| 1 | #!/usr/bin/env python3 |
| 2 | """Web image search CLI. |
| 3 | |
| 4 | Sister tool to ``image_gen.py``: search openly-licensed providers, either |
| 5 | download one metadata-ranked original or prepare a thumbnail-only visual |
| 6 | shortlist and defer the original until ``--promote``. |
| 7 | |
| 8 | Workflow: |
| 9 | 1. Build an :class:`ImageSearchRequest` from CLI args. |
| 10 | 2. Quality-first license search: |
| 11 | - Default: ask each provider for ``all`` allowed matches (CC0, |
| 12 | Public Domain, Pexels, Pixabay, CC BY, CC BY-SA), pick the |
| 13 | highest-scoring downloadable candidate, and record whether it |
| 14 | needs attribution. |
| 15 | - Strict mode: when ``--strict-no-attribution`` is set, ask only |
| 16 | for ``no-attribution-only`` matches and fail if none can be |
| 17 | downloaded. |
| 18 | 3. Best-only mode downloads the chosen original into ``--output``. |
| 19 | Thumbnail mode saves provider previews and stops for visual selection. |
| 20 | 4. Best-only / ``--promote`` appends a record to ``image_sources.json`` |
| 21 | (the single source of truth for downstream credit rendering). |
| 22 | |
| 23 | Examples: |
| 24 | # Default: zero-config, quality-first across allowed licenses |
| 25 | python3 scripts/image_search.py "offshore wind farm" \ |
| 26 | --filename cover_bg.jpg --slide 01_cover \ |
| 27 | --orientation landscape -o projects/demo/images |
| 28 | |
| 29 | # Strict mode: refuse anything that would require attribution |
| 30 | python3 scripts/image_search.py "abstract gradient" \ |
| 31 | --filename hero.jpg --strict-no-attribution \ |
| 32 | -o projects/demo/images |
| 33 | |
| 34 | # Pin a specific provider (useful when an API key is set) |
| 35 | python3 scripts/image_search.py "executive meeting" \ |
| 36 | --filename team.jpg --provider pexels \ |
| 37 | --orientation landscape -o projects/demo/images |
| 38 | """ |
| 39 | |
| 40 | from __future__ import annotations |
| 41 | |
| 42 | import argparse |
| 43 | import concurrent.futures |
| 44 | import importlib |
| 45 | import json |
| 46 | import os |
| 47 | import shutil |
| 48 | import sys |
| 49 | import tempfile |
| 50 | import threading |
| 51 | from dataclasses import dataclass, replace |
| 52 | from datetime import datetime, timezone |
| 53 | from pathlib import Path |
| 54 | from typing import Callable, Optional |
| 55 | |
| 56 | import requests |
| 57 | |
| 58 | # Make sibling modules importable when this script is invoked directly. |
| 59 | _SCRIPTS_DIR = Path(__file__).resolve().parent |
| 60 | if str(_SCRIPTS_DIR) not in sys.path: |
| 61 | sys.path.insert(0, str(_SCRIPTS_DIR)) |
| 62 | |
| 63 | from console_encoding import configure_utf8_stdio # noqa: E402 |
| 64 | from config import load_prefixed_env_file # noqa: E402 |
| 65 | from image_backends.backend_common import download_image, save_image_bytes # noqa: E402 |
| 66 | from image_sources.provider_common import ( # noqa: E402 |
| 67 | AssetCandidate, |
| 68 | ImageSearchRequest, |
| 69 | USER_AGENT, |
| 70 | build_attribution_text, |
| 71 | ensure_json_parent, |
| 72 | missing_required_terms, |
| 73 | score_candidate, |
| 74 | score_review_candidate, |
| 75 | ) |
| 76 | |
| 77 | configure_utf8_stdio() |
| 78 | |
| 79 | |
| 80 | # --------------------------------------------------------------------------- |
| 81 | # Provider registry |
| 82 | # --------------------------------------------------------------------------- |
| 83 | |
| 84 | PROVIDER_MODULES: dict[str, str] = { |
| 85 | "openverse": "image_sources.provider_openverse", |
| 86 | "wikimedia": "image_sources.provider_wikimedia", |
| 87 | "pexels": "image_sources.provider_pexels", |
| 88 | "pixabay": "image_sources.provider_pixabay", |
| 89 | } |
| 90 | |
| 91 | # Providers that work without configuration. ``image_search.py`` defaults |
| 92 | # to these so a fresh clone can search immediately. |
| 93 | ZERO_CONFIG_PROVIDERS: tuple[str, ...] = ("openverse", "wikimedia") |
| 94 | KEYED_PROVIDERS: tuple[str, ...] = ("pexels", "pixabay") |
| 95 | ALL_PROVIDERS: tuple[str, ...] = ZERO_CONFIG_PROVIDERS + KEYED_PROVIDERS |
| 96 | |
| 97 | ORIENTATION_CHOICES = ("any", "landscape", "portrait", "square") |
| 98 | |
| 99 | # --- Batch mode (`--batch image_queries.json`) ----------------------------- |
| 100 | # Web providers are politeness-sensitive (Wikimedia/Openverse expect a modest |
| 101 | # rate), so the default concurrency is deliberately low. Sister-tool |
| 102 | # `image_gen.py` hits a paid API and defaults higher; here 3 keeps several |
| 103 | # rows in flight without hammering any single free provider. Set to 1 to |
| 104 | # restore strict one-at-a-time pacing. |
| 105 | DEFAULT_SEARCH_CONCURRENCY = 3 |
| 106 | DEFAULT_CANDIDATE_PAGE_SIZE = 8 |
| 107 | |
| 108 | SEARCH_STATUS_PENDING = "Pending" |
| 109 | SEARCH_STATUS_SOURCED = "Sourced" |
| 110 | SEARCH_STATUS_FAILED = "Failed" |
| 111 | SEARCH_STATUS_NEEDS_SELECTION = "Needs-Selection" |
| 112 | SEARCH_STATUS_NEEDS_MANUAL = "Needs-Manual" |
| 113 | SEARCH_VALID_STATUSES = { |
| 114 | SEARCH_STATUS_PENDING, |
| 115 | SEARCH_STATUS_SOURCED, |
| 116 | SEARCH_STATUS_FAILED, |
| 117 | SEARCH_STATUS_NEEDS_SELECTION, |
| 118 | SEARCH_STATUS_NEEDS_MANUAL, |
| 119 | } |
| 120 | # Selection and manual rows are terminal until the agent promotes a candidate |
| 121 | # or materially changes the query; only Pending/Failed rows auto-retry. |
| 122 | SEARCH_RETRYABLE_STATUSES = {SEARCH_STATUS_PENDING, SEARCH_STATUS_FAILED} |
| 123 | SEARCH_REQUIRED_ITEM_FIELDS = ("filename", "query", "status") |
| 124 | SEARCH_FAILURE_NO_MATCH = "no-match" |
| 125 | SEARCH_FAILURE_RETRYABLE = "retryable" |
| 126 | _CANDIDATE_SELECTION_OUTPUT_FIELDS = ( |
| 127 | "review_sheet", |
| 128 | "candidate_count", |
| 129 | "candidate_total", |
| 130 | "has_more_candidates", |
| 131 | "next_candidate_page", |
| 132 | ) |
| 133 | |
| 134 | _WEAK_REQUIRED_TERM_PARTS = frozenset({ |
| 135 | "ancient town", |
| 136 | "bridge", |
| 137 | "canyon", |
| 138 | "cave", |
| 139 | "city", |
| 140 | "floating bridge", |
| 141 | "forest", |
| 142 | "gate", |
| 143 | "grand canyon", |
| 144 | "ground fissure", |
| 145 | "lake", |
| 146 | "monastery", |
| 147 | "monument", |
| 148 | "river", |
| 149 | "shrine", |
| 150 | "square", |
| 151 | "station", |
| 152 | "stone forest", |
| 153 | "stone pillar", |
| 154 | "stream", |
| 155 | "temple", |
| 156 | "valley", |
| 157 | "village", |
| 158 | "古城", |
| 159 | "古镇", |
| 160 | "地缝", |
| 161 | "大峡谷", |
| 162 | "寺", |
| 163 | "峡谷", |
| 164 | "广场", |
| 165 | "桥", |
| 166 | "洞", |
| 167 | "溪", |
| 168 | "石林", |
| 169 | "石柱", |
| 170 | }) |
| 171 | |
| 172 | |
| 173 | def _parse_required_terms(raw: object) -> tuple[str, ...]: |
| 174 | """Parse entity-safety terms from CLI / batch JSON. |
| 175 | |
| 176 | Multiple groups are ANDed. Alternatives inside one group are separated by |
| 177 | ``|``; comma splitting is accepted as CLI convenience. Examples: |
| 178 | ``["Chongqing", "Jiefangbei|Liberation Monument"]``. |
| 179 | """ |
| 180 | if raw is None: |
| 181 | return () |
| 182 | values: list[str] |
| 183 | if isinstance(raw, str): |
| 184 | values = [raw] |
| 185 | elif isinstance(raw, (list, tuple)): |
| 186 | values = [] |
| 187 | for item in raw: |
| 188 | if not isinstance(item, str): |
| 189 | raise ValueError("required_terms items must be strings") |
| 190 | values.append(item) |
| 191 | else: |
| 192 | raise ValueError("required_terms must be a string or list of strings") |
| 193 | |
| 194 | terms: list[str] = [] |
| 195 | for value in values: |
| 196 | for part in value.split(","): |
| 197 | part = part.strip() |
| 198 | if part: |
| 199 | terms.append(part) |
| 200 | return tuple(terms) |
| 201 | |
| 202 | |
| 203 | def _parse_query_variants(raw: object) -> tuple[str, ...]: |
| 204 | """Parse optional material query variants while preserving order.""" |
| 205 | if raw is None: |
| 206 | return () |
| 207 | if isinstance(raw, str): |
| 208 | values = [raw] |
| 209 | elif isinstance(raw, (list, tuple)): |
| 210 | values = list(raw) |
| 211 | else: |
| 212 | raise ValueError("query_variants must be a string or list of strings") |
| 213 | |
| 214 | variants: list[str] = [] |
| 215 | seen: set[str] = set() |
| 216 | for value in values: |
| 217 | if not isinstance(value, str) or not value.strip(): |
| 218 | raise ValueError("query_variants items must be non-empty strings") |
| 219 | normalized = value.strip() |
| 220 | key = normalized.casefold() |
| 221 | if key not in seen: |
| 222 | seen.add(key) |
| 223 | variants.append(normalized) |
| 224 | return tuple(variants) |
| 225 | |
| 226 | |
| 227 | def _search_request_variants( |
| 228 | request: ImageSearchRequest, |
| 229 | ) -> list[ImageSearchRequest]: |
| 230 | """Expand one intent into explicit provider queries without duplicates.""" |
| 231 | queries = _parse_query_variants((request.query, *request.query_variants)) |
| 232 | return [ |
| 233 | replace(request, query=query, query_variants=()) |
| 234 | for query in queries |
| 235 | ] |
| 236 | |
| 237 | |
| 238 | def _warn_weak_required_terms(required_terms: tuple[str, ...]) -> None: |
| 239 | """Warn when required_terms contain generic category words. |
| 240 | |
| 241 | These terms are useful in the query but dangerous as identity gates: |
| 242 | broadening a small Chinese attraction from its proper name to "canyon" / |
| 243 | "stone pillar" raises coverage while admitting wrong entities. |
| 244 | """ |
| 245 | weak: list[str] = [] |
| 246 | for group in required_terms: |
| 247 | for part in group.split("|"): |
| 248 | normalized = part.strip().lower() |
| 249 | if normalized in _WEAK_REQUIRED_TERM_PARTS: |
| 250 | weak.append(part.strip()) |
| 251 | if weak: |
| 252 | print( |
| 253 | " warning: required_terms contains generic category term(s) " |
| 254 | f"{weak}; keep proper-name / geography anchors too, and prefer " |
| 255 | "Needs-Manual or --from-url over loosening identity gates.", |
| 256 | file=sys.stderr, |
| 257 | ) |
| 258 | |
| 259 | |
| 260 | # --------------------------------------------------------------------------- |
| 261 | # .env loading |
| 262 | # --------------------------------------------------------------------------- |
| 263 | |
| 264 | |
| 265 | def _load_search_env_file() -> None: |
| 266 | """Load image-search keys from the shared PPT Master .env locations.""" |
| 267 | load_prefixed_env_file(("PEXELS_", "PIXABAY_")) |
| 268 | |
| 269 | |
| 270 | # --------------------------------------------------------------------------- |
| 271 | # Provider dispatch |
| 272 | # --------------------------------------------------------------------------- |
| 273 | |
| 274 | |
| 275 | def _load_provider(name: str): |
| 276 | return importlib.import_module(PROVIDER_MODULES[name]) |
| 277 | |
| 278 | |
| 279 | def _is_keyed_provider_unconfigured(provider_name: str, exc: Exception) -> bool: |
| 280 | """Treat 'API key missing' as a non-fatal skip so the default provider |
| 281 | chain can keep going.""" |
| 282 | if provider_name not in KEYED_PROVIDERS: |
| 283 | return False |
| 284 | return "API_KEY" in str(exc) |
| 285 | |
| 286 | |
| 287 | @dataclass |
| 288 | class SearchDownloadResult: |
| 289 | """Carry one search result without collapsing no-match and retryable failures.""" |
| 290 | |
| 291 | candidate: Optional[AssetCandidate] = None |
| 292 | provider_name: Optional[str] = None |
| 293 | stage: Optional[str] = None |
| 294 | actual_dimensions: Optional[tuple[int, int]] = None |
| 295 | staged_path: Optional[Path] = None |
| 296 | output_path: Optional[Path] = None |
| 297 | selection_required: bool = False |
| 298 | candidate_count: int = 0 |
| 299 | candidate_total: int = 0 |
| 300 | candidate_page: int = 1 |
| 301 | has_more_candidates: bool = False |
| 302 | review_sheet: Optional[Path] = None |
| 303 | failure_kind: Optional[str] = None |
| 304 | error: Optional[str] = None |
| 305 | |
| 306 | |
| 307 | class DownloadQualityError(ValueError): |
| 308 | """Signal a readable candidate that fails the requested image contract.""" |
| 309 | |
| 310 | |
| 311 | def _clear_candidate_selection_outputs(item: dict) -> None: |
| 312 | """Remove stale shortlist results while preserving query inputs.""" |
| 313 | for field_name in _CANDIDATE_SELECTION_OUTPUT_FIELDS: |
| 314 | item.pop(field_name, None) |
| 315 | |
| 316 | |
| 317 | def _is_pillow_decompression_error(exc: BaseException) -> bool: |
| 318 | """Recognize Pillow's safety exception without making Pillow a hard import.""" |
| 319 | cls = type(exc) |
| 320 | return ( |
| 321 | cls.__name__ == "DecompressionBombError" |
| 322 | and cls.__module__.startswith("PIL.") |
| 323 | ) |
| 324 | |
| 325 | |
| 326 | def _is_recoverable_image_error(exc: BaseException) -> bool: |
| 327 | """Return whether a provider/download/image failure can be reported cleanly.""" |
| 328 | return isinstance( |
| 329 | exc, |
| 330 | ( |
| 331 | requests.RequestException, |
| 332 | OSError, |
| 333 | RuntimeError, |
| 334 | SyntaxError, |
| 335 | ValueError, |
| 336 | ), |
| 337 | ) or _is_pillow_decompression_error(exc) |
| 338 | |
| 339 | |
| 340 | def _try_provider( |
| 341 | name: str, |
| 342 | request: ImageSearchRequest, |
| 343 | license_tier_filter: str, |
| 344 | *, |
| 345 | provider_is_explicit: bool = False, |
| 346 | ) -> tuple[Optional[list[AssetCandidate]], Optional[str]]: |
| 347 | """Run one provider while preserving whether it errored or returned no rows. |
| 348 | |
| 349 | An explicitly selected provider is required; a missing key is retryable |
| 350 | instead of an optional-provider skip. |
| 351 | """ |
| 352 | try: |
| 353 | module = _load_provider(name) |
| 354 | return module.search(request, license_tier_filter=license_tier_filter), None |
| 355 | except RuntimeError as exc: |
| 356 | if ( |
| 357 | not provider_is_explicit |
| 358 | and _is_keyed_provider_unconfigured(name, exc) |
| 359 | ): |
| 360 | print( |
| 361 | f" [{name}] skipped: {exc}", |
| 362 | file=sys.stderr, |
| 363 | ) |
| 364 | return None, None |
| 365 | else: |
| 366 | print(f" [{name}] error: {exc}", file=sys.stderr) |
| 367 | return None, f"{name}: {exc}" |
| 368 | except (requests.RequestException, OSError, ValueError) as exc: |
| 369 | print(f" [{name}] error: {exc}", file=sys.stderr) |
| 370 | return None, f"{name}: {exc}" |
| 371 | except ImportError as exc: |
| 372 | print(f" [{name}] error: {exc}", file=sys.stderr) |
| 373 | return None, f"{name}: provider import failed: {exc}" |
| 374 | |
| 375 | |
| 376 | # --------------------------------------------------------------------------- |
| 377 | # Post-download quality validation |
| 378 | # --------------------------------------------------------------------------- |
| 379 | |
| 380 | _MIN_DOWNLOAD_PIXELS = 800 * 600 # absolute floor for automated originals |
| 381 | |
| 382 | |
| 383 | def _validate_downloaded_quality( |
| 384 | path: Path, |
| 385 | *, |
| 386 | min_width: int = 0, |
| 387 | min_height: int = 0, |
| 388 | enforce_thumbnail_floor: bool = True, |
| 389 | ) -> bool: |
| 390 | """Reject unreadable images and actual EXIF-oriented dimensions below contract. |
| 391 | |
| 392 | Upstream metadata can be inaccurate (e.g. Openverse aggregates rawpixel |
| 393 | which only exposes a preview). This function checks what was actually |
| 394 | written to disk. Automated paths also reject thumbnails/previews; explicit |
| 395 | manual paths may disable that absolute floor while retaining their own |
| 396 | requested dimensions. |
| 397 | """ |
| 398 | try: |
| 399 | from PIL import Image, ImageOps # type: ignore |
| 400 | except ImportError as exc: |
| 401 | raise RuntimeError( |
| 402 | "Pillow is required to validate downloaded image dimensions. " |
| 403 | "Install it with: pip install Pillow" |
| 404 | ) from exc |
| 405 | try: |
| 406 | with Image.open(path) as im: |
| 407 | oriented = ImageOps.exif_transpose(im) |
| 408 | oriented.load() |
| 409 | w, h = oriented.size |
| 410 | if oriented is not im: |
| 411 | oriented.close() |
| 412 | if w < min_width or h < min_height: |
| 413 | print( |
| 414 | f" rejected: downloaded image dimensions {w}x{h} are below " |
| 415 | f"the requested minimum {min_width}x{min_height}", |
| 416 | file=sys.stderr, |
| 417 | ) |
| 418 | return False |
| 419 | if enforce_thumbnail_floor and w * h < _MIN_DOWNLOAD_PIXELS: |
| 420 | print( |
| 421 | f" rejected: downloaded image too small " |
| 422 | f"({w}x{h} = {w*h:,} px < {_MIN_DOWNLOAD_PIXELS:,} px minimum)", |
| 423 | file=sys.stderr, |
| 424 | ) |
| 425 | return False |
| 426 | return True |
| 427 | except Exception as exc: |
| 428 | if not _is_recoverable_image_error(exc): |
| 429 | raise |
| 430 | print( |
| 431 | f" rejected: downloaded file is not a readable image ({exc})", |
| 432 | file=sys.stderr, |
| 433 | ) |
| 434 | return False |
| 435 | |
| 436 | |
| 437 | def _stage_and_validate_image( |
| 438 | output_path: Path, |
| 439 | materialize: Callable[[Path], object], |
| 440 | *, |
| 441 | min_width: int, |
| 442 | min_height: int, |
| 443 | enforce_thumbnail_floor: bool = True, |
| 444 | ) -> tuple[Path, tuple[int, int]]: |
| 445 | """Materialize and validate beside the target without changing the canonical.""" |
| 446 | output_path.parent.mkdir(parents=True, exist_ok=True) |
| 447 | fd, temp_name = tempfile.mkstemp( |
| 448 | prefix=f".{output_path.stem}.", |
| 449 | suffix=output_path.suffix, |
| 450 | dir=str(output_path.parent), |
| 451 | ) |
| 452 | os.close(fd) |
| 453 | temp_path = Path(temp_name) |
| 454 | keep_temp = False |
| 455 | try: |
| 456 | materialize(temp_path) |
| 457 | if not _validate_downloaded_quality( |
| 458 | temp_path, |
| 459 | min_width=min_width, |
| 460 | min_height=min_height, |
| 461 | enforce_thumbnail_floor=enforce_thumbnail_floor, |
| 462 | ): |
| 463 | raise DownloadQualityError( |
| 464 | "downloaded image did not satisfy the requested dimensions/readability" |
| 465 | ) |
| 466 | actual_dimensions = _measure_actual_image(temp_path) |
| 467 | if actual_dimensions is None: |
| 468 | raise DownloadQualityError( |
| 469 | "downloaded image dimensions could not be measured" |
| 470 | ) |
| 471 | keep_temp = True |
| 472 | return temp_path, actual_dimensions |
| 473 | finally: |
| 474 | if not keep_temp: |
| 475 | try: |
| 476 | temp_path.unlink(missing_ok=True) |
| 477 | except OSError: |
| 478 | pass |
| 479 | |
| 480 | |
| 481 | def _stage_validated_image( |
| 482 | url: str, |
| 483 | output_path: Path, |
| 484 | *, |
| 485 | min_width: int, |
| 486 | min_height: int, |
| 487 | enforce_thumbnail_floor: bool = True, |
| 488 | ) -> tuple[Path, tuple[int, int]]: |
| 489 | """Download and validate beside the target without changing the canonical.""" |
| 490 | return _stage_and_validate_image( |
| 491 | output_path, |
| 492 | lambda temp_path: download_image( |
| 493 | url, |
| 494 | str(temp_path), |
| 495 | headers={"User-Agent": USER_AGENT}, |
| 496 | ), |
| 497 | min_width=min_width, |
| 498 | min_height=min_height, |
| 499 | enforce_thumbnail_floor=enforce_thumbnail_floor, |
| 500 | ) |
| 501 | |
| 502 | |
| 503 | def _stage_validated_candidate_copy( |
| 504 | source_path: Path, |
| 505 | output_path: Path, |
| 506 | *, |
| 507 | min_width: int, |
| 508 | min_height: int, |
| 509 | enforce_thumbnail_floor: bool = True, |
| 510 | ) -> tuple[Path, tuple[int, int]]: |
| 511 | """Stage a saved pool candidate, preserving target-extension correctness.""" |
| 512 | return _stage_and_validate_image( |
| 513 | output_path, |
| 514 | lambda temp_path: save_image_bytes( |
| 515 | source_path.read_bytes(), |
| 516 | str(temp_path), |
| 517 | ), |
| 518 | min_width=min_width, |
| 519 | min_height=min_height, |
| 520 | enforce_thumbnail_floor=enforce_thumbnail_floor, |
| 521 | ) |
| 522 | |
| 523 | |
| 524 | def _commit_staged_image( |
| 525 | staged_path: Path, |
| 526 | target_path: Path, |
| 527 | manifest_writer: Callable[[], Path], |
| 528 | ) -> Path: |
| 529 | """Install a staged image and roll it back if provenance cannot be written.""" |
| 530 | target_path.parent.mkdir(parents=True, exist_ok=True) |
| 531 | backup_path: Optional[Path] = None |
| 532 | installed = False |
| 533 | |
| 534 | try: |
| 535 | if target_path.exists(): |
| 536 | if not target_path.is_file(): |
| 537 | raise RuntimeError( |
| 538 | f"image target exists but is not a regular file: {target_path}" |
| 539 | ) |
| 540 | fd, backup_name = tempfile.mkstemp( |
| 541 | prefix=f".{target_path.stem}.backup.", |
| 542 | suffix=target_path.suffix, |
| 543 | dir=str(target_path.parent), |
| 544 | ) |
| 545 | os.close(fd) |
| 546 | reserved_backup_path = Path(backup_name) |
| 547 | reserved_backup_path.unlink() |
| 548 | os.replace(target_path, reserved_backup_path) |
| 549 | backup_path = reserved_backup_path |
| 550 | |
| 551 | os.replace(staged_path, target_path) |
| 552 | installed = True |
| 553 | written = manifest_writer() |
| 554 | except Exception as exc: |
| 555 | rollback_errors: list[str] = [] |
| 556 | if installed: |
| 557 | try: |
| 558 | target_path.unlink(missing_ok=True) |
| 559 | except OSError as rollback_exc: |
| 560 | rollback_errors.append(f"cannot remove new target: {rollback_exc}") |
| 561 | if backup_path is not None and backup_path.exists(): |
| 562 | try: |
| 563 | os.replace(backup_path, target_path) |
| 564 | except OSError as rollback_exc: |
| 565 | rollback_errors.append(f"cannot restore prior target: {rollback_exc}") |
| 566 | if rollback_errors: |
| 567 | raise RuntimeError( |
| 568 | f"{exc}; image rollback also failed: {'; '.join(rollback_errors)}" |
| 569 | ) from exc |
| 570 | raise |
| 571 | else: |
| 572 | if backup_path is not None: |
| 573 | try: |
| 574 | backup_path.unlink(missing_ok=True) |
| 575 | except OSError as exc: |
| 576 | print( |
| 577 | f" warning: could not remove image backup {backup_path}: {exc}", |
| 578 | file=sys.stderr, |
| 579 | ) |
| 580 | return written |
| 581 | finally: |
| 582 | try: |
| 583 | staged_path.unlink(missing_ok=True) |
| 584 | except OSError: |
| 585 | pass |
| 586 | |
| 587 | |
| 588 | def _write_review_copy( |
| 589 | src: Path, dest_dir: Path, name: str, max_side: int = 1024 |
| 590 | ) -> Optional[Path]: |
| 591 | """Write a downscaled JPEG review copy of ``src`` into ``dest_dir``. |
| 592 | |
| 593 | The placed / promoted asset is always the full-resolution original; this |
| 594 | bounded copy exists only so the agent can Read a sanely-sized image to |
| 595 | confirm suitability regardless of how large the source is. Best-effort — |
| 596 | returns None (non-fatal) if Pillow or the source is unavailable. |
| 597 | """ |
| 598 | try: |
| 599 | from PIL import Image, ImageOps # type: ignore |
| 600 | except ImportError: |
| 601 | return None |
| 602 | review_path: Optional[Path] = None |
| 603 | try: |
| 604 | dest_dir.mkdir(parents=True, exist_ok=True) |
| 605 | review_path = dest_dir / f"{Path(name).stem}.jpg" |
| 606 | with Image.open(src) as im: |
| 607 | oriented = ImageOps.exif_transpose(im) |
| 608 | review = oriented.convert("RGB") |
| 609 | review.thumbnail((max_side, max_side)) |
| 610 | review.save(review_path, "JPEG", quality=85) |
| 611 | review.close() |
| 612 | if oriented is not im: |
| 613 | oriented.close() |
| 614 | return review_path |
| 615 | except Exception as exc: |
| 616 | if not _is_recoverable_image_error(exc): |
| 617 | raise |
| 618 | if review_path is not None: |
| 619 | try: |
| 620 | review_path.unlink(missing_ok=True) |
| 621 | except OSError: |
| 622 | pass |
| 623 | return None |
| 624 | |
| 625 | |
| 626 | def _write_candidate_review_sheet( |
| 627 | cand_dir: Path, |
| 628 | pool: list[dict], |
| 629 | ) -> Optional[Path]: |
| 630 | """Build a contact sheet from only the current candidate pool reviews.""" |
| 631 | review_paths: list[Path] = [] |
| 632 | for entry in pool: |
| 633 | review = entry.get("review") |
| 634 | if not isinstance(review, str): |
| 635 | continue |
| 636 | review_path = cand_dir / review |
| 637 | if review_path.is_file(): |
| 638 | review_paths.append(review_path) |
| 639 | if not review_paths: |
| 640 | return None |
| 641 | |
| 642 | try: |
| 643 | from rotate_images import ImageRotator |
| 644 | except ImportError: |
| 645 | return None |
| 646 | |
| 647 | output_path = cand_dir / "review_sheet.jpg" |
| 648 | with tempfile.TemporaryDirectory( |
| 649 | prefix=".candidate-review-", |
| 650 | dir=str(cand_dir), |
| 651 | ) as temporary_dir: |
| 652 | staging_dir = Path(temporary_dir) |
| 653 | for review_path in review_paths: |
| 654 | shutil.copy2(review_path, staging_dir / review_path.name) |
| 655 | staged_sheet = staging_dir / output_path.name |
| 656 | ImageRotator().generate_contact_sheet(staging_dir, staged_sheet) |
| 657 | os.replace(staged_sheet, output_path) |
| 658 | return output_path |
| 659 | |
| 660 | |
| 661 | def _candidate_dedupe_key( |
| 662 | provider_name: str, |
| 663 | candidate: AssetCandidate, |
| 664 | ) -> str: |
| 665 | """Return a stable-enough key for one search window's cross-provider pool.""" |
| 666 | original_url = candidate.download_url.split("?", 1)[0].rstrip("/").casefold() |
| 667 | if original_url: |
| 668 | return original_url |
| 669 | return f"{provider_name}:{candidate.asset_id or candidate.source_page_url}" |
| 670 | |
| 671 | |
| 672 | def _dedupe_ranked_candidates( |
| 673 | ranked: list[tuple[float, str, AssetCandidate]], |
| 674 | ) -> list[tuple[float, str, AssetCandidate]]: |
| 675 | """Keep the highest-scoring occurrence of each cross-query asset.""" |
| 676 | deduped: list[tuple[float, str, AssetCandidate]] = [] |
| 677 | seen: set[str] = set() |
| 678 | for item in sorted(ranked, key=lambda entry: entry[0], reverse=True): |
| 679 | _score, provider_name, candidate = item |
| 680 | key = _candidate_dedupe_key(provider_name, candidate) |
| 681 | if key in seen: |
| 682 | continue |
| 683 | seen.add(key) |
| 684 | deduped.append(item) |
| 685 | return deduped |
| 686 | |
| 687 | |
| 688 | def _save_candidate_thumbnails( |
| 689 | ranked: list[tuple[float, str, AssetCandidate]], |
| 690 | output_dir: Path, |
| 691 | stem: str, |
| 692 | request: ImageSearchRequest, |
| 693 | license_stage: str, |
| 694 | max_candidates: int = DEFAULT_CANDIDATE_PAGE_SIZE, |
| 695 | candidate_page: int = 1, |
| 696 | ) -> tuple[list[dict], Optional[Path], list[str], int, bool]: |
| 697 | """Save one ranked page of previews without downloading originals.""" |
| 698 | cand_dir = output_dir / "candidates" / stem |
| 699 | cand_dir.mkdir(parents=True, exist_ok=True) |
| 700 | review_dir = cand_dir / "review" |
| 701 | review_dir.mkdir(parents=True, exist_ok=True) |
| 702 | |
| 703 | eligible: list[tuple[float, str, AssetCandidate]] = [] |
| 704 | seen_candidates: set[str] = set() |
| 705 | for ranked_item in ranked: |
| 706 | _score, provider_name, candidate = ranked_item |
| 707 | dedupe_key = _candidate_dedupe_key(provider_name, candidate) |
| 708 | if dedupe_key in seen_candidates: |
| 709 | continue |
| 710 | if ( |
| 711 | request.min_width |
| 712 | and candidate.width |
| 713 | and candidate.width < request.min_width |
| 714 | ): |
| 715 | continue |
| 716 | if ( |
| 717 | request.min_height |
| 718 | and candidate.height |
| 719 | and candidate.height < request.min_height |
| 720 | ): |
| 721 | continue |
| 722 | if not candidate.preview_url.strip(): |
| 723 | continue |
| 724 | seen_candidates.add(dedupe_key) |
| 725 | eligible.append(ranked_item) |
| 726 | |
| 727 | candidate_total = len(eligible) |
| 728 | if max_candidates == 0: |
| 729 | page_start = 0 |
| 730 | page_items = eligible |
| 731 | has_more_candidates = False |
| 732 | else: |
| 733 | page_start = (candidate_page - 1) * max_candidates |
| 734 | page_end = page_start + max_candidates |
| 735 | page_items = eligible[page_start:page_end] |
| 736 | has_more_candidates = page_end < candidate_total |
| 737 | |
| 738 | pool: list[dict] = [] |
| 739 | preview_errors: list[str] = [] |
| 740 | for global_rank, (score, provider_name, candidate) in enumerate( |
| 741 | page_items, |
| 742 | start=page_start + 1, |
| 743 | ): |
| 744 | preview_url = candidate.preview_url.strip() |
| 745 | cand_filename = f"candidate_{global_rank:02d}.jpg" |
| 746 | review_path = review_dir / cand_filename |
| 747 | staged_path: Optional[Path] = None |
| 748 | try: |
| 749 | staged_path, preview_dimensions = _stage_validated_image( |
| 750 | preview_url, |
| 751 | review_path, |
| 752 | min_width=1, |
| 753 | min_height=1, |
| 754 | enforce_thumbnail_floor=False, |
| 755 | ) |
| 756 | os.replace(staged_path, review_path) |
| 757 | staged_path = None |
| 758 | missing_terms = missing_required_terms( |
| 759 | candidate, |
| 760 | request.required_terms, |
| 761 | ) |
| 762 | pool.append({ |
| 763 | "rank": global_rank, |
| 764 | "score": round(score, 2), |
| 765 | "filename": cand_filename, |
| 766 | "review": f"review/{review_path.name}", |
| 767 | "provider": provider_name, |
| 768 | "title": candidate.title, |
| 769 | "author": candidate.author, |
| 770 | "source_page_url": candidate.source_page_url, |
| 771 | "download_url": candidate.download_url, |
| 772 | "preview_url": preview_url, |
| 773 | "matched_query": candidate.discovery_query or request.query, |
| 774 | "identity_evidence": ( |
| 775 | "metadata-verified" |
| 776 | if not missing_terms |
| 777 | else "visual-verification-required" |
| 778 | ), |
| 779 | "missing_required_terms": missing_terms, |
| 780 | "license_name": candidate.license_name, |
| 781 | "license_url": candidate.license_url, |
| 782 | "license_tier": candidate.license_tier, |
| 783 | "attribution_required": ( |
| 784 | candidate.license_tier == "attribution-required" |
| 785 | ), |
| 786 | "attribution_text": build_attribution_text( |
| 787 | request.filename, |
| 788 | candidate, |
| 789 | ), |
| 790 | "width": candidate.width, |
| 791 | "height": candidate.height, |
| 792 | "preview_width": preview_dimensions[0], |
| 793 | "preview_height": preview_dimensions[1], |
| 794 | }) |
| 795 | except Exception as exc: |
| 796 | if not _is_recoverable_image_error(exc): |
| 797 | raise |
| 798 | preview_errors.append( |
| 799 | f"{provider_name}/{candidate.title}: preview failed: {exc}" |
| 800 | ) |
| 801 | finally: |
| 802 | if staged_path is not None: |
| 803 | try: |
| 804 | staged_path.unlink(missing_ok=True) |
| 805 | except OSError: |
| 806 | pass |
| 807 | |
| 808 | meta = { |
| 809 | "schema_version": 3, |
| 810 | "candidate_storage": "thumbnail-only", |
| 811 | "target_filename": request.filename, |
| 812 | "selected": None, |
| 813 | "searched_at": datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"), |
| 814 | "license_stage": license_stage, |
| 815 | "candidate_page": candidate_page, |
| 816 | "page_size": max_candidates, |
| 817 | "candidate_total": candidate_total, |
| 818 | "has_more_candidates": has_more_candidates, |
| 819 | "request": { |
| 820 | "query": request.query, |
| 821 | "query_variants": list(request.query_variants), |
| 822 | "purpose": request.purpose, |
| 823 | "slide": request.slide, |
| 824 | "orientation": request.orientation or "any", |
| 825 | "required_terms": list(request.required_terms), |
| 826 | "min_width": request.min_width, |
| 827 | "min_height": request.min_height, |
| 828 | }, |
| 829 | "candidates": pool, |
| 830 | } |
| 831 | meta_path = cand_dir / "candidates.json" |
| 832 | _write_json_atomic(meta_path, meta) |
| 833 | |
| 834 | review_sheet: Optional[Path] = None |
| 835 | if pool: |
| 836 | print( |
| 837 | f" candidate thumbnails: {cand_dir}/ ({len(pool)} saved)", |
| 838 | file=sys.stderr, |
| 839 | ) |
| 840 | try: |
| 841 | review_sheet = _write_candidate_review_sheet(cand_dir, pool) |
| 842 | except Exception as exc: |
| 843 | if not _is_recoverable_image_error(exc): |
| 844 | raise |
| 845 | print( |
| 846 | f" warning: candidate review sheet could not be generated: {exc}", |
| 847 | file=sys.stderr, |
| 848 | ) |
| 849 | else: |
| 850 | if review_sheet is not None: |
| 851 | print( |
| 852 | f"[REPORT] Web image candidate review sheet: {review_sheet}", |
| 853 | file=sys.stderr, |
| 854 | ) |
| 855 | return ( |
| 856 | pool, |
| 857 | review_sheet, |
| 858 | preview_errors, |
| 859 | candidate_total, |
| 860 | has_more_candidates, |
| 861 | ) |
| 862 | |
| 863 | |
| 864 | def search_and_download( |
| 865 | providers: list[str], |
| 866 | request: ImageSearchRequest, |
| 867 | *, |
| 868 | output_path: Path, |
| 869 | strict_no_attribution: bool, |
| 870 | save_candidates: bool = False, |
| 871 | max_candidates: int = DEFAULT_CANDIDATE_PAGE_SIZE, |
| 872 | candidate_page: int = 1, |
| 873 | provider_is_explicit: bool = False, |
| 874 | ) -> SearchDownloadResult: |
| 875 | """Search candidates, then either shortlist previews or download one image. |
| 876 | |
| 877 | By default the best metadata match is downloaded. When ``save_candidates`` |
| 878 | is true, only metadata-qualified preview thumbnails are saved; the original |
| 879 | is deferred until the reviewer selects it with ``--promote``. |
| 880 | |
| 881 | Returns a structured result so batch mode can keep transient/provider |
| 882 | failures retryable while treating a complete no-match as terminal. |
| 883 | ``provider_is_explicit`` distinguishes a required provider from an optional |
| 884 | member of the default fallback chain. |
| 885 | """ |
| 886 | if max_candidates < 0: |
| 887 | raise ValueError("max_candidates must be zero or a positive integer") |
| 888 | if candidate_page < 1: |
| 889 | raise ValueError("candidate_page must be a positive integer") |
| 890 | if max_candidates == 0 and candidate_page != 1: |
| 891 | raise ValueError("candidate_page must be 1 when max_candidates is 0") |
| 892 | |
| 893 | license_filters: list[str] = ( |
| 894 | ["no-attribution-only"] if strict_no_attribution else ["all"] |
| 895 | ) |
| 896 | |
| 897 | provider_errors: list[str] = [] |
| 898 | download_errors: list[str] = [] |
| 899 | quality_rejections = 0 |
| 900 | request_variants = _search_request_variants(request) |
| 901 | scorer = score_review_candidate if save_candidates else score_candidate |
| 902 | |
| 903 | for stage in license_filters: |
| 904 | ranked: list[tuple[float, str, AssetCandidate]] = [] |
| 905 | for provider_name in providers: |
| 906 | provider_ranked: list[tuple[float, str, AssetCandidate]] = [] |
| 907 | for query_request in request_variants: |
| 908 | print( |
| 909 | f" -> trying {provider_name} ({stage}): " |
| 910 | f"{query_request.query!r}", |
| 911 | file=sys.stderr, |
| 912 | ) |
| 913 | candidates, provider_error = _try_provider( |
| 914 | provider_name, |
| 915 | query_request, |
| 916 | stage, |
| 917 | provider_is_explicit=provider_is_explicit, |
| 918 | ) |
| 919 | if provider_error: |
| 920 | provider_errors.append(provider_error) |
| 921 | break |
| 922 | if candidates is None: |
| 923 | break |
| 924 | for candidate in candidates: |
| 925 | candidate.discovery_query = query_request.query |
| 926 | score = scorer(candidate, query_request) |
| 927 | if score != float("-inf"): |
| 928 | provider_ranked.append( |
| 929 | (score, provider_name, candidate) |
| 930 | ) |
| 931 | if not provider_ranked: |
| 932 | reason = "query" |
| 933 | if request.required_terms: |
| 934 | reason += f" / required_terms={list(request.required_terms)}" |
| 935 | print( |
| 936 | f" no candidate matched {reason}; trying next provider/stage", |
| 937 | file=sys.stderr, |
| 938 | ) |
| 939 | continue |
| 940 | ranked.extend(provider_ranked) |
| 941 | |
| 942 | sorted_ranked = _dedupe_ranked_candidates(ranked) |
| 943 | |
| 944 | # --- Thumbnail-only visual selection (no original download) --- |
| 945 | if save_candidates and sorted_ranked: |
| 946 | stem = Path(output_path).stem |
| 947 | try: |
| 948 | ( |
| 949 | pool, |
| 950 | review_sheet, |
| 951 | preview_errors, |
| 952 | candidate_total, |
| 953 | has_more_candidates, |
| 954 | ) = _save_candidate_thumbnails( |
| 955 | sorted_ranked, |
| 956 | output_path.parent, |
| 957 | stem, |
| 958 | request, |
| 959 | stage, |
| 960 | max_candidates=max_candidates, |
| 961 | candidate_page=candidate_page, |
| 962 | ) |
| 963 | except Exception as exc: |
| 964 | if not _is_recoverable_image_error(exc): |
| 965 | raise |
| 966 | print( |
| 967 | f" warning: candidate pool could not be saved: {exc}", |
| 968 | file=sys.stderr, |
| 969 | ) |
| 970 | else: |
| 971 | download_errors.extend(preview_errors) |
| 972 | if pool: |
| 973 | return SearchDownloadResult( |
| 974 | stage=stage, |
| 975 | output_path=output_path, |
| 976 | selection_required=True, |
| 977 | candidate_count=len(pool), |
| 978 | candidate_total=candidate_total, |
| 979 | candidate_page=candidate_page, |
| 980 | has_more_candidates=has_more_candidates, |
| 981 | review_sheet=review_sheet, |
| 982 | ) |
| 983 | if candidate_total and not preview_errors: |
| 984 | return SearchDownloadResult( |
| 985 | stage=stage, |
| 986 | output_path=output_path, |
| 987 | candidate_total=candidate_total, |
| 988 | candidate_page=candidate_page, |
| 989 | failure_kind=SEARCH_FAILURE_NO_MATCH, |
| 990 | error=( |
| 991 | f"candidate page {candidate_page} is past the " |
| 992 | f"{candidate_total} available candidate(s)" |
| 993 | ), |
| 994 | ) |
| 995 | continue |
| 996 | |
| 997 | # --- Pick the best downloadable candidate --- |
| 998 | for _score, provider_name, candidate in sorted_ranked: |
| 999 | try: |
| 1000 | staged_path, actual_dimensions = _stage_validated_image( |
| 1001 | candidate.download_url, |
| 1002 | output_path, |
| 1003 | min_width=request.min_width, |
| 1004 | min_height=request.min_height, |
| 1005 | ) |
| 1006 | return SearchDownloadResult( |
| 1007 | candidate=candidate, |
| 1008 | provider_name=provider_name, |
| 1009 | stage=stage, |
| 1010 | actual_dimensions=actual_dimensions, |
| 1011 | staged_path=staged_path, |
| 1012 | output_path=output_path, |
| 1013 | ) |
| 1014 | except DownloadQualityError: |
| 1015 | quality_rejections += 1 |
| 1016 | continue |
| 1017 | except Exception as exc: |
| 1018 | if not _is_recoverable_image_error(exc): |
| 1019 | raise |
| 1020 | print( |
| 1021 | f" download failed for {candidate.title!r}: {exc}", |
| 1022 | file=sys.stderr, |
| 1023 | ) |
| 1024 | download_errors.append(f"{provider_name}/{candidate.title}: {exc}") |
| 1025 | continue |
| 1026 | |
| 1027 | retryable_errors = provider_errors + download_errors |
| 1028 | if retryable_errors: |
| 1029 | return SearchDownloadResult( |
| 1030 | failure_kind=SEARCH_FAILURE_RETRYABLE, |
| 1031 | error="; ".join(retryable_errors)[:500], |
| 1032 | ) |
| 1033 | |
| 1034 | detail = "no acceptable candidate across all providers/stages" |
| 1035 | if quality_rejections: |
| 1036 | detail += f" ({quality_rejections} candidate(s) failed actual-size/readability gates)" |
| 1037 | return SearchDownloadResult( |
| 1038 | failure_kind=SEARCH_FAILURE_NO_MATCH, |
| 1039 | error=detail, |
| 1040 | ) |
| 1041 | |
| 1042 | |
| 1043 | # --------------------------------------------------------------------------- |
| 1044 | # Manifest |
| 1045 | # --------------------------------------------------------------------------- |
| 1046 | |
| 1047 | |
| 1048 | def default_manifest_path(output_dir: str) -> Path: |
| 1049 | return Path(output_dir) / "image_sources.json" |
| 1050 | |
| 1051 | |
| 1052 | def _validate_bare_filename(value: str, *, field_name: str = "filename") -> str: |
| 1053 | """Require a bare filename with no absolute or parent path components.""" |
| 1054 | if ( |
| 1055 | not value.strip() |
| 1056 | or value in {".", ".."} |
| 1057 | or "/" in value |
| 1058 | or "\\" in value |
| 1059 | or ":" in value |
| 1060 | or Path(value).is_absolute() |
| 1061 | ): |
| 1062 | raise ValueError( |
| 1063 | f"{field_name} must be a bare filename without path components: {value!r}" |
| 1064 | ) |
| 1065 | return value |
| 1066 | |
| 1067 | |
| 1068 | def _measure_actual_image(path: Path) -> Optional[tuple[int, int]]: |
| 1069 | """Return ``(width, height)`` of the file actually saved at ``path``. |
| 1070 | |
| 1071 | Upstream metadata (``candidate.width``/``height``) describes the |
| 1072 | original image on the provider's server, which may differ from what |
| 1073 | we are allowed to download — for example, second-tier sources |
| 1074 | aggregated by Openverse (rawpixel etc.) often only expose a |
| 1075 | 1024px-wide preview. The Executor needs to know what is actually on |
| 1076 | disk for layout purposes; this function provides that ground truth. |
| 1077 | |
| 1078 | Returns ``None`` if Pillow is unavailable or the file is unreadable. |
| 1079 | """ |
| 1080 | try: |
| 1081 | from PIL import Image, ImageOps # type: ignore |
| 1082 | except ImportError: |
| 1083 | return None |
| 1084 | try: |
| 1085 | with Image.open(path) as im: |
| 1086 | oriented = ImageOps.exif_transpose(im) |
| 1087 | try: |
| 1088 | return int(oriented.width), int(oriented.height) |
| 1089 | finally: |
| 1090 | if oriented is not im: |
| 1091 | oriented.close() |
| 1092 | except Exception as exc: |
| 1093 | if not _is_recoverable_image_error(exc): |
| 1094 | raise |
| 1095 | return None |
| 1096 | |
| 1097 | |
| 1098 | def _candidate_to_manifest_item( |
| 1099 | candidate: AssetCandidate, |
| 1100 | args: argparse.Namespace, |
| 1101 | *, |
| 1102 | provider_name: str, |
| 1103 | stage: str, |
| 1104 | actual_dimensions: Optional[tuple[int, int]] = None, |
| 1105 | selection_method: str = "metadata-ranked", |
| 1106 | ) -> dict: |
| 1107 | """Build the manifest entry. |
| 1108 | |
| 1109 | ``width`` / ``height`` reflect the file actually saved to disk |
| 1110 | (measured by Pillow after download). The upstream-claimed dimensions |
| 1111 | are only kept under ``metadata_dimensions`` when they disagree with |
| 1112 | reality, which is the only case where this distinction matters. |
| 1113 | """ |
| 1114 | if actual_dimensions is not None: |
| 1115 | width, height = actual_dimensions |
| 1116 | else: |
| 1117 | width, height = candidate.width, candidate.height |
| 1118 | |
| 1119 | item = { |
| 1120 | "filename": args.filename, |
| 1121 | "slide": args.slide, |
| 1122 | "purpose": args.purpose, |
| 1123 | "search_query": args.query, |
| 1124 | "matched_query": candidate.discovery_query or args.query, |
| 1125 | "orientation": args.orientation, |
| 1126 | "provider": provider_name, |
| 1127 | "stage": stage, |
| 1128 | "title": candidate.title, |
| 1129 | "author": candidate.author, |
| 1130 | "source_page_url": candidate.source_page_url, |
| 1131 | "download_url": candidate.download_url, |
| 1132 | "license_name": candidate.license_name, |
| 1133 | "license_url": candidate.license_url, |
| 1134 | "license_tier": candidate.license_tier, |
| 1135 | "attribution_required": candidate.license_tier == "attribution-required", |
| 1136 | "width": width, |
| 1137 | "height": height, |
| 1138 | "attribution_text": build_attribution_text(args.filename, candidate), |
| 1139 | "selection_method": selection_method, |
| 1140 | "status": "sourced", |
| 1141 | } |
| 1142 | required_terms = _parse_required_terms( |
| 1143 | getattr(args, "required_terms", None) or getattr(args, "require_terms", None) |
| 1144 | ) |
| 1145 | if required_terms: |
| 1146 | item["required_terms"] = list(required_terms) |
| 1147 | query_variants = _parse_query_variants( |
| 1148 | getattr(args, "query_variants", None) |
| 1149 | or getattr(args, "query_variant", None) |
| 1150 | ) |
| 1151 | if query_variants: |
| 1152 | item["query_variants"] = list(query_variants) |
| 1153 | |
| 1154 | # Only carry upstream-claimed dimensions when they differ — this flags |
| 1155 | # cases where the provider returned a preview rather than the original. |
| 1156 | if ( |
| 1157 | actual_dimensions is not None |
| 1158 | and candidate.width |
| 1159 | and candidate.height |
| 1160 | and (candidate.width, candidate.height) != actual_dimensions |
| 1161 | ): |
| 1162 | item["metadata_dimensions"] = { |
| 1163 | "width": candidate.width, |
| 1164 | "height": candidate.height, |
| 1165 | "note": "upstream-reported size; actual downloaded file is smaller (likely a preview)", |
| 1166 | } |
| 1167 | |
| 1168 | return item |
| 1169 | |
| 1170 | |
| 1171 | def _read_existing_manifest(path: Path) -> dict: |
| 1172 | if not path.exists(): |
| 1173 | return {} |
| 1174 | try: |
| 1175 | payload = json.loads(path.read_text(encoding="utf-8")) |
| 1176 | except (OSError, json.JSONDecodeError) as exc: |
| 1177 | raise RuntimeError( |
| 1178 | f"existing image sources manifest is unreadable: {path} ({exc}); " |
| 1179 | "repair or restore it before continuing" |
| 1180 | ) from exc |
| 1181 | if not isinstance(payload, dict): |
| 1182 | raise RuntimeError( |
| 1183 | f"existing image sources manifest must be a JSON object: {path}" |
| 1184 | ) |
| 1185 | items = payload.get("items") |
| 1186 | if not isinstance(items, list): |
| 1187 | raise RuntimeError( |
| 1188 | f"existing image sources manifest must contain an 'items' array: {path}" |
| 1189 | ) |
| 1190 | if any(not isinstance(item, dict) for item in items): |
| 1191 | raise RuntimeError( |
| 1192 | f"existing image sources manifest contains a non-object item: {path}" |
| 1193 | ) |
| 1194 | seen_filenames: dict[str, str] = {} |
| 1195 | for index, item in enumerate(items): |
| 1196 | filename = item.get("filename") |
| 1197 | if not isinstance(filename, str): |
| 1198 | raise RuntimeError( |
| 1199 | f"existing image sources manifest items[{index}].filename " |
| 1200 | f"must be a non-empty bare filename: {path}" |
| 1201 | ) |
| 1202 | try: |
| 1203 | _validate_bare_filename(filename) |
| 1204 | except ValueError as exc: |
| 1205 | raise RuntimeError( |
| 1206 | f"existing image sources manifest items[{index}]: {exc}: {path}" |
| 1207 | ) from exc |
| 1208 | normalized_filename = filename.casefold() |
| 1209 | if normalized_filename in seen_filenames: |
| 1210 | raise RuntimeError( |
| 1211 | f"existing image sources manifest filename {filename!r} conflicts " |
| 1212 | f"with {seen_filenames[normalized_filename]!r} " |
| 1213 | f"(case-insensitive): {path}" |
| 1214 | ) |
| 1215 | seen_filenames[normalized_filename] = filename |
| 1216 | return payload |
| 1217 | |
| 1218 | |
| 1219 | def _write_json_atomic(path: str | Path, payload: dict) -> Path: |
| 1220 | """Write JSON through a same-directory temporary file and atomic rename.""" |
| 1221 | target = ensure_json_parent(path) |
| 1222 | fd, tmp_path = tempfile.mkstemp( |
| 1223 | prefix=target.stem + ".", suffix=".tmp", dir=str(target.parent) |
| 1224 | ) |
| 1225 | try: |
| 1226 | with os.fdopen(fd, "w", encoding="utf-8") as handle: |
| 1227 | json.dump(payload, handle, ensure_ascii=False, indent=2) |
| 1228 | handle.write("\n") |
| 1229 | os.replace(tmp_path, target) |
| 1230 | except Exception: |
| 1231 | try: |
| 1232 | os.unlink(tmp_path) |
| 1233 | except OSError: |
| 1234 | pass |
| 1235 | raise |
| 1236 | return target |
| 1237 | |
| 1238 | |
| 1239 | def write_sources_manifest(path: Path, item: dict) -> Path: |
| 1240 | """Append ``item`` to the manifest at ``path``, replacing any prior |
| 1241 | entry that targets the same filename.""" |
| 1242 | manifest_path = Path(path) |
| 1243 | payload = _read_existing_manifest(manifest_path) |
| 1244 | filename = item.get("filename") |
| 1245 | if not isinstance(filename, str): |
| 1246 | raise RuntimeError("new image source item requires a string filename") |
| 1247 | try: |
| 1248 | _validate_bare_filename(filename) |
| 1249 | except ValueError as exc: |
| 1250 | raise RuntimeError(f"new image source item: {exc}") from exc |
| 1251 | |
| 1252 | items: list[dict] = list(payload.get("items") or []) |
| 1253 | normalized_filename = filename.casefold() |
| 1254 | differently_cased = next( |
| 1255 | ( |
| 1256 | existing["filename"] |
| 1257 | for existing in items |
| 1258 | if isinstance(existing.get("filename"), str) |
| 1259 | and existing["filename"].casefold() == normalized_filename |
| 1260 | and existing["filename"] != filename |
| 1261 | ), |
| 1262 | None, |
| 1263 | ) |
| 1264 | if differently_cased is not None: |
| 1265 | raise RuntimeError( |
| 1266 | f"new image source filename {filename!r} conflicts with existing " |
| 1267 | f"{differently_cased!r} (case-insensitive)" |
| 1268 | ) |
| 1269 | items = [ |
| 1270 | i |
| 1271 | for i in items |
| 1272 | if not isinstance(i.get("filename"), str) |
| 1273 | or i["filename"].casefold() != normalized_filename |
| 1274 | ] |
| 1275 | items.append(item) |
| 1276 | |
| 1277 | payload["items"] = items |
| 1278 | payload["generated_at"] = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z") |
| 1279 | payload.setdefault( |
| 1280 | "license_verification", |
| 1281 | "provider metadata used; manual review recommended for external delivery", |
| 1282 | ) |
| 1283 | |
| 1284 | return _write_json_atomic(manifest_path, payload) |
| 1285 | |
| 1286 | |
| 1287 | # --------------------------------------------------------------------------- |
| 1288 | # Promote: replace primary image with a candidate |
| 1289 | # --------------------------------------------------------------------------- |
| 1290 | |
| 1291 | |
| 1292 | def promote_candidate( |
| 1293 | output_dir: Path, |
| 1294 | target_filename: str, |
| 1295 | candidate_filename: str, |
| 1296 | manifest_path: Optional[Path] = None, |
| 1297 | queries_manifest_path: Optional[Path] = None, |
| 1298 | ) -> int: |
| 1299 | """Download one selected candidate original and commit its provenance. |
| 1300 | |
| 1301 | Steps: |
| 1302 | 1. Resolve the selected thumbnail's original URL from candidates.json |
| 1303 | 2. Download and validate exactly that one original |
| 1304 | 3. Replace the canonical image and its provenance as one rollback unit |
| 1305 | 4. Advance ``candidates.json`` only after provenance succeeds |
| 1306 | """ |
| 1307 | target_filename = _validate_bare_filename( |
| 1308 | target_filename, field_name="target filename" |
| 1309 | ) |
| 1310 | candidate_filename = _validate_bare_filename( |
| 1311 | candidate_filename, field_name="candidate filename" |
| 1312 | ) |
| 1313 | mpath = manifest_path or default_manifest_path(str(output_dir)) |
| 1314 | stem = Path(target_filename).stem |
| 1315 | cand_dir = output_dir / "candidates" / stem |
| 1316 | cand_meta_path = cand_dir / "candidates.json" |
| 1317 | |
| 1318 | if not cand_meta_path.is_file(): |
| 1319 | print(f"Error: {cand_meta_path} not found.", file=sys.stderr) |
| 1320 | return 1 |
| 1321 | |
| 1322 | try: |
| 1323 | meta = json.loads(cand_meta_path.read_text(encoding="utf-8")) |
| 1324 | except (OSError, json.JSONDecodeError) as exc: |
| 1325 | print(f"Error: cannot read {cand_meta_path}: {exc}", file=sys.stderr) |
| 1326 | return 1 |
| 1327 | if not isinstance(meta, dict) or not isinstance(meta.get("candidates"), list): |
| 1328 | print( |
| 1329 | f"Error: {cand_meta_path} must contain a candidates array.", |
| 1330 | file=sys.stderr, |
| 1331 | ) |
| 1332 | return 1 |
| 1333 | candidates = meta.get("candidates", []) |
| 1334 | |
| 1335 | entry = next( |
| 1336 | ( |
| 1337 | candidate |
| 1338 | for candidate in candidates |
| 1339 | if isinstance(candidate, dict) |
| 1340 | and candidate.get("filename") == candidate_filename |
| 1341 | ), |
| 1342 | None, |
| 1343 | ) |
| 1344 | if entry is None: |
| 1345 | names = [ |
| 1346 | str(candidate.get("filename")) |
| 1347 | for candidate in candidates |
| 1348 | if isinstance(candidate, dict) and candidate.get("filename") |
| 1349 | ] |
| 1350 | print( |
| 1351 | f"Error: '{candidate_filename}' not found. Available: {', '.join(names)}", |
| 1352 | file=sys.stderr, |
| 1353 | ) |
| 1354 | return 1 |
| 1355 | |
| 1356 | dst_path = output_dir / target_filename |
| 1357 | declared_target = meta.get("target_filename") |
| 1358 | if declared_target != target_filename: |
| 1359 | print( |
| 1360 | f"Error: candidate pool targets {declared_target!r}, not " |
| 1361 | f"{target_filename!r}.", |
| 1362 | file=sys.stderr, |
| 1363 | ) |
| 1364 | return 1 |
| 1365 | |
| 1366 | request_meta = meta.get("request") |
| 1367 | if not isinstance(request_meta, dict): |
| 1368 | request_meta = {} |
| 1369 | min_width = int(request_meta.get("min_width") or 0) |
| 1370 | min_height = int(request_meta.get("min_height") or 0) |
| 1371 | selected_candidate = AssetCandidate( |
| 1372 | provider=str(entry.get("provider") or ""), |
| 1373 | title=str(entry.get("title") or ""), |
| 1374 | source_page_url=str(entry.get("source_page_url") or ""), |
| 1375 | license_name=str(entry.get("license_name") or ""), |
| 1376 | license_url=str(entry.get("license_url") or ""), |
| 1377 | license_tier=str(entry.get("license_tier") or ""), |
| 1378 | width=int(entry.get("width") or 0), |
| 1379 | height=int(entry.get("height") or 0), |
| 1380 | download_url=str(entry.get("download_url") or ""), |
| 1381 | preview_url=str(entry.get("preview_url") or ""), |
| 1382 | discovery_query=str(entry.get("matched_query") or ""), |
| 1383 | author=str(entry.get("author") or ""), |
| 1384 | ) |
| 1385 | if not selected_candidate.download_url: |
| 1386 | print( |
| 1387 | f"Error: {candidate_filename} has no original download URL.", |
| 1388 | file=sys.stderr, |
| 1389 | ) |
| 1390 | return 1 |
| 1391 | |
| 1392 | staged_path: Optional[Path] = None |
| 1393 | try: |
| 1394 | legacy_src_path = cand_dir / candidate_filename |
| 1395 | if ( |
| 1396 | meta.get("candidate_storage") != "thumbnail-only" |
| 1397 | and legacy_src_path.is_file() |
| 1398 | ): |
| 1399 | staged_path, actual_dimensions = _stage_validated_candidate_copy( |
| 1400 | legacy_src_path, |
| 1401 | dst_path, |
| 1402 | min_width=min_width, |
| 1403 | min_height=min_height, |
| 1404 | ) |
| 1405 | else: |
| 1406 | staged_path, actual_dimensions = _stage_validated_image( |
| 1407 | selected_candidate.download_url, |
| 1408 | dst_path, |
| 1409 | min_width=min_width, |
| 1410 | min_height=min_height, |
| 1411 | ) |
| 1412 | |
| 1413 | item_args = argparse.Namespace( |
| 1414 | filename=target_filename, |
| 1415 | slide=str(request_meta.get("slide") or ""), |
| 1416 | purpose=str(request_meta.get("purpose") or ""), |
| 1417 | query=str(request_meta.get("query") or ""), |
| 1418 | orientation=str(request_meta.get("orientation") or "any"), |
| 1419 | required_terms=_parse_required_terms( |
| 1420 | request_meta.get("required_terms") |
| 1421 | ), |
| 1422 | query_variants=_parse_query_variants( |
| 1423 | request_meta.get("query_variants") |
| 1424 | ), |
| 1425 | ) |
| 1426 | source_item = _candidate_to_manifest_item( |
| 1427 | selected_candidate, |
| 1428 | item_args, |
| 1429 | provider_name=selected_candidate.provider, |
| 1430 | stage=str(meta.get("license_stage") or "all"), |
| 1431 | actual_dimensions=actual_dimensions, |
| 1432 | selection_method="visual-thumbnail", |
| 1433 | ) |
| 1434 | source_item["status"] = "promoted" |
| 1435 | |
| 1436 | _commit_staged_image( |
| 1437 | staged_path, |
| 1438 | dst_path, |
| 1439 | lambda: write_sources_manifest(mpath, source_item), |
| 1440 | ) |
| 1441 | except Exception as exc: |
| 1442 | if not _is_recoverable_image_error(exc): |
| 1443 | raise |
| 1444 | print(f"Error: candidate promotion failed: {exc}", file=sys.stderr) |
| 1445 | return 1 |
| 1446 | finally: |
| 1447 | if staged_path is not None: |
| 1448 | try: |
| 1449 | staged_path.unlink(missing_ok=True) |
| 1450 | except OSError: |
| 1451 | pass |
| 1452 | |
| 1453 | print(f" promoted: {candidate_filename} → {target_filename}", file=sys.stderr) |
| 1454 | print(f" manifest updated: {mpath}", file=sys.stderr) |
| 1455 | |
| 1456 | # The selection marker must never move ahead of canonical provenance. |
| 1457 | meta["selected"] = candidate_filename |
| 1458 | try: |
| 1459 | _write_json_atomic(cand_meta_path, meta) |
| 1460 | except OSError as exc: |
| 1461 | print( |
| 1462 | f"Error: image was promoted, but {cand_meta_path} could not be updated: " |
| 1463 | f"{exc}", |
| 1464 | file=sys.stderr, |
| 1465 | ) |
| 1466 | return 1 |
| 1467 | |
| 1468 | if queries_manifest_path is not None: |
| 1469 | try: |
| 1470 | queries_manifest = load_search_manifest(str(queries_manifest_path)) |
| 1471 | query_row = next( |
| 1472 | ( |
| 1473 | row |
| 1474 | for row in queries_manifest["items"] |
| 1475 | if row["filename"].casefold() == target_filename.casefold() |
| 1476 | ), |
| 1477 | None, |
| 1478 | ) |
| 1479 | if query_row is None or query_row["filename"] != target_filename: |
| 1480 | raise RuntimeError( |
| 1481 | f"no exact query row for {target_filename!r}" |
| 1482 | ) |
| 1483 | query_row["status"] = SEARCH_STATUS_SOURCED |
| 1484 | query_row["provider"] = selected_candidate.provider |
| 1485 | query_row["license_tier"] = selected_candidate.license_tier |
| 1486 | query_row.pop("last_error", None) |
| 1487 | _clear_candidate_selection_outputs(query_row) |
| 1488 | save_search_manifest(str(queries_manifest_path), queries_manifest) |
| 1489 | except (OSError, RuntimeError, ValueError) as exc: |
| 1490 | print( |
| 1491 | f"Error: image was promoted, but query status could not be " |
| 1492 | f"updated: {exc}", |
| 1493 | file=sys.stderr, |
| 1494 | ) |
| 1495 | return 1 |
| 1496 | print( |
| 1497 | f" query status updated: {queries_manifest_path}", |
| 1498 | file=sys.stderr, |
| 1499 | ) |
| 1500 | |
| 1501 | review = _write_review_copy(dst_path, output_dir / ".review", target_filename) |
| 1502 | if review is not None: |
| 1503 | print(f" review copy: {review}", file=sys.stderr) |
| 1504 | return 0 |
| 1505 | |
| 1506 | |
| 1507 | # --------------------------------------------------------------------------- |
| 1508 | # Manual URL replacement (model-agnostic) |
| 1509 | # --------------------------------------------------------------------------- |
| 1510 | |
| 1511 | |
| 1512 | def fetch_url_replace( |
| 1513 | output_dir: Path, |
| 1514 | target_filename: str, |
| 1515 | url: str, |
| 1516 | manifest_path: Optional[Path] = None, |
| 1517 | *, |
| 1518 | slide: str = "", |
| 1519 | purpose: str = "", |
| 1520 | search_query: str = "", |
| 1521 | orientation: str = "", |
| 1522 | required_terms: tuple[str, ...] = (), |
| 1523 | min_width: int = 1200, |
| 1524 | min_height: int = 800, |
| 1525 | ) -> int: |
| 1526 | """Download a directly selected image URL into the target and record it. |
| 1527 | |
| 1528 | The model-agnostic manual path: when an automated best match is not |
| 1529 | suitable (or the running model cannot see images at all), a human finds a |
| 1530 | good image, passes its URL, and it replaces the target. License is unknown |
| 1531 | for an arbitrary URL, so the manifest marks it ``manual`` and notes that |
| 1532 | verifying usage rights is the user's responsibility. |
| 1533 | """ |
| 1534 | target_filename = _validate_bare_filename( |
| 1535 | target_filename, field_name="target filename" |
| 1536 | ) |
| 1537 | mpath = manifest_path or default_manifest_path(str(output_dir)) |
| 1538 | try: |
| 1539 | existing_manifest = _read_existing_manifest(mpath) |
| 1540 | except RuntimeError as exc: |
| 1541 | print(f"Error: {exc}", file=sys.stderr) |
| 1542 | return 1 |
| 1543 | prior = next( |
| 1544 | ( |
| 1545 | item |
| 1546 | for item in existing_manifest.get("items", []) |
| 1547 | if isinstance(item.get("filename"), str) |
| 1548 | and item["filename"].casefold() == target_filename.casefold() |
| 1549 | ), |
| 1550 | {}, |
| 1551 | ) |
| 1552 | if prior and prior["filename"] != target_filename: |
| 1553 | print( |
| 1554 | f"Error: target filename casing {target_filename!r} conflicts with " |
| 1555 | f"provenance filename {prior['filename']!r}.", |
| 1556 | file=sys.stderr, |
| 1557 | ) |
| 1558 | return 1 |
| 1559 | try: |
| 1560 | inherited_required_terms = _parse_required_terms( |
| 1561 | prior.get("required_terms") |
| 1562 | ) |
| 1563 | except ValueError as exc: |
| 1564 | print( |
| 1565 | f"Error: existing provenance has invalid required_terms: {exc}", |
| 1566 | file=sys.stderr, |
| 1567 | ) |
| 1568 | return 1 |
| 1569 | final_required_terms = inherited_required_terms or required_terms |
| 1570 | |
| 1571 | output_dir.mkdir(parents=True, exist_ok=True) |
| 1572 | dst_path = output_dir / target_filename |
| 1573 | try: |
| 1574 | staged_path, actual_dim = _stage_validated_image( |
| 1575 | url, |
| 1576 | dst_path, |
| 1577 | min_width=min_width, |
| 1578 | min_height=min_height, |
| 1579 | enforce_thumbnail_floor=False, |
| 1580 | ) |
| 1581 | except ( |
| 1582 | DownloadQualityError, |
| 1583 | requests.RequestException, |
| 1584 | OSError, |
| 1585 | RuntimeError, |
| 1586 | ValueError, |
| 1587 | ) as exc: |
| 1588 | print(f"Error: failed to download {url}: {exc}", file=sys.stderr) |
| 1589 | return 1 |
| 1590 | |
| 1591 | # Inherit page context (which slide / purpose / query this image serves) |
| 1592 | # from the entry being replaced; override only source / license / size / |
| 1593 | # status so the audit trail survives a manual swap. |
| 1594 | item = { |
| 1595 | "filename": target_filename, |
| 1596 | "slide": prior.get("slide") or slide, |
| 1597 | "purpose": prior.get("purpose") or purpose, |
| 1598 | "search_query": prior.get("search_query") or search_query, |
| 1599 | "orientation": prior.get("orientation") or orientation, |
| 1600 | "provider": "manual", |
| 1601 | "title": "", |
| 1602 | "author": "", |
| 1603 | "source_page_url": url, |
| 1604 | "download_url": url, |
| 1605 | "license_name": "unverified — direct URL", |
| 1606 | "license_url": "", |
| 1607 | "license_tier": "manual", |
| 1608 | "attribution_required": False, |
| 1609 | "width": actual_dim[0], |
| 1610 | "height": actual_dim[1], |
| 1611 | "attribution_text": "", |
| 1612 | "status": "manual", |
| 1613 | "note": ( |
| 1614 | "Direct image URL; verifying usage rights is the user's " |
| 1615 | "responsibility." |
| 1616 | ), |
| 1617 | } |
| 1618 | if final_required_terms: |
| 1619 | item["required_terms"] = list(final_required_terms) |
| 1620 | |
| 1621 | try: |
| 1622 | written = _commit_staged_image( |
| 1623 | staged_path, |
| 1624 | dst_path, |
| 1625 | lambda: write_sources_manifest(mpath, item), |
| 1626 | ) |
| 1627 | except ( |
| 1628 | OSError, |
| 1629 | RuntimeError, |
| 1630 | ValueError, |
| 1631 | ) as exc: |
| 1632 | print( |
| 1633 | f"Error: failed to replace {target_filename} and record provenance: {exc}", |
| 1634 | file=sys.stderr, |
| 1635 | ) |
| 1636 | return 1 |
| 1637 | finally: |
| 1638 | try: |
| 1639 | staged_path.unlink(missing_ok=True) |
| 1640 | except OSError: |
| 1641 | pass |
| 1642 | |
| 1643 | print(f" fetched: {url} -> {target_filename}", file=sys.stderr) |
| 1644 | print(f" manifest updated: {written}", file=sys.stderr) |
| 1645 | review = _write_review_copy(dst_path, output_dir / ".review", target_filename) |
| 1646 | if review is not None: |
| 1647 | print(f" review copy: {review}", file=sys.stderr) |
| 1648 | return 0 |
| 1649 | |
| 1650 | |
| 1651 | # --------------------------------------------------------------------------- |
| 1652 | # Batch mode (`--batch image_queries.json`) |
| 1653 | # --------------------------------------------------------------------------- |
| 1654 | |
| 1655 | |
| 1656 | def load_search_manifest(path: str) -> dict: |
| 1657 | """Load and validate an ``image_queries.json`` batch manifest. |
| 1658 | |
| 1659 | Schema (top level): ``{"items": [ ... ]}``. Each item requires |
| 1660 | ``filename``, ``query``, ``status``. Optional per-item overrides: |
| 1661 | ``query_variants``, ``candidate_page``, ``slide``, ``purpose``, |
| 1662 | ``orientation``, ``provider``, ``strict_no_attribution``, ``min_width``, |
| 1663 | ``min_height``, ``last_error``. |
| 1664 | """ |
| 1665 | try: |
| 1666 | data = json.loads(Path(path).read_text(encoding="utf-8")) |
| 1667 | except OSError as exc: |
| 1668 | raise ValueError(f"Cannot read {path}: {exc}") from exc |
| 1669 | except json.JSONDecodeError as exc: |
| 1670 | raise ValueError( |
| 1671 | f"Invalid JSON in {path}: {exc.msg} " |
| 1672 | f"(line {exc.lineno}, col {exc.colno})" |
| 1673 | ) from exc |
| 1674 | |
| 1675 | if not isinstance(data, dict): |
| 1676 | raise ValueError( |
| 1677 | f"{path}: top level must be a JSON object, got {type(data).__name__}" |
| 1678 | ) |
| 1679 | |
| 1680 | items = data.get("items") |
| 1681 | if not isinstance(items, list) or not items: |
| 1682 | raise ValueError(f"{path}: 'items' must be a non-empty array") |
| 1683 | |
| 1684 | seen_filenames: dict[str, str] = {} |
| 1685 | for i, item in enumerate(items): |
| 1686 | prefix = f"{path}: items[{i}]" |
| 1687 | if not isinstance(item, dict): |
| 1688 | raise ValueError(f"{prefix} must be an object") |
| 1689 | for field in SEARCH_REQUIRED_ITEM_FIELDS: |
| 1690 | if field not in item: |
| 1691 | raise ValueError(f"{prefix} missing required field '{field}'") |
| 1692 | if not isinstance(item[field], str) or not item[field].strip(): |
| 1693 | raise ValueError( |
| 1694 | f"{prefix} field '{field}' must be a non-empty string" |
| 1695 | ) |
| 1696 | if item["status"] not in SEARCH_VALID_STATUSES: |
| 1697 | raise ValueError( |
| 1698 | f"{prefix} status '{item['status']}' is invalid. " |
| 1699 | f"Valid: {sorted(SEARCH_VALID_STATUSES)}" |
| 1700 | ) |
| 1701 | if "required_terms" in item: |
| 1702 | try: |
| 1703 | _parse_required_terms(item["required_terms"]) |
| 1704 | except ValueError as exc: |
| 1705 | raise ValueError(f"{prefix} {exc}") from exc |
| 1706 | if "query_variants" in item: |
| 1707 | try: |
| 1708 | _parse_query_variants(item["query_variants"]) |
| 1709 | except ValueError as exc: |
| 1710 | raise ValueError(f"{prefix} {exc}") from exc |
| 1711 | if "candidate_page" in item: |
| 1712 | candidate_page = item["candidate_page"] |
| 1713 | if ( |
| 1714 | not isinstance(candidate_page, int) |
| 1715 | or isinstance(candidate_page, bool) |
| 1716 | or candidate_page < 1 |
| 1717 | ): |
| 1718 | raise ValueError( |
| 1719 | f"{prefix} field 'candidate_page' must be a positive integer" |
| 1720 | ) |
| 1721 | for dimension_field in ("min_width", "min_height"): |
| 1722 | if dimension_field not in item: |
| 1723 | continue |
| 1724 | value = item[dimension_field] |
| 1725 | if ( |
| 1726 | not isinstance(value, int) |
| 1727 | or isinstance(value, bool) |
| 1728 | or value < 1 |
| 1729 | ): |
| 1730 | raise ValueError( |
| 1731 | f"{prefix} field '{dimension_field}' must be a positive integer" |
| 1732 | ) |
| 1733 | fname = item["filename"] |
| 1734 | try: |
| 1735 | _validate_bare_filename(fname) |
| 1736 | except ValueError as exc: |
| 1737 | raise ValueError(f"{prefix} {exc}") from exc |
| 1738 | normalized_filename = fname.casefold() |
| 1739 | if normalized_filename in seen_filenames: |
| 1740 | raise ValueError( |
| 1741 | f"{prefix} filename {fname!r} conflicts with " |
| 1742 | f"{seen_filenames[normalized_filename]!r} (case-insensitive)" |
| 1743 | ) |
| 1744 | seen_filenames[normalized_filename] = fname |
| 1745 | |
| 1746 | return data |
| 1747 | |
| 1748 | |
| 1749 | def save_search_manifest(path: str, data: dict) -> None: |
| 1750 | """Atomically write the batch manifest back (tmp file + rename).""" |
| 1751 | try: |
| 1752 | _write_json_atomic(path, data) |
| 1753 | except OSError as exc: |
| 1754 | raise RuntimeError( |
| 1755 | f"cannot update image query manifest {path}: {exc}" |
| 1756 | ) from exc |
| 1757 | |
| 1758 | |
| 1759 | def _resolve_search_concurrency(cli_value: Optional[int]) -> int: |
| 1760 | """CLI value wins over IMAGE_SEARCH_CONCURRENCY env; default 3.""" |
| 1761 | if cli_value is not None: |
| 1762 | return max(1, cli_value) |
| 1763 | env_val = os.environ.get("IMAGE_SEARCH_CONCURRENCY", "").strip() |
| 1764 | if env_val.isdigit(): |
| 1765 | return max(1, int(env_val)) |
| 1766 | return DEFAULT_SEARCH_CONCURRENCY |
| 1767 | |
| 1768 | |
| 1769 | def _search_one_item( |
| 1770 | item: dict, |
| 1771 | *, |
| 1772 | output_dir: Path, |
| 1773 | save_candidates: bool, |
| 1774 | max_candidates: int, |
| 1775 | default_candidate_page: int, |
| 1776 | default_provider: Optional[str], |
| 1777 | default_strict: bool, |
| 1778 | default_min_width: int, |
| 1779 | default_min_height: int, |
| 1780 | ) -> tuple[ |
| 1781 | Optional[dict], |
| 1782 | Optional[str], |
| 1783 | bool, |
| 1784 | bool, |
| 1785 | Optional[Path], |
| 1786 | Optional[Path], |
| 1787 | ]: |
| 1788 | """Run one batch item's search or thumbnail-shortlist stage. |
| 1789 | |
| 1790 | Returns ``(manifest_item, message, retryable, selection_required, |
| 1791 | staged_path, output_path)``. |
| 1792 | Only network and staged-file work happens here; canonical replacement and |
| 1793 | all manifest writes are serialized by the caller. |
| 1794 | """ |
| 1795 | filename = _validate_bare_filename(item["filename"]) |
| 1796 | orientation = item.get("orientation", "any") or "any" |
| 1797 | strict = bool(item.get("strict_no_attribution", default_strict)) |
| 1798 | required_terms = _parse_required_terms(item.get("required_terms")) |
| 1799 | candidate_page = int(item.get("candidate_page", default_candidate_page)) |
| 1800 | if max_candidates == 0 and candidate_page != 1: |
| 1801 | raise ValueError( |
| 1802 | "candidate_page must be 1 when max_candidates is 0" |
| 1803 | ) |
| 1804 | _warn_weak_required_terms(required_terms) |
| 1805 | request = ImageSearchRequest( |
| 1806 | query=item["query"], |
| 1807 | purpose=item.get("purpose", ""), |
| 1808 | orientation="" if orientation == "any" else orientation, |
| 1809 | filename=filename, |
| 1810 | slide=item.get("slide", ""), |
| 1811 | min_width=int(item.get("min_width", default_min_width)), |
| 1812 | min_height=int(item.get("min_height", default_min_height)), |
| 1813 | required_terms=required_terms, |
| 1814 | query_variants=_parse_query_variants(item.get("query_variants")), |
| 1815 | ) |
| 1816 | |
| 1817 | pinned = item.get("provider") or default_provider |
| 1818 | providers = [pinned] if pinned else _default_provider_chain() |
| 1819 | output_path = output_dir / filename |
| 1820 | |
| 1821 | result = search_and_download( |
| 1822 | providers, |
| 1823 | request, |
| 1824 | output_path=output_path, |
| 1825 | strict_no_attribution=strict, |
| 1826 | save_candidates=save_candidates, |
| 1827 | max_candidates=max_candidates, |
| 1828 | candidate_page=candidate_page, |
| 1829 | provider_is_explicit=bool(pinned), |
| 1830 | ) |
| 1831 | if result.selection_required: |
| 1832 | sheet = result.review_sheet or ( |
| 1833 | output_dir / "candidates" / Path(filename).stem / "review_sheet.jpg" |
| 1834 | ) |
| 1835 | return ( |
| 1836 | { |
| 1837 | "candidate_page": result.candidate_page, |
| 1838 | "candidate_count": result.candidate_count, |
| 1839 | "candidate_total": result.candidate_total, |
| 1840 | "has_more_candidates": result.has_more_candidates, |
| 1841 | "next_candidate_page": ( |
| 1842 | result.candidate_page + 1 |
| 1843 | if result.has_more_candidates |
| 1844 | else None |
| 1845 | ), |
| 1846 | "review_sheet": ( |
| 1847 | f"candidates/{Path(filename).stem}/{sheet.name}" |
| 1848 | ), |
| 1849 | }, |
| 1850 | ( |
| 1851 | f"{result.candidate_count}/{result.candidate_total} thumbnail " |
| 1852 | f"candidate(s), page {result.candidate_page}: {sheet}" |
| 1853 | + ( |
| 1854 | f"; next page: {result.candidate_page + 1}" |
| 1855 | if result.has_more_candidates |
| 1856 | else "; pool exhausted" |
| 1857 | ) |
| 1858 | ), |
| 1859 | False, |
| 1860 | True, |
| 1861 | None, |
| 1862 | None, |
| 1863 | ) |
| 1864 | if result.candidate is None: |
| 1865 | return ( |
| 1866 | None, |
| 1867 | result.error or "search failed", |
| 1868 | result.failure_kind == SEARCH_FAILURE_RETRYABLE, |
| 1869 | False, |
| 1870 | None, |
| 1871 | None, |
| 1872 | ) |
| 1873 | if result.staged_path is None or result.output_path is None: |
| 1874 | return ( |
| 1875 | None, |
| 1876 | "search succeeded without a staged output", |
| 1877 | True, |
| 1878 | False, |
| 1879 | None, |
| 1880 | None, |
| 1881 | ) |
| 1882 | |
| 1883 | try: |
| 1884 | item_args = argparse.Namespace( |
| 1885 | filename=filename, |
| 1886 | slide=item.get("slide", ""), |
| 1887 | purpose=item.get("purpose", ""), |
| 1888 | query=item["query"], |
| 1889 | orientation=orientation, |
| 1890 | required_terms=request.required_terms, |
| 1891 | query_variants=request.query_variants, |
| 1892 | ) |
| 1893 | manifest_item = _candidate_to_manifest_item( |
| 1894 | result.candidate, |
| 1895 | item_args, |
| 1896 | provider_name=result.provider_name or "", |
| 1897 | stage=result.stage or "", |
| 1898 | actual_dimensions=result.actual_dimensions, |
| 1899 | ) |
| 1900 | return ( |
| 1901 | manifest_item, |
| 1902 | None, |
| 1903 | False, |
| 1904 | False, |
| 1905 | result.staged_path, |
| 1906 | result.output_path, |
| 1907 | ) |
| 1908 | except Exception: |
| 1909 | if result.staged_path is not None: |
| 1910 | try: |
| 1911 | result.staged_path.unlink(missing_ok=True) |
| 1912 | except OSError: |
| 1913 | pass |
| 1914 | raise |
| 1915 | |
| 1916 | |
| 1917 | def run_search_manifest( |
| 1918 | manifest: dict, |
| 1919 | manifest_path: str, |
| 1920 | *, |
| 1921 | output_dir: Path, |
| 1922 | sources_manifest_path: Path, |
| 1923 | concurrency: int, |
| 1924 | save_candidates: bool, |
| 1925 | max_candidates: int, |
| 1926 | candidate_page: int, |
| 1927 | default_provider: Optional[str], |
| 1928 | default_strict: bool, |
| 1929 | default_min_width: int, |
| 1930 | default_min_height: int, |
| 1931 | ) -> tuple[int, int, int, int, int]: |
| 1932 | """Process all Pending/Failed rows concurrently with a bounded pool. |
| 1933 | |
| 1934 | On success the rich provenance entry is appended to ``image_sources.json`` |
| 1935 | (the credit source of truth) and the row's status flips to ``Sourced``. |
| 1936 | Thumbnail-only rows become ``Needs-Selection`` until promoted or re-queried. |
| 1937 | A row that exhausts the provider/stage chain becomes ``Needs-Manual``. |
| 1938 | Status is written back after each completion, so an interrupt preserves |
| 1939 | finished rows. Returns |
| 1940 | ``(sourced, needs_selection, needs_manual, failed, skipped)``. |
| 1941 | """ |
| 1942 | sources_manifest = _read_existing_manifest(sources_manifest_path) |
| 1943 | items = manifest["items"] |
| 1944 | |
| 1945 | provenance_filenames = { |
| 1946 | item["filename"].casefold(): item["filename"] |
| 1947 | for item in sources_manifest.get("items", []) |
| 1948 | if isinstance(item.get("filename"), str) |
| 1949 | } |
| 1950 | repaired_sourced = False |
| 1951 | for item in items: |
| 1952 | if item["status"] != SEARCH_STATUS_SOURCED: |
| 1953 | continue |
| 1954 | filename = item["filename"] |
| 1955 | target_path = output_dir / filename |
| 1956 | reasons: list[str] = [] |
| 1957 | provenance_filename = provenance_filenames.get(filename.casefold()) |
| 1958 | if provenance_filename is None: |
| 1959 | reasons.append("image_sources.json has no matching provenance entry") |
| 1960 | elif provenance_filename != filename: |
| 1961 | reasons.append( |
| 1962 | "image_sources.json filename casing does not match " |
| 1963 | f"({provenance_filename!r} vs {filename!r})" |
| 1964 | ) |
| 1965 | if not target_path.is_file(): |
| 1966 | reasons.append("target file is missing") |
| 1967 | else: |
| 1968 | min_width = int(item.get("min_width", default_min_width)) |
| 1969 | min_height = int(item.get("min_height", default_min_height)) |
| 1970 | try: |
| 1971 | target_is_valid = _validate_downloaded_quality( |
| 1972 | target_path, |
| 1973 | min_width=min_width, |
| 1974 | min_height=min_height, |
| 1975 | ) |
| 1976 | except RuntimeError as exc: |
| 1977 | reasons.append(f"target validation unavailable: {exc}") |
| 1978 | else: |
| 1979 | if not target_is_valid: |
| 1980 | reasons.append( |
| 1981 | "target file is unreadable or below requested dimensions" |
| 1982 | ) |
| 1983 | if reasons: |
| 1984 | item["status"] = SEARCH_STATUS_FAILED |
| 1985 | item["last_error"] = ( |
| 1986 | "Sourced state validation failed: " + "; ".join(reasons) |
| 1987 | )[:500] |
| 1988 | repaired_sourced = True |
| 1989 | print(f" [RETRY] {filename} — {item['last_error']}") |
| 1990 | if repaired_sourced: |
| 1991 | save_search_manifest(manifest_path, manifest) |
| 1992 | |
| 1993 | pending_idx = [ |
| 1994 | i for i, it in enumerate(items) |
| 1995 | if it["status"] in SEARCH_RETRYABLE_STATUSES |
| 1996 | ] |
| 1997 | total = len(pending_idx) |
| 1998 | skipped = len(items) - total |
| 1999 | |
| 2000 | if total == 0: |
| 2001 | print( |
| 2002 | f"[Batch] Nothing to do — all {len(items)} row(s) already in a " |
| 2003 | "terminal state (Sourced / Needs-Selection / Needs-Manual)." |
| 2004 | ) |
| 2005 | return 0, 0, 0, 0, skipped |
| 2006 | |
| 2007 | print( |
| 2008 | f"\n[Batch] {total} row(s) to search, {skipped} already done. " |
| 2009 | f"concurrency={concurrency}\n" |
| 2010 | ) |
| 2011 | |
| 2012 | sourced_count = 0 |
| 2013 | needs_selection_count = 0 |
| 2014 | needs_manual_count = 0 |
| 2015 | failed_count = 0 |
| 2016 | write_lock = threading.Lock() |
| 2017 | |
| 2018 | def _one(idx: int): |
| 2019 | try: |
| 2020 | manifest_item, error, retryable, selection_required, staged_path, target_path = ( |
| 2021 | _search_one_item( |
| 2022 | items[idx], |
| 2023 | output_dir=output_dir, |
| 2024 | save_candidates=save_candidates, |
| 2025 | max_candidates=max_candidates, |
| 2026 | default_candidate_page=candidate_page, |
| 2027 | default_provider=default_provider, |
| 2028 | default_strict=default_strict, |
| 2029 | default_min_width=default_min_width, |
| 2030 | default_min_height=default_min_height, |
| 2031 | ) |
| 2032 | ) |
| 2033 | return ( |
| 2034 | idx, |
| 2035 | manifest_item, |
| 2036 | error, |
| 2037 | retryable, |
| 2038 | selection_required, |
| 2039 | staged_path, |
| 2040 | target_path, |
| 2041 | ) |
| 2042 | except Exception as exc: # noqa: BLE001 — provider code raises freely |
| 2043 | return idx, None, str(exc)[:500], True, False, None, None |
| 2044 | |
| 2045 | futures: list[concurrent.futures.Future] = [] |
| 2046 | try: |
| 2047 | with concurrent.futures.ThreadPoolExecutor(max_workers=concurrency) as ex: |
| 2048 | futures = [ex.submit(_one, i) for i in pending_idx] |
| 2049 | for fut in concurrent.futures.as_completed(futures): |
| 2050 | ( |
| 2051 | idx, |
| 2052 | manifest_item, |
| 2053 | error, |
| 2054 | retryable, |
| 2055 | selection_required, |
| 2056 | staged_path, |
| 2057 | target_path, |
| 2058 | ) = fut.result() |
| 2059 | item = items[idx] |
| 2060 | with write_lock: |
| 2061 | if ( |
| 2062 | manifest_item is not None |
| 2063 | and staged_path is not None |
| 2064 | and target_path is not None |
| 2065 | ): |
| 2066 | try: |
| 2067 | _commit_staged_image( |
| 2068 | staged_path, |
| 2069 | target_path, |
| 2070 | lambda: write_sources_manifest( |
| 2071 | sources_manifest_path, |
| 2072 | manifest_item, |
| 2073 | ), |
| 2074 | ) |
| 2075 | except Exception as exc: |
| 2076 | if not _is_recoverable_image_error(exc): |
| 2077 | raise |
| 2078 | item["status"] = SEARCH_STATUS_FAILED |
| 2079 | item["last_error"] = ( |
| 2080 | f"canonical/provenance commit failed: {exc}" |
| 2081 | )[:500] |
| 2082 | _clear_candidate_selection_outputs(item) |
| 2083 | failed_count += 1 |
| 2084 | print( |
| 2085 | f" [FAIL] {item['filename']} — " |
| 2086 | f"{item['last_error']}" |
| 2087 | ) |
| 2088 | else: |
| 2089 | item["status"] = SEARCH_STATUS_SOURCED |
| 2090 | item["provider"] = manifest_item.get("provider", "") |
| 2091 | item["license_tier"] = manifest_item.get( |
| 2092 | "license_tier", |
| 2093 | "", |
| 2094 | ) |
| 2095 | item.pop("last_error", None) |
| 2096 | _clear_candidate_selection_outputs(item) |
| 2097 | sourced_count += 1 |
| 2098 | review = _write_review_copy( |
| 2099 | target_path, |
| 2100 | output_dir / ".review", |
| 2101 | target_path.name, |
| 2102 | ) |
| 2103 | if review is not None: |
| 2104 | print( |
| 2105 | f" review copy: {review}", |
| 2106 | file=sys.stderr, |
| 2107 | ) |
| 2108 | print( |
| 2109 | f" [OK] {item['filename']} " |
| 2110 | f"({item['provider']})" |
| 2111 | ) |
| 2112 | elif selection_required: |
| 2113 | item["status"] = SEARCH_STATUS_NEEDS_SELECTION |
| 2114 | _clear_candidate_selection_outputs(item) |
| 2115 | if manifest_item is not None: |
| 2116 | item.update(manifest_item) |
| 2117 | item.pop("last_error", None) |
| 2118 | item.pop("provider", None) |
| 2119 | item.pop("license_tier", None) |
| 2120 | needs_selection_count += 1 |
| 2121 | print(f" [REVIEW] {item['filename']} — {error}") |
| 2122 | elif retryable: |
| 2123 | item["status"] = SEARCH_STATUS_FAILED |
| 2124 | item["last_error"] = error or "provider/download failure" |
| 2125 | _clear_candidate_selection_outputs(item) |
| 2126 | failed_count += 1 |
| 2127 | print( |
| 2128 | f" [FAIL] {item['filename']} — {item['last_error']}" |
| 2129 | ) |
| 2130 | else: |
| 2131 | item["status"] = SEARCH_STATUS_NEEDS_MANUAL |
| 2132 | item["last_error"] = error or "search failed" |
| 2133 | _clear_candidate_selection_outputs(item) |
| 2134 | needs_manual_count += 1 |
| 2135 | print( |
| 2136 | f" [MANUAL] {item['filename']} — " |
| 2137 | f"{item['last_error']}" |
| 2138 | ) |
| 2139 | save_search_manifest(manifest_path, manifest) |
| 2140 | finally: |
| 2141 | # Workers only stage files. Any result not committed because of an |
| 2142 | # interrupt or a later manifest error must not leave candidate residue. |
| 2143 | for future in futures: |
| 2144 | if not future.done(): |
| 2145 | continue |
| 2146 | try: |
| 2147 | outcome = future.result() |
| 2148 | except BaseException: |
| 2149 | continue |
| 2150 | staged_path = outcome[5] |
| 2151 | if staged_path is not None: |
| 2152 | try: |
| 2153 | staged_path.unlink(missing_ok=True) |
| 2154 | except OSError: |
| 2155 | pass |
| 2156 | |
| 2157 | print( |
| 2158 | f"\n[Batch] Done: {sourced_count} sourced / " |
| 2159 | f"{needs_selection_count} needs-selection / {failed_count} failed / " |
| 2160 | f"{needs_manual_count} needs-manual ({skipped} pre-skipped). " |
| 2161 | f"Manifest: {manifest_path}" |
| 2162 | ) |
| 2163 | return ( |
| 2164 | sourced_count, |
| 2165 | needs_selection_count, |
| 2166 | needs_manual_count, |
| 2167 | failed_count, |
| 2168 | skipped, |
| 2169 | ) |
| 2170 | |
| 2171 | |
| 2172 | # --------------------------------------------------------------------------- |
| 2173 | # CLI |
| 2174 | # --------------------------------------------------------------------------- |
| 2175 | |
| 2176 | |
| 2177 | def build_parser() -> argparse.ArgumentParser: |
| 2178 | parser = argparse.ArgumentParser( |
| 2179 | description=( |
| 2180 | "Search openly-licensed web images, shortlist thumbnails, or " |
| 2181 | "download one selected original. Sister to image_gen.py." |
| 2182 | ), |
| 2183 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 2184 | ) |
| 2185 | parser.add_argument( |
| 2186 | "query", |
| 2187 | nargs="?", |
| 2188 | default=None, |
| 2189 | help="Search query (1-4 concrete keywords work best). Omit in --batch mode.", |
| 2190 | ) |
| 2191 | parser.add_argument( |
| 2192 | "--query-variant", |
| 2193 | action="append", |
| 2194 | default=None, |
| 2195 | metavar="QUERY", |
| 2196 | help=( |
| 2197 | "Add a materially different spelling, translation, or entity " |
| 2198 | "query. Repeatable; results are aggregated and deduplicated." |
| 2199 | ), |
| 2200 | ) |
| 2201 | parser.add_argument( |
| 2202 | "--filename", |
| 2203 | default=None, |
| 2204 | help=( |
| 2205 | "Local filename for the chosen image (e.g. cover_bg.jpg). " |
| 2206 | "Required for single-query, --promote, and --from-url modes; " |
| 2207 | "ignored only during batch search." |
| 2208 | ), |
| 2209 | ) |
| 2210 | parser.add_argument( |
| 2211 | "-o", |
| 2212 | "--output", |
| 2213 | default=".", |
| 2214 | help="Output directory. Manifest defaults to <output>/image_sources.json.", |
| 2215 | ) |
| 2216 | parser.add_argument( |
| 2217 | "--provider", |
| 2218 | choices=ALL_PROVIDERS, |
| 2219 | default=None, |
| 2220 | help=( |
| 2221 | "Pin one provider. Default: try zero-config providers (openverse, " |
| 2222 | "wikimedia) plus any keyed provider whose API key is set." |
| 2223 | ), |
| 2224 | ) |
| 2225 | parser.add_argument( |
| 2226 | "--orientation", |
| 2227 | choices=ORIENTATION_CHOICES, |
| 2228 | default="any", |
| 2229 | help="Preferred orientation.", |
| 2230 | ) |
| 2231 | parser.add_argument( |
| 2232 | "--purpose", |
| 2233 | default="", |
| 2234 | help="Purpose tag stored in the manifest (e.g. background, hero, side).", |
| 2235 | ) |
| 2236 | parser.add_argument( |
| 2237 | "--slide", |
| 2238 | default="", |
| 2239 | help="Slide identifier the image belongs to (e.g. 01_cover).", |
| 2240 | ) |
| 2241 | parser.add_argument( |
| 2242 | "--strict-no-attribution", |
| 2243 | action="store_true", |
| 2244 | help=( |
| 2245 | "Refuse CC BY / CC BY-SA results. If no attribution-free match is " |
| 2246 | "downloadable, exit non-zero." |
| 2247 | ), |
| 2248 | ) |
| 2249 | parser.add_argument( |
| 2250 | "--min-width", |
| 2251 | type=int, |
| 2252 | default=1200, |
| 2253 | help="Minimum acceptable image width in pixels (default: 1200).", |
| 2254 | ) |
| 2255 | parser.add_argument( |
| 2256 | "--min-height", |
| 2257 | type=int, |
| 2258 | default=800, |
| 2259 | help="Minimum acceptable image height in pixels (default: 800).", |
| 2260 | ) |
| 2261 | parser.add_argument( |
| 2262 | "--require-terms", |
| 2263 | action="append", |
| 2264 | default=None, |
| 2265 | metavar="TERM[,TERM...]", |
| 2266 | help=( |
| 2267 | "Entity-safety gate: require each metadata term group before a " |
| 2268 | "candidate can be accepted. Repeatable; comma separates groups; " |
| 2269 | "'A|B' means aliases within one group. Example: " |
| 2270 | "--require-terms Chongqing --require-terms 'Jiefangbei|Liberation Monument'." |
| 2271 | ), |
| 2272 | ) |
| 2273 | parser.add_argument( |
| 2274 | "--manifest", |
| 2275 | default=None, |
| 2276 | help="Override manifest path. Defaults to <output>/image_sources.json.", |
| 2277 | ) |
| 2278 | parser.add_argument( |
| 2279 | "--batch", |
| 2280 | default=None, |
| 2281 | metavar="QUERIES_JSON", |
| 2282 | help=( |
| 2283 | "Process a batch of search requests from an image_queries.json " |
| 2284 | "manifest concurrently, writing provenance into image_sources.json " |
| 2285 | "and status back into the queries manifest. In --promote mode, " |
| 2286 | "reconcile the selected row to Sourced." |
| 2287 | ), |
| 2288 | ) |
| 2289 | parser.add_argument( |
| 2290 | "--concurrency", |
| 2291 | type=int, |
| 2292 | default=None, |
| 2293 | help=( |
| 2294 | "Max concurrent searches in --batch mode. Defaults to " |
| 2295 | f"IMAGE_SEARCH_CONCURRENCY env or {DEFAULT_SEARCH_CONCURRENCY}. " |
| 2296 | "Keep modest — free providers are rate-sensitive; use 1 for " |
| 2297 | "strict one-at-a-time pacing." |
| 2298 | ), |
| 2299 | ) |
| 2300 | parser.add_argument( |
| 2301 | "--save-candidates", |
| 2302 | action="store_true", |
| 2303 | help=( |
| 2304 | "Thumbnail-selection mode: save one ranked page of review-eligible " |
| 2305 | "previews and a labeled contact sheet, but download no original. " |
| 2306 | "Use --promote after visual selection." |
| 2307 | ), |
| 2308 | ) |
| 2309 | parser.add_argument( |
| 2310 | "--max-candidates", |
| 2311 | type=int, |
| 2312 | default=DEFAULT_CANDIDATE_PAGE_SIZE, |
| 2313 | help=( |
| 2314 | "Thumbnail page size with --save-candidates (default: " |
| 2315 | f"{DEFAULT_CANDIDATE_PAGE_SIZE}); 0 explicitly keeps every " |
| 2316 | "qualified candidate." |
| 2317 | ), |
| 2318 | ) |
| 2319 | parser.add_argument( |
| 2320 | "--candidate-page", |
| 2321 | type=int, |
| 2322 | default=1, |
| 2323 | help=( |
| 2324 | "Ranked thumbnail page to fetch with --save-candidates " |
| 2325 | "(default: 1). Batch items may override this with candidate_page." |
| 2326 | ), |
| 2327 | ) |
| 2328 | parser.add_argument( |
| 2329 | "--promote", |
| 2330 | default=None, |
| 2331 | metavar="CANDIDATE_FILE", |
| 2332 | help=( |
| 2333 | "Download one selected candidate original and make it primary. " |
| 2334 | "Example: --promote candidate_03.jpg --filename 05_wulong.jpg -o images/" |
| 2335 | ), |
| 2336 | ) |
| 2337 | parser.add_argument( |
| 2338 | "--from-url", |
| 2339 | default=None, |
| 2340 | metavar="URL", |
| 2341 | help=( |
| 2342 | "Manual replacement: download a directly selected image URL into " |
| 2343 | "--filename and record it (license marked 'manual'). Works without " |
| 2344 | "a multimodal model. Example: --from-url https://… --filename team.jpg -o images/" |
| 2345 | ), |
| 2346 | ) |
| 2347 | return parser |
| 2348 | |
| 2349 | |
| 2350 | def _default_provider_chain() -> list[str]: |
| 2351 | """Keyed high-quality providers first; zero-config providers as fallback. |
| 2352 | This is the search order when ``--provider`` is unset.""" |
| 2353 | chain: list[str] = [] |
| 2354 | if os.environ.get("PEXELS_API_KEY"): |
| 2355 | chain.append("pexels") |
| 2356 | if os.environ.get("PIXABAY_API_KEY"): |
| 2357 | chain.append("pixabay") |
| 2358 | chain.extend(ZERO_CONFIG_PROVIDERS) |
| 2359 | return chain |
| 2360 | |
| 2361 | |
| 2362 | def main(argv: Optional[list[str]] = None) -> int: |
| 2363 | _load_search_env_file() |
| 2364 | |
| 2365 | parser = build_parser() |
| 2366 | args = parser.parse_args(argv) |
| 2367 | |
| 2368 | try: |
| 2369 | if args.filename: |
| 2370 | args.filename = _validate_bare_filename(args.filename) |
| 2371 | if args.promote: |
| 2372 | args.promote = _validate_bare_filename( |
| 2373 | args.promote, field_name="--promote candidate filename" |
| 2374 | ) |
| 2375 | except ValueError as exc: |
| 2376 | parser.error(str(exc)) |
| 2377 | if args.min_width < 1 or args.min_height < 1: |
| 2378 | parser.error("--min-width and --min-height must both be positive integers") |
| 2379 | if args.max_candidates < 0: |
| 2380 | parser.error("--max-candidates must be zero or a positive integer") |
| 2381 | if args.candidate_page < 1: |
| 2382 | parser.error("--candidate-page must be a positive integer") |
| 2383 | if args.max_candidates == 0 and args.candidate_page != 1: |
| 2384 | parser.error("--candidate-page must be 1 when --max-candidates is 0") |
| 2385 | |
| 2386 | output_dir = Path(args.output) |
| 2387 | |
| 2388 | # --- Promote mode --- |
| 2389 | if args.promote: |
| 2390 | if not args.filename: |
| 2391 | parser.error("--filename is required in --promote mode") |
| 2392 | return promote_candidate( |
| 2393 | output_dir, |
| 2394 | args.filename, |
| 2395 | args.promote, |
| 2396 | manifest_path=Path(args.manifest) if args.manifest else None, |
| 2397 | queries_manifest_path=Path(args.batch) if args.batch else None, |
| 2398 | ) |
| 2399 | |
| 2400 | # --- Manual URL replacement --- |
| 2401 | if args.from_url: |
| 2402 | if not args.filename: |
| 2403 | parser.error("--filename is required with --from-url") |
| 2404 | return fetch_url_replace( |
| 2405 | output_dir, |
| 2406 | args.filename, |
| 2407 | args.from_url, |
| 2408 | manifest_path=Path(args.manifest) if args.manifest else None, |
| 2409 | slide=args.slide, |
| 2410 | purpose=args.purpose, |
| 2411 | search_query=args.query or "", |
| 2412 | orientation="" if args.orientation == "any" else args.orientation, |
| 2413 | required_terms=_parse_required_terms(args.require_terms), |
| 2414 | min_width=args.min_width, |
| 2415 | min_height=args.min_height, |
| 2416 | ) |
| 2417 | |
| 2418 | # --- Batch mode --- |
| 2419 | if args.batch: |
| 2420 | if not os.path.isfile(args.batch): |
| 2421 | print(f"Error: queries manifest not found: {args.batch}", file=sys.stderr) |
| 2422 | return 1 |
| 2423 | try: |
| 2424 | manifest = load_search_manifest(args.batch) |
| 2425 | except ValueError as exc: |
| 2426 | print(f"Error: {exc}", file=sys.stderr) |
| 2427 | return 1 |
| 2428 | batch_output_dir = ( |
| 2429 | output_dir if args.output != "." else Path(args.batch).parent |
| 2430 | ) |
| 2431 | batch_output_dir.mkdir(parents=True, exist_ok=True) |
| 2432 | sources_manifest_path = ( |
| 2433 | Path(args.manifest) if args.manifest |
| 2434 | else default_manifest_path(str(batch_output_dir)) |
| 2435 | ) |
| 2436 | try: |
| 2437 | _, needs_selection, needs_manual, failed, _ = run_search_manifest( |
| 2438 | manifest, |
| 2439 | args.batch, |
| 2440 | output_dir=batch_output_dir, |
| 2441 | sources_manifest_path=sources_manifest_path, |
| 2442 | concurrency=_resolve_search_concurrency(args.concurrency), |
| 2443 | save_candidates=args.save_candidates, |
| 2444 | max_candidates=args.max_candidates, |
| 2445 | candidate_page=args.candidate_page, |
| 2446 | default_provider=args.provider, |
| 2447 | default_strict=args.strict_no_attribution, |
| 2448 | default_min_width=args.min_width, |
| 2449 | default_min_height=args.min_height, |
| 2450 | ) |
| 2451 | except KeyboardInterrupt: |
| 2452 | print("\n\nInterrupted by user. Partial progress preserved in manifest.") |
| 2453 | return 130 |
| 2454 | except RuntimeError as exc: |
| 2455 | print(f"Error: {exc}", file=sys.stderr) |
| 2456 | return 1 |
| 2457 | if needs_selection: |
| 2458 | print( |
| 2459 | f"[Batch] {needs_selection} row(s) await thumbnail selection." |
| 2460 | ) |
| 2461 | # A successfully prepared shortlist is an intermediate success. Only |
| 2462 | # provider failures or exhausted/manual rows make acquisition fail. |
| 2463 | return 1 if needs_manual or failed else 0 |
| 2464 | |
| 2465 | # --- Single-query search mode --- |
| 2466 | if not args.query: |
| 2467 | parser.error("query is required unless --batch, --promote, or --from-url is used") |
| 2468 | if not args.filename: |
| 2469 | parser.error("--filename is required in single-query mode") |
| 2470 | |
| 2471 | request = ImageSearchRequest( |
| 2472 | query=args.query, |
| 2473 | purpose=args.purpose, |
| 2474 | orientation="" if args.orientation == "any" else args.orientation, |
| 2475 | filename=args.filename, |
| 2476 | slide=args.slide, |
| 2477 | min_width=args.min_width, |
| 2478 | min_height=args.min_height, |
| 2479 | required_terms=_parse_required_terms(args.require_terms), |
| 2480 | query_variants=_parse_query_variants(args.query_variant), |
| 2481 | ) |
| 2482 | _warn_weak_required_terms(request.required_terms) |
| 2483 | |
| 2484 | providers = [args.provider] if args.provider else _default_provider_chain() |
| 2485 | |
| 2486 | manifest_path = ( |
| 2487 | Path(args.manifest) if args.manifest else default_manifest_path(args.output) |
| 2488 | ) |
| 2489 | try: |
| 2490 | _read_existing_manifest(manifest_path) |
| 2491 | except RuntimeError as exc: |
| 2492 | print(f"Error: {exc}", file=sys.stderr) |
| 2493 | return 1 |
| 2494 | |
| 2495 | output_dir.mkdir(parents=True, exist_ok=True) |
| 2496 | output_path = output_dir / args.filename |
| 2497 | |
| 2498 | print(f"Searching providers: {', '.join(providers)}", file=sys.stderr) |
| 2499 | result = search_and_download( |
| 2500 | providers, |
| 2501 | request, |
| 2502 | output_path=output_path, |
| 2503 | strict_no_attribution=args.strict_no_attribution, |
| 2504 | save_candidates=args.save_candidates, |
| 2505 | max_candidates=args.max_candidates, |
| 2506 | candidate_page=args.candidate_page, |
| 2507 | provider_is_explicit=bool(args.provider), |
| 2508 | ) |
| 2509 | |
| 2510 | if result.selection_required: |
| 2511 | continuation = ( |
| 2512 | f" Next page: --candidate-page {result.candidate_page + 1}." |
| 2513 | if result.has_more_candidates |
| 2514 | else " Candidate pool exhausted." |
| 2515 | ) |
| 2516 | print( |
| 2517 | f" [REVIEW] {result.candidate_count}/{result.candidate_total} " |
| 2518 | f"thumbnail candidate(s), page {result.candidate_page}: " |
| 2519 | f"{result.review_sheet}.{continuation}", |
| 2520 | file=sys.stderr, |
| 2521 | ) |
| 2522 | print( |
| 2523 | " No original image or provenance record was written. " |
| 2524 | "Select one with --promote, or change the query and shortlist again.", |
| 2525 | file=sys.stderr, |
| 2526 | ) |
| 2527 | return 0 |
| 2528 | |
| 2529 | if result.candidate is None: |
| 2530 | print( |
| 2531 | f"{result.error or 'Image search failed'}. " |
| 2532 | "Try a shorter query, use default attribution mode if strict mode " |
| 2533 | "is enabled, or set an API key for a keyed provider.", |
| 2534 | file=sys.stderr, |
| 2535 | ) |
| 2536 | return 1 |
| 2537 | |
| 2538 | print( |
| 2539 | f" picked: {result.candidate.title!r} from {result.provider_name} " |
| 2540 | f"({result.candidate.license_name or 'no license string'}, " |
| 2541 | f"{result.candidate.license_tier})", |
| 2542 | file=sys.stderr, |
| 2543 | ) |
| 2544 | |
| 2545 | # The staged file has already been measured; upstream metadata can still be |
| 2546 | # off (e.g. Openverse aggregates rawpixel which only exposes previews). |
| 2547 | actual_dimensions = result.actual_dimensions |
| 2548 | if ( |
| 2549 | actual_dimensions is not None |
| 2550 | and result.candidate.width |
| 2551 | and result.candidate.height |
| 2552 | and actual_dimensions[0] * actual_dimensions[1] |
| 2553 | < 0.5 * result.candidate.width * result.candidate.height |
| 2554 | ): |
| 2555 | print( |
| 2556 | f"\n[!] Downloaded image is much smaller than upstream metadata " |
| 2557 | f"({actual_dimensions[0]}x{actual_dimensions[1]} vs " |
| 2558 | f"{result.candidate.width}x{result.candidate.height}). The provider " |
| 2559 | f"likely only exposes a preview here. Layout based on the manifest's " |
| 2560 | f"width/height will be accurate; the metadata_dimensions field " |
| 2561 | f"is preserved for reference.", |
| 2562 | file=sys.stderr, |
| 2563 | ) |
| 2564 | |
| 2565 | if result.staged_path is None or result.output_path is None: |
| 2566 | print("Error: image search returned no staged output.", file=sys.stderr) |
| 2567 | if result.staged_path is not None: |
| 2568 | try: |
| 2569 | result.staged_path.unlink(missing_ok=True) |
| 2570 | except OSError: |
| 2571 | pass |
| 2572 | return 1 |
| 2573 | try: |
| 2574 | item = _candidate_to_manifest_item( |
| 2575 | result.candidate, |
| 2576 | args, |
| 2577 | provider_name=result.provider_name or "", |
| 2578 | stage=result.stage or "", |
| 2579 | actual_dimensions=actual_dimensions, |
| 2580 | ) |
| 2581 | written = _commit_staged_image( |
| 2582 | result.staged_path, |
| 2583 | result.output_path, |
| 2584 | lambda: write_sources_manifest(manifest_path, item), |
| 2585 | ) |
| 2586 | except (OSError, RuntimeError, ValueError) as exc: |
| 2587 | print(f"Error: {exc}", file=sys.stderr) |
| 2588 | return 1 |
| 2589 | finally: |
| 2590 | try: |
| 2591 | result.staged_path.unlink(missing_ok=True) |
| 2592 | except OSError: |
| 2593 | pass |
| 2594 | print(f" manifest: {written}", file=sys.stderr) |
| 2595 | review = _write_review_copy( |
| 2596 | result.output_path, |
| 2597 | result.output_path.parent / ".review", |
| 2598 | result.output_path.name, |
| 2599 | ) |
| 2600 | if review is not None: |
| 2601 | print(f" review copy: {review}", file=sys.stderr) |
| 2602 | |
| 2603 | if result.candidate.license_tier == "attribution-required": |
| 2604 | print( |
| 2605 | "\n[!] This image requires on-slide attribution. " |
| 2606 | "Executor should add a small credit element to the slide using " |
| 2607 | "the 'attribution_text' field in the manifest.", |
| 2608 | file=sys.stderr, |
| 2609 | ) |
| 2610 | |
| 2611 | return 0 |
| 2612 | |
| 2613 | |
| 2614 | if __name__ == "__main__": |
| 2615 | raise SystemExit(main()) |
| 2616 |