322 lines
12 KiB
Python
322 lines
12 KiB
Python
from __future__ import annotations
|
|
|
|
from contextlib import contextmanager
|
|
from dataclasses import dataclass
|
|
from datetime import datetime
|
|
import fcntl
|
|
import math
|
|
from pathlib import Path
|
|
import re
|
|
from typing import Any, Iterable
|
|
from zoneinfo import ZoneInfo
|
|
|
|
from page_state import MarkerError, PageMarker, find_marker
|
|
from publication_meta import (
|
|
SLUG_RE,
|
|
mark_blocked,
|
|
mark_published,
|
|
mark_publishing,
|
|
same_meta,
|
|
)
|
|
from sanitize import assert_publishable, sanitize_text
|
|
|
|
RIYADH = ZoneInfo("Asia/Riyadh")
|
|
|
|
|
|
class AmbiguousWrite(RuntimeError):
|
|
"""The remote side effect may have succeeded; reconcile before retrying."""
|
|
|
|
|
|
class PublicVerificationError(RuntimeError):
|
|
pass
|
|
|
|
|
|
class DeterministicFailure(RuntimeError):
|
|
def __init__(self, reason: str, message: str):
|
|
super().__init__(message)
|
|
self.reason = reason
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class Candidate:
|
|
page: dict[str, Any]
|
|
marker: PageMarker
|
|
blocks: list[dict[str, Any]]
|
|
|
|
|
|
@contextmanager
|
|
def exclusive_run_lock(path: Path):
|
|
path.parent.mkdir(parents=True, exist_ok=True)
|
|
with path.open("a+") as handle:
|
|
try:
|
|
fcntl.flock(handle.fileno(), fcntl.LOCK_EX | fcntl.LOCK_NB)
|
|
except BlockingIOError:
|
|
yield False
|
|
return
|
|
try:
|
|
yield True
|
|
finally:
|
|
fcntl.flock(handle.fileno(), fcntl.LOCK_UN)
|
|
|
|
|
|
def _queued_at(candidate: Candidate) -> datetime:
|
|
return datetime.fromisoformat(candidate.marker.meta["queued_at"].replace("Z", "+00:00"))
|
|
|
|
|
|
def _lexical(article: dict[str, Any]) -> dict[str, Any]:
|
|
children: list[dict[str, Any]] = []
|
|
for section in article["sections"]:
|
|
children.append({
|
|
"type": "heading",
|
|
"tag": "h2",
|
|
"version": 1,
|
|
"children": [{"type": "text", "version": 1, "text": section["heading"]}],
|
|
})
|
|
for paragraph in section["paragraphs"]:
|
|
children.append({
|
|
"type": "paragraph",
|
|
"version": 1,
|
|
"children": [{"type": "text", "version": 1, "text": paragraph}],
|
|
})
|
|
if section.get("code"):
|
|
children.append({
|
|
"type": "code",
|
|
"version": 1,
|
|
"language": "text",
|
|
"children": [{"type": "text", "version": 1, "text": section["code"]}],
|
|
})
|
|
return {
|
|
"root": {
|
|
"type": "root",
|
|
"format": "",
|
|
"indent": 0,
|
|
"version": 1,
|
|
"direction": "rtl",
|
|
"children": children,
|
|
}
|
|
}
|
|
|
|
|
|
def _validate_article(article: Any, access: str, expected_slug: str | None = None) -> dict[str, Any]:
|
|
if not isinstance(article, dict):
|
|
raise DeterministicFailure("article_schema", "article must be an object")
|
|
for field in ("title", "slug", "dek", "sections"):
|
|
if field not in article:
|
|
raise DeterministicFailure("article_schema", f"article field missing: {field}")
|
|
for field in ("title", "slug", "dek"):
|
|
if not isinstance(article[field], str) or not article[field].strip():
|
|
raise DeterministicFailure("article_schema", f"article field invalid: {field}")
|
|
if not SLUG_RE.fullmatch(article["slug"]):
|
|
raise DeterministicFailure("article_schema", "article slug must be lowercase ASCII kebab-case")
|
|
if expected_slug is not None and article["slug"] != expected_slug:
|
|
raise DeterministicFailure("article_schema", "republish slug cannot change")
|
|
if access == "vip" and (not isinstance(article.get("teaser"), str) or not article["teaser"].strip()):
|
|
raise DeterministicFailure("article_schema", "VIP teaser is required")
|
|
|
|
sections = article["sections"]
|
|
if not isinstance(sections, list) or not 3 <= len(sections) <= 6:
|
|
raise DeterministicFailure("article_schema", "article must contain 3 to 6 sections")
|
|
for section in sections:
|
|
if not isinstance(section, dict):
|
|
raise DeterministicFailure("article_schema", "section must be an object")
|
|
if not isinstance(section.get("heading"), str) or not section["heading"].strip():
|
|
raise DeterministicFailure("article_schema", "section heading is required")
|
|
paragraphs = section.get("paragraphs")
|
|
if not isinstance(paragraphs, list) or not paragraphs or not all(
|
|
isinstance(value, str) and value.strip() for value in paragraphs
|
|
):
|
|
raise DeterministicFailure("article_schema", "section paragraphs are invalid")
|
|
return article
|
|
|
|
|
|
def _riyadh_iso(value: str) -> str:
|
|
parsed = datetime.fromisoformat(value.replace("Z", "+00:00"))
|
|
if parsed.tzinfo is None or parsed.utcoffset() is None:
|
|
raise ValueError("publishedAt must include a timezone")
|
|
return parsed.astimezone(RIYADH).isoformat()
|
|
|
|
|
|
class Publisher:
|
|
def __init__(self, *, appflowy, api, generator, media, lock_path: Path, allowed_hosts: Iterable[str] = ("anwarpy.ai",)):
|
|
self.appflowy = appflowy
|
|
self.api = api
|
|
self.generator = generator
|
|
self.media = media
|
|
self.lock_path = lock_path
|
|
self.allowed_hosts = set(allowed_hosts)
|
|
self.marker_errors: list[str] = []
|
|
|
|
def run(self) -> str:
|
|
with exclusive_run_lock(self.lock_path) as acquired:
|
|
if not acquired:
|
|
return "busy"
|
|
candidates = self._candidates()
|
|
publishing = sorted(
|
|
(candidate for candidate in candidates if candidate.marker.meta["status"] == "publishing"),
|
|
key=lambda candidate: (_queued_at(candidate), candidate.page["view_id"]),
|
|
)
|
|
if publishing:
|
|
return self._reconcile(publishing[0])
|
|
|
|
eligible = sorted(
|
|
(
|
|
candidate
|
|
for candidate in candidates
|
|
if candidate.marker.meta["status"] in ("pending", "republish")
|
|
),
|
|
key=lambda candidate: (_queued_at(candidate), candidate.page["view_id"]),
|
|
)
|
|
if not eligible:
|
|
return "idle"
|
|
return self._publish(eligible[0])
|
|
|
|
def _candidates(self) -> list[Candidate]:
|
|
found: list[Candidate] = []
|
|
self.marker_errors = []
|
|
for page in self.appflowy.list_pages():
|
|
page_id = page["view_id"]
|
|
blocks = self.appflowy.get_blocks(page_id)
|
|
try:
|
|
marker = find_marker(blocks)
|
|
except MarkerError as error:
|
|
self.marker_errors.append(f"{page_id}: {error}")
|
|
continue
|
|
if marker is not None:
|
|
found.append(Candidate(page=page, marker=marker, blocks=blocks))
|
|
return found
|
|
|
|
def _read_marker(self, candidate: Candidate) -> PageMarker:
|
|
marker = find_marker(self.appflowy.get_blocks(candidate.page["view_id"]))
|
|
if marker is None:
|
|
raise ValueError("publication marker disappeared")
|
|
return marker
|
|
|
|
def _edit_and_verify(self, candidate: Candidate, marker_text: str) -> PageMarker:
|
|
self.appflowy.edit_block(candidate.page["view_id"], candidate.marker.block_id, marker_text)
|
|
readback = self._read_marker(candidate)
|
|
if not same_meta(marker_text, readback.raw):
|
|
raise ValueError("AppFlowy marker read-back mismatch")
|
|
return readback
|
|
|
|
def _evidence(self, candidate: Candidate) -> str:
|
|
texts = [
|
|
str(block.get("text") or "")
|
|
for block in candidate.blocks
|
|
if block.get("id") != candidate.marker.block_id
|
|
]
|
|
sanitized = sanitize_text("\n\n".join(texts), allowed_hosts=self.allowed_hosts)
|
|
assert_publishable(sanitized, allowed_hosts=self.allowed_hosts)
|
|
if not sanitized.strip():
|
|
raise DeterministicFailure("sanitizer", "source is empty after sanitization")
|
|
return sanitized
|
|
|
|
def _block(self, candidate: Candidate, reason: str) -> str:
|
|
current = self._read_marker(candidate)
|
|
blocked = mark_blocked(current.raw, reason)
|
|
self._edit_and_verify(Candidate(candidate.page, current, candidate.blocks), blocked)
|
|
return f"blocked:{reason}"
|
|
|
|
def _publish(self, candidate: Candidate) -> str:
|
|
original = candidate.marker
|
|
meta = original.meta
|
|
try:
|
|
evidence = self._evidence(candidate)
|
|
article = _validate_article(
|
|
self.generator.generate(candidate.page["name"], evidence, meta),
|
|
meta["access"],
|
|
meta.get("slug") if meta["status"] == "republish" else None,
|
|
)
|
|
media = self.media.render(article, meta)
|
|
if not isinstance(media, dict) or "coverImage" not in media:
|
|
raise DeterministicFailure("cover", "cover output is incomplete")
|
|
except DeterministicFailure as error:
|
|
return self._block(candidate, error.reason)
|
|
except ValueError:
|
|
return self._block(candidate, "sanitizer")
|
|
|
|
current = self._read_marker(candidate)
|
|
if not same_meta(original.raw, current.raw):
|
|
raise ValueError("AppFlowy marker changed during generation")
|
|
|
|
claimed_text = mark_publishing(current.raw, article["slug"])
|
|
claimed = self._edit_and_verify(Candidate(candidate.page, current, candidate.blocks), claimed_text)
|
|
claimed_candidate = Candidate(candidate.page, claimed, candidate.blocks)
|
|
|
|
body = self._post_body(article, meta, media["coverImage"])
|
|
existing = self.api.lookup(article["slug"])
|
|
record = None
|
|
if meta["status"] == "republish":
|
|
if existing is None:
|
|
raise AmbiguousWrite("republish target is missing; POST is forbidden")
|
|
record = self.api.update(article["slug"], body)
|
|
elif existing is None:
|
|
try:
|
|
record = self.api.create(body)
|
|
except AmbiguousWrite:
|
|
record = self.api.lookup(article["slug"])
|
|
if record is None:
|
|
raise
|
|
elif existing.get("status") == "draft":
|
|
record = self.api.update(article["slug"], {"publish": True})
|
|
else:
|
|
record = existing
|
|
|
|
return self._finish(claimed_candidate, record)
|
|
|
|
def _reconcile(self, candidate: Candidate) -> str:
|
|
slug = candidate.marker.meta["slug"]
|
|
record = self.api.lookup(slug)
|
|
if record is None:
|
|
return f"awaiting-reconciliation:{slug}"
|
|
if record.get("status") == "draft":
|
|
record = self.api.update(slug, {"publish": True})
|
|
result = self._finish(candidate, record)
|
|
return result
|
|
|
|
def _finish(self, candidate: Candidate, record: Any) -> str:
|
|
if not isinstance(record, dict):
|
|
raise ValueError("agent API returned an invalid record")
|
|
expected = candidate.marker.meta
|
|
slug = expected["slug"]
|
|
for field in ("slug", "access", "category"):
|
|
if record.get(field) != expected[field]:
|
|
raise ValueError(f"agent API {field} mismatch")
|
|
if record.get("status") != "published":
|
|
raise ValueError("agent API post is not published")
|
|
published_at = record.get("publishedAt")
|
|
if not isinstance(published_at, str):
|
|
raise ValueError("agent API publishedAt is missing")
|
|
url = record.get("url") or f"https://anwarpy.ai/posts/{slug}"
|
|
if not self.api.verify_public(record):
|
|
raise PublicVerificationError(f"public verification failed for {slug}")
|
|
|
|
published = mark_published(candidate.marker.raw, _riyadh_iso(published_at), url)
|
|
self._edit_and_verify(candidate, published)
|
|
return f"published:{slug}"
|
|
|
|
@staticmethod
|
|
def _post_body(article: dict[str, Any], meta: dict[str, str], cover_image: dict[str, Any]) -> dict[str, Any]:
|
|
words = re.findall(r"\S+", " ".join(
|
|
paragraph
|
|
for section in article["sections"]
|
|
for paragraph in section["paragraphs"]
|
|
))
|
|
body: dict[str, Any] = {
|
|
"title": article["title"],
|
|
"slug": article["slug"],
|
|
"dek": article["dek"],
|
|
"content": _lexical(article),
|
|
"category": meta["category"],
|
|
"access": meta["access"],
|
|
"readingMinutes": max(1, math.ceil(len(words) / 200)),
|
|
"publish": True,
|
|
"coverImage": cover_image,
|
|
}
|
|
if meta["access"] == "vip":
|
|
body["teaser"] = article["teaser"]
|
|
if body["category"] != meta["category"]:
|
|
raise ValueError("category changed")
|
|
if body["access"] != meta["access"]:
|
|
raise ValueError("access changed")
|
|
return body
|