takahe/api/pagination.py

261 lines
8.3 KiB
Python
Raw Normal View History

import dataclasses
import urllib.parse
from collections.abc import Callable
from typing import Any, Generic, Protocol, TypeVar
2022-12-11 23:54:51 -08:00
from django.db import models
2023-03-09 18:46:57 -08:00
from django.db.models.expressions import Case, F, When
from django.http import HttpRequest
from hatchway.http import ApiResponse
2023-03-09 18:46:57 -08:00
from activities.models import PostInteraction, TimelineEvent
T = TypeVar("T")
class SchemaWithId(Protocol):
"""
Little protocol type to represent schemas that have an ID attribute
"""
id: str
TI = TypeVar("TI", bound=SchemaWithId)
TM = TypeVar("TM", bound=models.Model)
class PaginatingApiResponse(ApiResponse[list[TI]]):
"""
An ApiResponse subclass that also handles pagination link headers
"""
def __init__(
self,
data: list[TI],
request: HttpRequest,
include_params: list[str],
**kwargs,
):
# Call superclass
super().__init__(data, **kwargs)
# Figure out if we need link headers
self._request = request
self.extra_params = self.filter_params(self._request, include_params)
link_header = self.build_link_header()
if link_header:
self.headers["link"] = link_header
@staticmethod
def filter_params(request: HttpRequest, allowed_params: list[str]):
params = {}
for key in allowed_params:
value = request.GET.get(key, None)
if value:
params[key] = value
return params
def get_part(self, data_index: int, param_name: str, rel: str) -> str | None:
"""
Used to get next/prev URLs
"""
if not self.data:
return None
# Use the ID of the last object for the next page start
params = dict(self.extra_params)
params[param_name] = self.data[data_index].id
return (
"<"
+ self._request.build_absolute_uri(self._request.path)
+ "?"
+ urllib.parse.urlencode(params)
+ f'>; rel="{rel}"'
)
def build_link_header(self):
parts = [
entry
for entry in [
self.get_part(-1, "max_id", "next"),
self.get_part(0, "min_id", "prev"),
]
if entry
]
if not parts:
return None
return ", ".join(parts)
@dataclasses.dataclass
class PaginationResult(Generic[T]):
"""
Represents a pagination result for Mastodon (it does Link header stuff)
"""
#: A list of objects that matched the pagination query.
results: list[T]
#: The actual applied limit, which may be different from what was requested.
limit: int
#: A list of transformed JSON objects
json_results: list[dict] | None = None
@classmethod
def empty(cls):
return cls(results=[], limit=20)
def next(self, request: HttpRequest, allowed_params: list[str]):
"""
Returns a URL to the next page of results.
"""
if not self.results:
return None
if self.json_results is None:
raise ValueError("You must JSONify the results first")
params = self.filter_params(request, allowed_params)
params["max_id"] = self.json_results[-1]["id"]
return f"{request.build_absolute_uri(request.path)}?{urllib.parse.urlencode(params)}"
def prev(self, request: HttpRequest, allowed_params: list[str]):
"""
Returns a URL to the previous page of results.
"""
if not self.results:
return None
if self.json_results is None:
raise ValueError("You must JSONify the results first")
params = self.filter_params(request, allowed_params)
params["min_id"] = self.json_results[0]["id"]
return f"{request.build_absolute_uri(request.path)}?{urllib.parse.urlencode(params)}"
def link_header(self, request: HttpRequest, allowed_params: list[str]):
"""
Creates a link header for the given request
"""
return ", ".join(
(
f'<{self.next(request, allowed_params)}>; rel="next"',
f'<{self.prev(request, allowed_params)}>; rel="prev"',
)
)
def jsonify_results(self, map_function: Callable[[Any], Any]):
"""
Replaces our results with ones transformed via map_function
"""
self.json_results = [map_function(result) for result in self.results]
def jsonify_posts(self, identity):
"""
Predefined way of JSON-ifying Post objects
"""
interactions = PostInteraction.get_post_interactions(self.results, identity)
self.jsonify_results(
lambda post: post.to_mastodon_json(
interactions=interactions, identity=identity
)
)
def jsonify_status_events(self, identity):
"""
Predefined way of JSON-ifying TimelineEvent objects representing statuses
"""
interactions = PostInteraction.get_event_interactions(self.results, identity)
self.jsonify_results(
lambda event: event.to_mastodon_status_json(
interactions=interactions, identity=identity
)
)
def jsonify_notification_events(self, identity):
"""
Predefined way of JSON-ifying TimelineEvent objects representing notifications
"""
interactions = PostInteraction.get_event_interactions(self.results, identity)
self.jsonify_results(
lambda event: event.to_mastodon_notification_json(interactions=interactions)
)
def jsonify_identities(self):
"""
Predefined way of JSON-ifying Identity objects
"""
self.jsonify_results(lambda identity: identity.to_mastodon_json())
@staticmethod
def filter_params(request: HttpRequest, allowed_params: list[str]):
params = {}
for key in allowed_params:
value = request.GET.get(key, None)
if value:
params[key] = value
return params
2022-12-11 23:54:51 -08:00
2022-12-11 23:38:02 -08:00
class MastodonPaginator:
"""
Paginates in the Mastodon style (max_id, min_id, etc).
2023-01-08 22:06:09 -08:00
Note that this basically _requires_ us to always do it on IDs, so we do.
2022-12-11 23:38:02 -08:00
"""
def __init__(
self,
default_limit: int = 20,
max_limit: int = 40,
):
self.default_limit = default_limit
self.max_limit = max_limit
def paginate(
self,
queryset: models.QuerySet[TM],
2022-12-11 23:38:02 -08:00
min_id: str | None,
max_id: str | None,
since_id: str | None,
limit: int | None,
2023-03-10 08:10:34 -08:00
home: bool = False,
) -> PaginationResult[TM]:
2023-01-08 22:06:09 -08:00
limit = min(limit or self.default_limit, self.max_limit)
2023-03-10 08:10:34 -08:00
filters = {}
id_field = "id"
reverse = False
if home:
# The home timeline interleaves Post IDs and PostInteraction IDs in an
# annotated field called "subject_id".
id_field = "subject_id"
queryset = queryset.annotate(
subject_id=Case(
When(type=TimelineEvent.Types.post, then=F("subject_post_id")),
default=F("subject_post_interaction"),
)
2022-12-11 23:38:02 -08:00
)
2023-03-10 08:10:34 -08:00
# These "does not start with interaction" checks can be removed after a
# couple months, when clients have flushed them out.
2023-03-09 18:46:57 -08:00
if max_id and not max_id.startswith("interaction"):
2023-03-10 08:10:34 -08:00
filters[f"{id_field}__lt"] = max_id
if since_id and not since_id.startswith("interaction"):
2023-03-10 08:10:34 -08:00
filters[f"{id_field}__gt"] = since_id
if min_id and not min_id.startswith("interaction"):
2022-12-11 23:38:02 -08:00
# Min ID requires items _immediately_ newer than specified, so we
# invert the ordering to accommodate
2023-03-10 08:10:34 -08:00
filters[f"{id_field}__gt"] = min_id
reverse = True
# Default is to order by ID descending (newest first), except for min_id
# queries, which should order by ID for limiting, then reverse the results to be
# consistent. The clearest explanation of this I've found so far is this:
# https://mastodon.social/@Gargron/100846335353411164
ordering = id_field if reverse else f"-{id_field}"
results = list(queryset.filter(**filters).order_by(ordering)[:limit])
if reverse:
results.reverse()
return PaginationResult(
2023-03-10 08:10:34 -08:00
results=results,
limit=limit,
)