import hashlib
import re
from collections.abc import AsyncIterator
from datetime import UTC, datetime
from urllib.parse import unquote, urljoin, urlparse
from xml.etree.ElementTree import Element, fromstring
import httpx
from haiku.rag.ingester.sources.base import (
FetchResult,
RevisionSnapshot,
SourceEvent,
SourceEventKind,
check_file_size,
)
from haiku.rag.ingester.sources.filter import (
FileFilter,
_default_supported_extensions,
)
# WebDAV PROPFIND uses XML with the DAV: namespace. Element tags arrive as
# Clark-notation strings like "{DAV:}response", so we precompile them once.
_DAV_NS = "DAV:"
_TAG_RESPONSE = f"{{{_DAV_NS}}}response"
_TAG_HREF = f"{{{_DAV_NS}}}href"
_TAG_PROPSTAT = f"{{{_DAV_NS}}}propstat"
_TAG_PROP = f"{{{_DAV_NS}}}prop"
_TAG_STATUS = f"{{{_DAV_NS}}}status"
_TAG_RESOURCETYPE = f"{{{_DAV_NS}}}resourcetype"
_TAG_COLLECTION = f"{{{_DAV_NS}}}collection"
_TAG_GETETAG = f"{{{_DAV_NS}}}getetag"
_TAG_GETLASTMODIFIED = f"{{{_DAV_NS}}}getlastmodified"
_TAG_GETCONTENTTYPE = f"{{{_DAV_NS}}}getcontenttype"
_MAX_PROPFIND_REDIRECTS = 5
_PROPFIND_BODY = b"""
"""
class _PropfindEntry:
"""One element decoded into the fields the source actually
uses. `is_collection` separates folders from files; revision is ETag
when present, otherwise the Last-Modified header value."""
__slots__ = ("href", "is_collection", "revision", "content_type")
def __init__(
self,
href: str,
*,
is_collection: bool,
revision: str | None,
content_type: str | None,
) -> None:
self.href = href
self.is_collection = is_collection
self.revision = revision
self.content_type = content_type
# Matches an ETag value with optional leading whitespace, optional weak
# marker ``W/``, optional surrounding double quotes, and optional trailing
# whitespace. The non-greedy capture pulls out just the opaque inner value.
_ETAG_RE = re.compile(r'^\s*(?:W/)?"?(.*?)"?\s*$')
def _strip_etag(value: str | None) -> str | None:
"""Return the opaque part of an ETag header value (or ``getetag`` element):
strip surrounding whitespace, the optional ``W/`` weak marker, and
optional surrounding double quotes. Returns ``None`` for empty input."""
if value is None:
return None
match = _ETAG_RE.match(value)
cleaned = match.group(1) if match else value.strip()
return cleaned or None
def _entry_from_response(response: Element) -> _PropfindEntry | None:
"""Decode a single . Returns None if the prop block is missing
or the entry didn't return HTTP 200 (e.g. 404 for a known-bad path)."""
href_el = response.find(_TAG_HREF)
if href_el is None or not href_el.text:
return None
href = href_el.text
is_collection = False
revision: str | None = None
last_modified: str | None = None
content_type: str | None = None
ok = False
for propstat in response.findall(_TAG_PROPSTAT):
status_el = propstat.find(_TAG_STATUS)
if status_el is None or not status_el.text:
continue
if " 200 " not in status_el.text:
continue
ok = True
prop = propstat.find(_TAG_PROP)
if prop is None:
continue
resourcetype = prop.find(_TAG_RESOURCETYPE)
if resourcetype is not None and resourcetype.find(_TAG_COLLECTION) is not None:
is_collection = True
etag_el = prop.find(_TAG_GETETAG)
if etag_el is not None and etag_el.text:
revision = _strip_etag(etag_el.text)
lm_el = prop.find(_TAG_GETLASTMODIFIED)
if lm_el is not None and lm_el.text:
last_modified = lm_el.text.strip() or None
ct_el = prop.find(_TAG_GETCONTENTTYPE)
if ct_el is not None and ct_el.text:
content_type = ct_el.text.split(";")[0].strip().lower() or None
if not ok:
return None
# Prefer ETag — stronger validator. Fall back to Last-Modified so revision
# detection still works against servers that don't return ETags on PROPFIND.
return _PropfindEntry(
href=href,
is_collection=is_collection,
revision=revision or last_modified,
content_type=content_type,
)
def _parse_multistatus(body: bytes) -> list[_PropfindEntry]:
"""Top-level multistatus parser. Raises ValueError on garbage XML so the
poller's circuit breaker can record a failure."""
try:
root = fromstring(body)
except Exception as exc: # ParseError + any defensive surprise
raise ValueError(f"Invalid PROPFIND response XML: {exc}") from exc
return [
entry
for response in root.findall(_TAG_RESPONSE)
if (entry := _entry_from_response(response)) is not None
]
def _resolve_href(href: str, base_url: str) -> str:
"""PROPFIND href values can be either absolute URLs or server-relative
paths. Resolve to absolute against base_url either way, then URL-decode
the path so the URI we store matches what a user would type."""
absolute = urljoin(base_url, href)
parsed = urlparse(absolute)
decoded_path = unquote(parsed.path)
rebuilt = parsed._replace(path=decoded_path)
return rebuilt.geturl()
class WebDAVSource:
def __init__(
self,
*,
source_id: str,
base_url: str,
username: str | None = None,
password: str | None = None,
headers: dict[str, str] | None = None,
ignore_patterns: list[str] | None = None,
include_patterns: list[str] | None = None,
supported_extensions: list[str] | None = None,
transport: httpx.AsyncBaseTransport | None = None,
max_file_size: int | None = None,
) -> None:
self.source_id = source_id
# Trailing slash matters: urljoin treats path-without-slash as a sibling
# link, so "https://srv/dav" joined with "subdir/x" gives ".../x".
self.base_url = base_url if base_url.endswith("/") else base_url + "/"
self.username = username
self.password = password
self.headers = dict(headers or {})
self.supported_extensions = (
list(supported_extensions)
if supported_extensions is not None
else _default_supported_extensions()
)
self.filter = FileFilter(
ignore_patterns=ignore_patterns,
include_patterns=include_patterns,
supported_extensions=self.supported_extensions,
)
# transport is for testing — production callers leave it None.
self._max_file_size = max_file_size
auth = (
(self.username, self.password)
if self.username is not None and self.password is not None
else None
)
# Follow redirects: Plone and other front-ended WebDAV servers 30x on
# trailing-slash normalisation and virtual-host rewrites; httpx defaults
# follow_redirects to False, which would surface those as errors. GET
# and HEAD (fetch) follow at the client level. PROPFIND (head/discover)
# is followed by hand in `_propfind` because httpx downgrades PROPFIND to
# GET on 302/303, which the final endpoint would reject or answer with a
# non-multistatus body.
self._http = httpx.AsyncClient(
auth=auth,
headers=self.headers,
transport=transport,
follow_redirects=True,
)
def supports(self, uri: str) -> bool:
return uri.startswith(self.base_url)
async def aclose(self) -> None:
await self._http.aclose()
async def _propfind(self, url: str, *, depth: str) -> tuple[httpx.Response, str]:
"""Issue a PROPFIND, following redirects by hand so the method and body
are preserved (httpx turns a redirected PROPFIND into a GET on 302/303).
Returns the final response and the URL it was served from.
Each hop re-issues through the client, which re-applies `auth`, so a
cross-host redirect is refused *before* the request is sent — otherwise
we would leak Basic credentials to the redirect target (httpx only
strips auth cross-host for its own auto-followed redirects)."""
host = urlparse(url).netloc
current = url
for _ in range(_MAX_PROPFIND_REDIRECTS):
response = await self._http.request(
"PROPFIND",
current,
headers={"Depth": depth, "Content-Type": "application/xml"},
content=_PROPFIND_BODY,
follow_redirects=False,
)
location = response.headers.get("location")
if response.is_redirect and location:
current = urljoin(current, location)
if urlparse(current).netloc != host:
raise ValueError(
f"WebDAV PROPFIND {url!r} redirected to a different host "
f"{current!r}; refusing to send credentials. Update "
f"base_url to the new host."
)
continue
return response, current
raise httpx.TooManyRedirects(
f"Exceeded {_MAX_PROPFIND_REDIRECTS} redirects for PROPFIND {url}",
request=response.request,
)
async def head(self, uri: str) -> str | None:
response, _ = await self._propfind(uri, depth="0")
if response.is_error:
return None
entries = _parse_multistatus(response.content)
if not entries:
return None
return entries[0].revision
async def fetch(self, uri: str) -> FetchResult:
if self._max_file_size is not None:
head = await self._http.head(uri)
content_length = head.headers.get("content-length")
if content_length is not None:
check_file_size(int(content_length), self._max_file_size, uri)
response = await self._http.get(uri)
response.raise_for_status()
body = response.content
content_type = (
response.headers.get("content-type", "application/octet-stream")
.split(";")[0]
.strip()
.lower()
)
# ETag from the GET response is the freshest revision; fall back to
# Last-Modified, matching HTTPSource's preference order.
revision = (
_strip_etag(response.headers.get("etag"))
or (response.headers.get("last-modified") or "").strip()
or None
)
extra: dict[str, str] = {}
last_modified = (response.headers.get("last-modified") or "").strip()
if last_modified:
extra["last_modified"] = last_modified
return FetchResult(
uri=uri,
body=body,
content_type=content_type,
content_hash=hashlib.md5(body, usedforsecurity=False).hexdigest(),
revision=revision,
extra_metadata=extra,
)
async def discover(
self,
since: RevisionSnapshot | None = None,
*,
known_uris: set[str] | None = None,
) -> AsyncIterator[SourceEvent]:
snapshot: dict[str, str] = dict(since) if since else {}
known = known_uris or set()
now = datetime.now(UTC)
seen: set[str] = set()
response, final_url = await self._propfind(self.base_url, depth="infinity")
response.raise_for_status()
# hrefs are resolved and filtered against base_url, and the worker
# resolves URIs back to this source by base_url too. A redirect that
# moves the collection to a different path would make every href fall
# outside base_url — silently emitting DELETEs for all known docs. Fail
# loudly instead so the operator points base_url at the new location.
# (_propfind already rejects cross-host redirects, so only the path can
# differ here; a same-path scheme upgrade stays transparent.)
if urlparse(final_url).path.rstrip("/") != urlparse(self.base_url).path.rstrip(
"/"
):
raise ValueError(
f"WebDAV collection {self.base_url!r} redirected to a different "
f"path {final_url!r}; update base_url to the new location."
)
entries = _parse_multistatus(response.content)
for entry in entries:
if entry.is_collection:
continue
uri = _resolve_href(entry.href, self.base_url)
# The base URL itself sometimes appears as a non-collection on
# broken servers; skip anything that's not strictly under it.
if uri == self.base_url.rstrip("/") or not uri.startswith(self.base_url):
continue
if not self.filter.include_file(uri):
continue
seen.add(uri)
revision = entry.revision
if revision is not None and snapshot.get(uri) == revision:
kind = SourceEventKind.UNCHANGED
elif revision is None and uri in known:
kind = SourceEventKind.UNCHANGED
else:
kind = SourceEventKind.UPSERT
yield SourceEvent(
source_id=self.source_id,
uri=uri,
kind=kind,
revision=revision,
discovered_at=now,
)
for uri in known - seen:
yield SourceEvent(
source_id=self.source_id,
uri=uri,
kind=SourceEventKind.DELETE,
revision=None,
discovered_at=now,
)