Synapse cuts out some words from search leading to confusion

[gdvine]

Description

Desktop client in OS X returns no results, ever.

Steps to reproduce

• Click on lupe button
• Type any search term, such as "the", click submit lupe. (Either "This Room" or "All Rooms")
• Lupe flashes for 10 seconds or so. Below: "No results"

Expected: some results.

Log: sent

• Desktop
• OS X
• Version 0.12.2

[t3chguy]

@gdvine are you in an E2E encrypted room? Searches are currently server-side and thus cannot be done on E2E messages

[turt2live]

Strangely, I can reproduce this on Chrome 60 web as well:

The response is a 200 OK from the server:
{"search_categories":{"room_events":{"count":0,"highlights":[],"results":[]}}}

Other terms, like e and hello work, but the fails every time.

[t3chguy]

maybe Synapse cuts the and other such words out of the search phrase to get more results
because it does do quite a wide search

[turt2live]

I'm going to also blame synapse. Words like it, my, and on also fail. Other stop words like must work though.

[gdvine]

E2E encryption is on, so that explains why the searching fails. Perhaps a conditional should be introduced which bypasses E2E encrypted rooms, and returns a message if no unencrypted rooms are found.

Is there a proposal for client-side search anywhere?

[gdvine]

Bad example – not only "the" but any string whatsoever returns no results as all my rooms are E2E.

[t3chguy]

@gdvine #2548

turning this thread into better tracking of the implicit behaviour of the and co

[lampholder]

Yuck!

Whatever else we decide to do, we should tell the user what we're doing if we're modifiying their search.

How easily could we do something like:

[ara4n]

This is particularly unpleasant when searching for terms in URLs: https://github.com/vector-im/riot-web/issues/4486

[grahamperrin]

Today's https://matrix.to/#/!GdmnfkDxuKZAICnOhe:matrix.org/$15054991507323727AFZiy:matrix.org ends with this string:

to be /theme and /doc :D

Seeking /theme in that room does not find that line.

---

Side note: I stumbled across that example, and others involving non-alphanumeric characters, whilst trying to make #5007 reproducible.

[t3chguy]

Closing in favour of matrix-org/synapse#10559