chemenu/tools/chemenu/kb_scan.py

"""Scan kb/ into Page objects and build the wikilink graph."""
from __future__ import annotations

import re
from collections import Counter
from pathlib import Path
from typing import Iterator

from chemenu.frontmatter_io import read_page
from chemenu.markdown_code import strip_code_spans
from chemenu.page import Page

WIKILINK_RE = re.compile(r"\[\[([^\]|#]+)")


# Root-level files under kb/ that are not pages: the generated catalog map, log
# and provenance index, plus the two documents that constrain the tree rather
# than living in it - the stack's contract and this instance's own conventions.
_KB_META_FILES = {"index.md", "log.md", "provenance.md", "CONTRACT.md", "CONVENTIONS.md"}

# The per-collection authoring contract. Unlike the meta files above it is never
# at the kb root - it sits one level down, in every collection - so it has to be
# excluded by name at any depth rather than by parent directory.
_COLLECTION_CONTRACT = "COLLECTION.md"

# The generated per-collection/per-area catalog shard. Excluded by name at any
# depth for the same reason as the contract, and for one more: it lists every
# page in its subtree as a wikilink, so treating it as a page would make every
# page look linked-to and silence the orphan check entirely.
GENERATED_INDEX = "INDEX.md"


def is_page_path(relative: str) -> bool:
    """Whether a `kb/`-relative path names a page rather than routing material.

    Stated over a plain path, not a filesystem entry, so callers that read a
    *past* revision out of git can apply the identical rule - `migrate verify`
    does. Two different answers to "is this a page" would report every
    COLLECTION.md and INDEX.md as a page that has since disappeared.
    """
    parts = relative.split("/")
    if parts[-1] in (_COLLECTION_CONTRACT, GENERATED_INDEX):
        return False
    if len(parts) == 1 and parts[0] in _KB_META_FILES:
        return False
    return parts[-1].endswith(".md")


def iter_kb_pages(kb_dir: Path) -> Iterator[Path]:
    """Yield every page under kb_dir.

    Three kinds of file are skipped: the kb-root meta files (generated catalog,
    log, provenance, and the kb contract), every COLLECTION.md, and every
    generated INDEX.md. None carry page frontmatter. A README.md *inside* a
    collection is an ordinary page - only kb-root files are routing material.
    """
    for path in sorted(kb_dir.rglob("*.md")):
        if is_page_path(path.relative_to(kb_dir).as_posix()):
            yield path


def load_kb_pages(kb_dir: Path) -> dict[str, Page]:
    """Load every markdown page under kb_dir, keyed by title (filename stem).

    If two files share a stem (a naming collision), the later one (by sorted
    path order) wins here; `wikitool lint` explicitly detects and reports such
    collisions so they don't go unnoticed.
    """
    pages: dict[str, Page] = {}
    for path in iter_kb_pages(kb_dir):
        frontmatter, body = read_page(path)
        pages[path.stem] = Page(path=path, frontmatter=frontmatter, body=body)
    return pages


def find_duplicate_title_paths(kb_dir: Path, root: Path) -> list[dict]:
    """Return stem collisions as {"stem": str, "paths": [str, ...]}.

    Paths are repo-root-relative and sorted for stable output.
    """
    by_stem: dict[str, list[str]] = {}
    for path in iter_kb_pages(kb_dir):
        try:
            rel = str(path.relative_to(root))
        except ValueError:
            rel = str(path.relative_to(kb_dir.parent))
        by_stem.setdefault(path.stem, []).append(rel)
    return [
        {"stem": stem, "paths": sorted(paths)}
        for stem, paths in sorted(by_stem.items())
        if len(paths) > 1
    ]


def extract_wikilinks(body: str) -> set[str]:
    """Which pages this body links to, as a set.

    The right shape for `lint` and the link graph, whose question is "does
    this reference resolve" - asked once per distinct target. It is the wrong
    shape for asking whether a rewrite *dropped* a link: use
    `count_wikilinks` for that.

    Code is masked out first (see markdown_code.strip_code_spans): a
    `[[Wikilink]]` shown inside a fence or backticks is an example of the
    notation, and counting it made a page that documents the wiki look like it
    linked to something that need not exist.
    """
    return {m.group(1).strip() for m in WIKILINK_RE.finditer(strip_code_spans(body))}


def count_wikilinks(body: str) -> Counter[str]:
    """How often this body links to each page.

    The counting sibling of `extract_wikilinks`, and the reason it exists: a
    page citing `[[X]]` twice that comes back citing it once has the same link
    *set* and a different link *multiset*. Three of the four defects found in
    the 248-page German translation were exactly that shape, and a set-based
    comparison reported all three as clean.
    """
    return Counter(m.group(1).strip() for m in WIKILINK_RE.finditer(strip_code_spans(body)))


def build_link_graph(pages: dict[str, Page]) -> dict[str, set[str]]:
    """Map each page title to the set of titles it links to."""
    return {title: extract_wikilinks(page.body) for title, page in pages.items()}


def inbound_links(graph: dict[str, set[str]]) -> dict[str, set[str]]:
    """Map each page title to the set of titles that link to it."""
    inbound: dict[str, set[str]] = {title: set() for title in graph}
    for source, targets in graph.items():
        for target in targets:
            if target in inbound:
                inbound[target].add(source)
    return inbound