#!/usr/bin/env python3
# A documented one-shot, not part of the OCaml build (spec's "no hand-written
# JSON parser" rule is about colitur's OWN sexp format -- reading someone
# else's JSON needs a reader, same reasoning as tools/bootstrap_sanctoral.ml's
# own header comment on its INI reader). Task 16: extracts the missalemeum
# oracle fixture (test/fixtures/missalemeum-ef-2026-2027.txt) that
# test/test_oracle.ml compares colitur against -- there is no JSON library in
# this project's frozen deps, so the fixture is shaped here, once, outside
# the test, exactly as test/test_differential.ml's lectio fixture already is
# (see that file's own header comment).
#
# Usage:
#   python3 tools/extract_missalemeum_oracle.py <snapshot-dir> <out-file>
#
# <snapshot-dir> is ~/git/projects/lectio/sources/ already unpacked (`tar xzf
# snapshot.tar.gz` in that directory -- lectio's own scripts/snapshot-
# sources.sh does this), so <snapshot-dir>/missalemeum/en/YYYY-MM-DD.json
# exists for every day.
#
# One line per day, pipe-separated:
#   date|rank|colors|title|tempora|commemorations|displaced|n_masses|commemoration_ids|first|gospel
#
# - rank/colors/title/tempora/commemorations/displaced are info.rank,
#   info.colors (sorted, concatenated, e.g. "pv"), info.title, info.tempora
#   ("-" for JSON null), the ";"-joined titles of info.commemorations and
#   info.displaced ("-" for an empty list), verbatim from the JSON's own
#   English strings -- not translated or slugified, so the comparator (and a
#   human auditor) can match colitur's own celebration names against them.
# - n_masses is len(the day's JSON array). Four days (Christmas, All Souls)
#   carry more than one Mass; entry[0]'s info block is used throughout (rank/
#   colors/tempora/commemorations/displaced are IDENTICAL across every Mass
#   of the same day for all 730 days -- checked below, not assumed; see the
#   task report for why entry[0] loses nothing).
# - commemoration_ids (Task B, branch ef-rg16a) is the ";"-joined info.
#   commemorations[*].id, index-aligned with the commemorations field ("-"
#   for an empty list, same convention) -- e.g. "sancti:01-05:4:r". Added so
#   test_oracle.ml's identity comparison has a second, independent signal
#   beyond the title text (the id encodes the commemorated saint's own rank
#   and colour too, verified across the whole fixture to always carry the
#   SAME calendar date as the day itself -- a commemoration's own natural
#   fixed date, by construction, since only a saint impeded on their own day
#   is ever commemorated there). Never parsed for its date component by this
#   fixture or the comparator (info.rank/colors already give the day's own
#   values); kept opaque and compared as a plain string.
# - first/gospel (Task 9) are the day's own Epistle/Lesson and Gospel
#   citations, extracted from entry[0]["sections"] -- the section with
#   id "Lectio" ("Epistle") and "Evangelium" ("Gospel") respectively. Each
#   section's own body is a [[english, latin]] pair; the ENGLISH text
#   (body[0][0]) embeds the citation as one of possibly several
#   "*...*"-wrapped (markdown-italic) spans -- e.g. "Lesson from the letter
#   of St. Paul... \n*Titus 2:11-15*\nBeloved: ...". It is NOT always the
#   FIRST such span: Holy Saturday's own Gospel section opens with a rubric
#   note, itself "*"-wrapped ("*While singing the Gospel candles are not
#   being hold.*"), before the real citation "*Matt 28:1-7*" -- checked
#   directly against the raw JSON, not assumed. The citation span is
#   identified structurally instead of positionally: the first "*...*" span
#   under 40 characters that contains a chapter:verse-shaped digit pair
#   (`\d+\s*[:,.]\s*\d+`, matching all three separators actually used in
#   this source -- "Titus 2:11-15", "4 Kings, 5:1-15", "John 20. 19-31").
#   Verified exhaustively over the whole fixture (all 1 456 Lectio/
#   Evangelium sections, 728 days x 2) -- CORRECTED, fix round 1
#   (coordinator review): this used to claim
#   "exactly one candidate per section, zero ambiguous". WRONG: 2 of the
#   1 456 sections (Holy Saturday's own "Lectio", both years) yield TWO
#   candidates, ['Col 3:1-4', 'Ps 117, 1'] -- the second is the Tractus
#   verse that follows the real Epistle citation in the same paragraph,
#   also matching the verse-shaped regex. The fixture is still correct on
#   both rows (Col 3:1-4 IS the real citation) because [extract_citation]
#   below returns candidates[0] and the real citation happens to come
#   FIRST in the text -- correctness here rests on ORDERING, not
#   uniqueness, and this header previously overclaimed the latter. TWO
#   days (Good Friday, both years -- the "Missa Praesanctificatorum"
#   liturgy) have no "Lectio"/"Evangelium" section at
#   all (a multi-lesson structure instead, "Lectiones"/"Passio", with no
#   single reading occupying the Epistle/Gospel slot this schema assumes,
#   the SAME shape colitur's own test_lectionary.ml records for its own
#   hand-authored Holy Week data) -- "-" for both fields on those two rows,
#   a WARNING on stderr naming the date, never a silent guess. Extracted
#   VERBATIM (only .strip()ped of surrounding whitespace) -- not normalized,
#   not re-punctuated: the comparator's job, not this extractor's, matching
#   the same "verbatim, not translated or slugified" discipline the fields
#   above already state.
# - Spaces in tempora are turned to "_" (matching the "no field has an
#   internal space" convention test/fixtures/lectio-ef-2005-2050.txt already
#   uses, so this fixture can be read the same simple way -- split on '|',
#   not on whitespace, so title/commemoration/displaced text keeps its own
#   spaces; only tempora, which nothing in this project parses further than
#   pass-through display, is space-collapsed for a cheap column-count check).
import json
import os
import re
import sys

STAR_RE = re.compile(r"\*([^*]+)\*")
VERSE_RE = re.compile(r"\d+\s*[:,.]\s*\d+")


def extract_citation(sections, section_id, date, label):
    for sec in sections:
        if sec["id"] != section_id:
            continue
        text = sec["body"][0][0]
        candidates = [s.strip() for s in STAR_RE.findall(text) if VERSE_RE.search(s) and len(s) < 40]
        if candidates:
            return candidates[0]
        print(f"WARNING: {date}: {section_id!r} section found but no citation-shaped span in it ({label})",
              file=sys.stderr)
        return "-"
    print(f"WARNING: {date}: no {section_id!r} section ({label})", file=sys.stderr)
    return "-"


def main():
    # <expected-days> (added 2026-08-17, the 2038 oracle extension) is the
    # number of day-files this run MUST produce, stated by the caller rather
    # than hardcoded. It used to be a bare `!= 730`, which silently forbade
    # extracting any window other than the original 2026-2027 pair. It stays
    # MANDATORY-in-effect via its default so the existing documented command
    # keeps its own guard unchanged: a partial fetch (a dropped HTTP request,
    # an interrupted unpack) must fail loudly here rather than quietly
    # produce a short fixture that then "passes" a comparison it never ran.
    if len(sys.argv) not in (3, 4):
        print(f"usage: {sys.argv[0]} <snapshot-dir> <out-file> [expected-days]", file=sys.stderr)
        return 2
    snapshot_dir, out_path = sys.argv[1], sys.argv[2]
    expected_days = int(sys.argv[3]) if len(sys.argv) == 4 else 730
    src = os.path.join(snapshot_dir, "missalemeum", "en")
    rows = []
    for fname in sorted(os.listdir(src)):
        if not fname.endswith(".json"):
            continue
        date = fname[:-5]
        with open(os.path.join(src, fname), encoding="utf-8") as f:
            data = json.load(f)
        first = data[0]["info"]
        for other in data[1:]:
            o = other["info"]
            for k in ("rank", "colors", "tempora", "commemorations", "displaced"):
                if first[k] != o[k]:
                    print(f"WARNING: {date} masses disagree on {k}: {first[k]!r} vs {o[k]!r}", file=sys.stderr)
        comm_titles = [c["title"] for c in first["commemorations"]]
        comm_ids = [c["id"] for c in first["commemorations"]]
        disp_titles = [d["title"] for d in first["displaced"]]
        for t in [first["title"]] + comm_titles + comm_ids + disp_titles:
            if "|" in t or ";" in t:
                print(f"ERROR: {date}: field {t!r} contains a delimiter this fixture uses", file=sys.stderr)
                return 1
        colors = "".join(sorted(first["colors"]))
        tempora = (first["tempora"] or "-").replace(" ", "_")
        comms = ";".join(comm_titles) or "-"
        disp = ";".join(disp_titles) or "-"
        ids = ";".join(comm_ids) or "-"
        sections = data[0]["sections"]
        cit_first = extract_citation(sections, "Lectio", date, "Epistle")
        cit_gospel = extract_citation(sections, "Evangelium", date, "Gospel")
        for t in (cit_first, cit_gospel):
            if "|" in t:
                print(f"ERROR: {date}: citation {t!r} contains a delimiter this fixture uses", file=sys.stderr)
                return 1
        rows.append(
            f"{date}|{first['rank']}|{colors}|{first['title']}|{tempora}|{comms}|{disp}|{len(data)}|{ids}"
            f"|{cit_first}|{cit_gospel}"
        )

    if len(rows) != expected_days:
        print(f"ERROR: expected {expected_days} days, got {len(rows)}", file=sys.stderr)
        return 1

    with open(out_path, "w", encoding="utf-8") as f:
        f.write("\n".join(rows) + "\n")
    print(f"wrote {len(rows)} lines to {out_path}", file=sys.stderr)
    return 0


if __name__ == "__main__":
    sys.exit(main())