025 | assert object_records.normalize_filter_value(field, "eq", "1") == "true" |
| 3026 | assert object_records.normalize_filter_value(field, "eq", "yes") == "true" |
| 3027 | assert object_records.normalize_filter_value(field, "eq", "0") == "false" |
| 3028 | with pytest.raises(object_records.InvalidRecordPayloadError): |
| 3029 | object_records.normalize_filter_value(field, "eq", "maybe") |
| 3030 | |
| 3031 | |
| 3032 | def test_normalize_filter_value_validates_types(tmp_path): |
| 3033 | int_field = {"name": "total", "type": "integer"} |
| 3034 | date_field = {"name": "created_at", "type": "date"} |
| 3035 | enum_field = {"name": "status", "type": "enum", "enum": ["draft", "live"]} |
| 3036 | |
| 3037 | assert object_records.normalize_filter_value(int_field, "gte", "150") == "150" |
| 3038 | with pytest.raises(object_records.InvalidRecordPayloadError): |
| 3039 | object_records.normalize_filter_value(int_field, "eq", "not-a-number") |
| 3040 | |
| 3041 | assert object_records.normalize_filter_value(date_field, "gte", "2026-07-01") == "2026-07-01" |
| 3042 | with pytest.raises(object_records.InvalidRecordPayloadError): |
| 3043 | object_records.normalize_filter_value(date_field, "eq", "not-a-date") |
| 3044 | |
| 3045 | assert object_records.normalize_filter_value(enum_field, "eq", "live") == "live" |
| 3046 | with pytest.raises(object_records.InvalidRecordPayloadError): |
| 3047 | object_records.normalize_filter_value(enum_field, "eq", "bogus") |
| 3048 | |
| 3049 | |
| 3050 | def test_normalize_filter_value_rejects_ordered_op_on_unsupported_type(tmp_path): |
| 3051 | text_field = {"name": "name", "type": "text"} |
| 3052 | |
| 3053 | with pytest.raises(object_records.InvalidRecordPayloadError, match="does not support operator"): |
| 3054 | object_records.normalize_filter_value(text_field, "gte", "A") |
| 3055 | |
| 3056 | # eq/ne/in stay fine for text. |
| 3057 | assert object_records.normalize_filter_value(text_field, "eq", "Ada") == "Ada" |
| 3058 | |
| 3059 | |
| 3060 | def test_normalize_filter_value_permissive_when_field_is_none(tmp_path): |
| 3061 | """A schemaless/derived collection has no field def to validate |
| 3062 | against; 58 treats that as permissive, matching every other |
| 3063 | schema-optional check in this module.""" |
| 3064 | assert object_records.normalize_filter_value(None, "gte", "anything") == "anything" |