[Python lark-oapi 1.6.9] InboundMessage.mentioned_all is always False for post (rich-text) @all messages

[wuzhenhua24]

Summary

For post (rich-text) messages that mention everyone (@所有人 / @all),
InboundMessage.mentioned_all is always False. As a result PolicyGate
never emits policy_mention_all_blocked, so a bot configured with
respond_to_mention_all=False still responds to broadcast @all posts.

Plain text @all messages are handled correctly — only the post code
path is affected.

Root cause

In lark_oapi/channel/normalize/pipeline.py, the PostContent branch of
InboundPipeline.process derives mentioned_all only from:

1. ext.mentioned_all (the event mentions[] array), and
2. text_has_mention_all(content.text) / parse_at_tags(content.text)
   (the rendered text).

But for a post:

• Feishu does not populate mentions[] for @all (already acknowledged in the
  code comments on the text branch).
• The @all signal lives in the post AST as an at node with
  user_id == "all" ({"tag":"at","user_id":"all"}). converters/post.py
  renders that node to the literal string @所有人 / @<user_name> — not
  the @_all placeholder that text_has_mention_all looks for, and not an
  <at> tag that parse_at_tags matches.

So neither probe fires and mentioned_all stays False. The post AST is never
inspected for the mention-all node.

Reproduction (confirmed on 1.6.9, also reproduces on 1.6.6)

import asyncio, json
from lark_oapi.channel.normalize.pipeline import InboundPipeline, PipelineConfig, PipelineDeps
from lark_oapi.channel.config import PolicyConfig
from lark_oapi.channel.safety.policy_gate import PolicyGate

async def main():
    pipe = InboundPipeline(PipelineConfig(), PipelineDeps())
    # A post that @-mentions everyone, like a broadcast notice
    content = {"content": [[
        {"tag": "at", "user_id": "all", "user_name": "Everyone"},
        {"tag": "text", "text": " heads up everyone"},
    ]]}
    event = {"message_id": "om", "chat_id": "oc", "chat_type": "group",
             "message_type": "post", "content": json.dumps(content), "mentions": []}
    sender = {"sender_id": {"open_id": "ou_asker"}, "sender_type": "user"}
    inbound = await pipe.process("e", event, sender)
    print("post  mentioned_all =", inbound.mentioned_all)

    gate = PolicyGate(PolicyConfig(require_mention=False, respond_to_mention_all=False))
    gate.set_bot_open_id("ou_bot")
    print("post  decision      =", gate.evaluate(inbound))

    # same intent as plain text — handled correctly
    tevent = dict(event, message_type="text", content=json.dumps({"text": "@_all heads up"}))
    tin = await pipe.process("e2", tevent, sender)
    print("text  mentioned_all =", tin.mentioned_all)
    print("text  decision      =", gate.evaluate(tin))

asyncio.run(main())

Output:

post  mentioned_all = False
post  decision      = PolicyDecision(allowed=True, reason=None)
text  mentioned_all = True
text  decision      = PolicyDecision(allowed=False, reason='policy_mention_all_blocked')

Expected: the post case should match the text case
(mentioned_all = True, decision policy_mention_all_blocked).

Suggested fix

In the PostContent branch, probe the raw post AST for an at node with
user_id == "all" (mirroring is_mention_all):

elif isinstance(content, PostContent):
    if not mentioned_all and text_has_mention_all(content.text):
        mentioned_all = True
    # post @all is an `at` node (user_id == "all") in the AST, not an
    # @_all placeholder in the rendered text — probe the raw post.
    if not mentioned_all and _post_has_mention_all(content.post):
        mentioned_all = True
    ...

where _post_has_mention_all walks the post documents/paragraphs for any at
element whose user_id (or open_id) equals "all".

Environment

• lark-oapi 1.6.9 (also 1.6.6)
• Python 3.11

Possibly related: #134 (InboundMessage.mentioned_bot always False for IM
messages) — same family of post/IM mention-resolution gaps.

[wuzhenhua24]

Update: the real wire value is @_all, not all

Captured an actual @all post off the wire and the mention-all marker does
not match the is_mention_all convention. The at node carries
user_id == "@_all" (with the @_ prefix), not "all":

// message_type = "post", event mentions[] = []  (empty)
{
  "title": "",
  "content": [
    [
      {"tag": "at", "user_id": "@_all", "user_name": "所有人", "style": []},
      {"tag": "text", "text": " ", "style": []}
    ],
    [{"tag": "img", "image_key": "img_v3_...", "width": 1263, "height": 347}],
    [{"tag": "text", "text": " some text", "style": []}]
  ]
}

For this message InboundMessage.mentioned_all is False and the rendered
content_text is @所有人 \n\n![image](img_v3_...)\n\n some text — so the only
faithful signal is the at node's user_id.

This means the gap is a bit wider than the original report:

• is_mention_all() (in normalize/mentions.py) only matches
  id.user_id == "all", and parse_at_tags() only matches uid == "all".
  Neither catches the @_all value seen on the wire — so even if the post AST
  were probed, matching "all" alone would still miss real @all posts.

Suggest accepting both values wherever the mention-all id is checked, e.g.:

_MENTION_ALL_IDS = frozenset({"all", "@_all"})
...
if uid in _MENTION_ALL_IDS:
    mentioned_all = True

(applied to is_mention_all, parse_at_tags, and the proposed PostContent AST
probe).

Tested on 1.6.9 and 1.6.6.