Feature request: is it possible to add a query filter on "topics:"

[hanskr]

No description provided.

[keegancsmith]

This is quite github specific feature request. However, maybe a more generic "repository labels" would make sense here. For example we already have this sort of idea for supporting filtering by archived/fork/etc but that is a static list. Then we could expose in the query parser topic:.

• RawConfig is how we do archived/forked/etc
  https://sourcegraph.com/github.com/sourcegraph/zoekt@df7a7e7162cf7d7af4d4cdde3701c57950830676/-/blob/query/query.go?L40:6-40:15#tab=references
• I imagine to encode topics we would store a Set (map[string]struct{}) per repo. Probably doesn't need to live in memory per shard given the low amount of data.

FYI I don't think this would get worked on anytime soon, but happy to guide anyone who is motivated to do it.

[hanskr]

A generic labels/tags/topics functionality that could be populated by "something" per type of backend (topics for GH) (not sure if others have similar functionality) sound like a good approach, and would solve my needs fine.

For now, I have worked my way around it with a script that uses the GH api to translate topics => repo-names and construct a zoekt query with a repo regex.

[xavier-calland]

A generic labels/tags/topics feature would be great.

[xavier-calland]

@keegancsmith I can try to look at this feature, but I need details on the different modifications to be made.
Can you guide me?

[stefanhengl]

@xavier-calland

Here is how I would get started

1. The labels would probably have to live here. Why? For Sourcegraph we have an optimization that updates to metadata don't cause a reindex (see .meta files in the code). I think repository labels shouldn't cause a full reindex.
2. Indexing: Pick a code host (probably GitHub) and pipe the data from the code host all the way to the builder, which writes the metadata to disk.
3. Query language: you have to extend the query language. You could add EG a label: filter. Check out query.go and parse.go for inspiration.
4. Matchtree: The final piece is to use a match tree to skip the repos that don't have the correct label. Checkout RepoSet in matchtree.go or any of the other match trees there for inspiration.

EDIT:
5. UI: It would be nice to show the labels in UI, too.