False negative of a filter on a hashtag

[AdeptVeritatis]

Steps to reproduce the problem

1. Followed the hashtag #aurora a while ago.
2. Filtered the term "aurora" for a week.
3. One post was visible, although I am not following this account and the only connection seems to be the hashtag.

Expected behaviour

The post should not have shown up, but filtered correctly.

Actual behaviour

I could see the post on my starting page.

Detailed description

the affected post: https://aaronparecki.com/2024/05/11/3/aurora

I could not see the hashtags of this post at all. Only when I opened the original page.

Mastodon instance

social.tchncs.de

Mastodon version

v4.2.8

Browser name and version

Firefox 125.0.3

Operating system

Arch

Technical details

No response

[trwnh]

My guess is that the filter checks for the content of the post, not the actual hashtag itself... whereas the hashtag follow checks for the hashtag itself, not the content of the post.

[renchap]

Looks like out-of-band hashtags are not checked for filtering, only the post content (which might filter some hashtags if they also appear in the text representation)

[ClearlyClaire]

Yeah, filters apply to the contents of the post (text, poll options and media descriptions), but not out-of-band hashtags. I wonder if we should extend the filtering to out-of-band hashtags as well, or add a specific filter type for hashtags.

[renchap]

I think I would apply the filter to out-of-band hashtags as well, because this is the behaviour we currently have for in-band hashtags

[ThisIsMissEm]

Huh, interesting, this may have implications for using searchable_text on the streaming server (this improves streaming's performance by not using JSDOM). Additionally, there is a bug regarding filtering hashtags using keyword filters, see #27990 — the PR for that would add a proper CustomFilterHashtag implementation.

[trwnh]

I wonder if we should extend the filtering to out-of-band hashtags as well, or add a specific filter type for hashtags.

i'd be in favor of having both a CustomFilterHashtag and a CustomFilterMention as these are distinct from the textual content, and can be presented in multiple forms (the text inside the anchor link can be anything, including not present). etc etc of course, see #18955 for that