The agent that grows with you
Python aiai-agentai-agentsanthropicchatgptclaudeclaude-codecodexhermeshermes-agentllmnous-researchopenai
"""Every event the Matrix plugin builds fits the homeserver's byte cap, whatever the script or shape of
the text: sends, stream/progress edits at the adapter's advertised budget, and the standalone sender."""
import json
import sys
import types
import pytest
from gateway.config import PlatformConfig
# The homeserver rejects an event whose canonical JSON exceeds 65,536 bytes (M_TOO_LARGE). E2EE
# base64-inflates the content by 4/3, and the envelope (ids, hashes, signatures) needs room too.
_EVENT_CAP = 65_536
_LINES = {
"english": "The **quick** brown fox jumps over the *lazy* dog, and `code` too.\n",
"cyrillic": "Быстрая **коричневая** лиса прыгает через *ленивую* собаку, и `код` тоже.\n",
"cjk": "敏捷的**棕色**狐狸跳过了*懒惰的*狗,还有`代码`也是。这是一个很长的段落。\n",
"code": "```html\n<a href=\"x&y\">it's <b><bold></b></a>\n```\n",
}
def _texts(chars: int) -> dict:
"""One text per script/shape, about *chars* characters long."""
texts = {name: line * (chars // len(line)) for name, line in _LINES.items()}
# Distinct Matrix IDs: each lands in m.mentions, which an edit carries twice.
texts["mentions"] = " ".join(f"@u{i}:a.b" for i in range(chars // 8))
return texts
def _fits_one_event(content: dict) -> bool:
size = len(json.dumps(content, ensure_ascii=False, separators=(",", ":")).encode("utf-8"))
return size * 4 // 3 + 4096 <= _EVENT_CAP
def _make_adapter():
from plugins.platforms.matrix.adapter import MatrixAdapter
return MatrixAdapter(PlatformConfig(enabled=True, token="syt_test_token", extra={
"homeserver": "https://matrix.example.org", "user_id": "@bot:example.org"}))
@pytest.mark.asyncio
async def test_sends_and_budget_sized_edits_fit_one_event():
from gateway.platforms.base import SendResult, _custom_unit_to_cp
adapter = _make_adapter()
sent, edits = [], []
async def send_room_message(chat_id, content):
sent.append(content)
return "$sent"
async def send_content_event(chat_id, content):
edits.append(content)
return SendResult(success=True, message_id="$edit")
adapter._send_room_message = send_room_message
adapter._send_content_event = send_content_event
chat = "!room:example.org"
# Stream previews and progress edits stay within the adapter's advertised per-chat budget.
limit, len_fn = adapter.max_message_length_for_chat(chat), adapter.message_len_fn_for_chat(chat)
for name, text in _texts(40_000).items():
assert (await adapter.send(chat, text)).success, name
preview = text[: _custom_unit_to_cp(text, limit, len_fn)]
assert (await adapter.edit_message(chat, "$preview", preview)).success, name
oversized = [len(json.dumps(c, ensure_ascii=False)) for c in sent + edits if not _fits_one_event(c)]
assert oversized == []
# Chunking keeps Markdown rendering: an ordinary chunk still carries its HTML.
assert all("formatted_body" in c for c in sent if "quick" in c["body"])
@pytest.mark.asyncio
async def test_standalone_sender_events_fit_one_event(monkeypatch):
from plugins.platforms.matrix.adapter import DEFAULT_MAX_MESSAGE_LENGTH, _standalone_send
payloads = []
class _Response:
status = 200
async def __aenter__(self):
return self
async def __aexit__(self, *exc):
return None
async def json(self):
return {"event_id": f"$e{len(payloads)}"}
class _Session:
async def __aenter__(self):
return self
async def __aexit__(self, *exc):
return None
def put(self, url, **kwargs):
payloads.append(kwargs["json"])
return _Response()
monkeypatch.setitem(sys.modules, "aiohttp", types.SimpleNamespace(ClientSession=_Session))
pconfig = PlatformConfig(enabled=True, token="syt_test_token", extra={"homeserver": "https://matrix.example.org"})
# send_message chunks plugin platforms on characters at the registry's max_message_length.
for name, text in _texts(DEFAULT_MAX_MESSAGE_LENGTH).items():
result = await _standalone_send(pconfig, "!room:example.org", text[:DEFAULT_MAX_MESSAGE_LENGTH])
assert result["success"], (name, result)
assert payloads
assert [len(json.dumps(p, ensure_ascii=False)) for p in payloads if not _fits_one_event(p)] == []