|
2 | 2 |
|
3 | 3 | import abc |
4 | 4 | import io |
| 5 | +import re |
5 | 6 | import stat |
6 | 7 | from collections.abc import Mapping, Sequence |
7 | 8 | from dataclasses import dataclass, field |
@@ -380,36 +381,134 @@ def _get_manifest_entry_by_path(manifest: Manifest, path: Path) -> BaseEntry | N |
380 | 381 | return None |
381 | 382 |
|
382 | 383 |
|
| 384 | +_BLOCK_SCALAR_HEADER = re.compile(r"[>|](?:[1-9][+-]?|[+-][1-9]?)?(?:[ \t]+#.*)?") |
| 385 | + |
| 386 | + |
| 387 | +def _indent_of(line: str) -> int: |
| 388 | + """Count indentation spaces, leaving tabs in scalar content.""" |
| 389 | + |
| 390 | + return len(line) - len(line.lstrip(" ")) |
| 391 | + |
| 392 | + |
| 393 | +def _fold_lines(lines: list[str]) -> str: |
| 394 | + """Fold ordinary lines, retaining paragraphs and more-indented block content.""" |
| 395 | + |
| 396 | + parts: list[str] = [] |
| 397 | + previous = "" |
| 398 | + blank_lines = 0 |
| 399 | + for line in lines: |
| 400 | + if not line: |
| 401 | + blank_lines += 1 |
| 402 | + continue |
| 403 | + if parts: |
| 404 | + more_indented = previous.startswith((" ", "\t")) or line.startswith((" ", "\t")) |
| 405 | + parts.append("\n" * (blank_lines + 1) if more_indented else "\n" * blank_lines or " ") |
| 406 | + parts.append(line) |
| 407 | + previous = line |
| 408 | + blank_lines = 0 |
| 409 | + return "".join(parts) |
| 410 | + |
| 411 | + |
| 412 | +def _take_continuation_lines( |
| 413 | + lines: list[str], start: int, end: int, key_indent: int, *, block_header: str | None |
| 414 | +) -> tuple[list[str], int]: |
| 415 | + """Consume a field's indented body and strip a block scalar's content indentation.""" |
| 416 | + |
| 417 | + content_indent: int | None = None |
| 418 | + if block_header is not None: |
| 419 | + content_indent = next( |
| 420 | + (key_indent + int(char) for char in block_header[1:] if char.isdigit()), None |
| 421 | + ) |
| 422 | + body: list[str] = [] |
| 423 | + index = start |
| 424 | + while index < end: |
| 425 | + line = lines[index] |
| 426 | + stripped = line.strip() |
| 427 | + # Plain-scalar comments may have any indentation and are not scalar content. |
| 428 | + if block_header is None and stripped.startswith("#"): |
| 429 | + index += 1 |
| 430 | + continue |
| 431 | + if stripped: |
| 432 | + indent = _indent_of(line) |
| 433 | + # A dedented comment ends a block just like any other nonblank line. |
| 434 | + if indent < (content_indent if content_indent is not None else key_indent + 1): |
| 435 | + break |
| 436 | + if block_header is not None and content_indent is None: |
| 437 | + content_indent = indent |
| 438 | + body.append(line) |
| 439 | + index += 1 |
| 440 | + while body and not body[-1].strip(): |
| 441 | + body.pop() |
| 442 | + if content_indent is not None: |
| 443 | + body = [line[content_indent:] for line in body] |
| 444 | + return body, index |
| 445 | + |
| 446 | + |
383 | 447 | def _parse_frontmatter(markdown: str) -> dict[str, str]: |
384 | | - """Parse the simple YAML frontmatter shape used by skill indexes.""" |
| 448 | + """Read the string metadata subset used by skill indexes, without a YAML dependency. |
| 449 | +
|
| 450 | + Supports single-line values, indented wrapped plain values, and folded/literal blocks |
| 451 | + with optional indentation/chomping indicators and header comments. Block values have |
| 452 | + outer whitespace removed for the index; chomping indicators are accepted but do not |
| 453 | + change that normalization. Internal paragraph breaks and block indentation survive. |
| 454 | +
|
| 455 | + This is not a general YAML loader: tags, aliases, collections, multiline quoted values, |
| 456 | + and quoted escape decoding are not interpreted. Nested fields are consumed without |
| 457 | + promoting their keys. Single-line quote stripping and inline hash text retain their |
| 458 | + historical behavior. Supply explicit Skill metadata or a LazySkillSource when a full |
| 459 | + YAML loader is needed. |
| 460 | + """ |
385 | 461 |
|
386 | 462 | lines = markdown.splitlines() |
387 | 463 | if not lines or lines[0].strip() != "---": |
388 | 464 | return {} |
389 | 465 |
|
390 | | - end_index: int | None = None |
391 | | - for index, line in enumerate(lines[1:], start=1): |
392 | | - if line.strip() == "---": |
393 | | - end_index = index |
394 | | - break |
| 466 | + # Indented document markers inside scalar content do not close the frontmatter. |
| 467 | + delimiter_indent = _indent_of(lines[0]) |
| 468 | + end_index = next( |
| 469 | + ( |
| 470 | + index |
| 471 | + for index, line in enumerate(lines[1:], start=1) |
| 472 | + if line.strip() == "---" and _indent_of(line) <= delimiter_indent |
| 473 | + ), |
| 474 | + None, |
| 475 | + ) |
395 | 476 | if end_index is None: |
396 | 477 | return {} |
397 | 478 |
|
398 | 479 | metadata: dict[str, str] = {} |
399 | | - for line in lines[1:end_index]: |
| 480 | + index = 1 |
| 481 | + while index < end_index: |
| 482 | + line = lines[index] |
| 483 | + index += 1 |
400 | 484 | stripped = line.strip() |
401 | | - if stripped == "" or stripped.startswith("#") or ":" not in stripped: |
| 485 | + if not stripped or stripped.startswith("#") or ":" not in stripped: |
402 | 486 | continue |
403 | 487 | key, value = stripped.split(":", 1) |
404 | | - parsed_key = key.strip() |
405 | 488 | parsed_value = value.strip() |
406 | | - if ( |
| 489 | + # Header comments cannot supply an indentation or chomping indicator. |
| 490 | + block_header = ( |
| 491 | + parsed_value.split()[0] if _BLOCK_SCALAR_HEADER.fullmatch(parsed_value) else None |
| 492 | + ) |
| 493 | + key_indent = _indent_of(line) |
| 494 | + continuation, index = _take_continuation_lines( |
| 495 | + lines, index, end_index, key_indent, block_header=block_header |
| 496 | + ) |
| 497 | + |
| 498 | + if block_header is not None: |
| 499 | + parsed_value = ( |
| 500 | + "\n".join(continuation) if block_header[0] == "|" else _fold_lines(continuation) |
| 501 | + ).strip() |
| 502 | + elif ( |
407 | 503 | len(parsed_value) >= 2 |
408 | 504 | and parsed_value[0] == parsed_value[-1] |
409 | 505 | and parsed_value[0] in {"'", '"'} |
410 | 506 | ): |
411 | 507 | parsed_value = parsed_value[1:-1] |
412 | | - metadata[parsed_key] = parsed_value |
| 508 | + elif parsed_value and continuation: |
| 509 | + parsed_value = _fold_lines([parsed_value, *(item.strip() for item in continuation)]) |
| 510 | + |
| 511 | + metadata[key.strip()] = parsed_value |
413 | 512 | return metadata |
414 | 513 |
|
415 | 514 |
|
|
0 commit comments