Search Bar for Issues and Pull Requests

[ethantkoenig]

Add a search bar for searching issues and pull requests. This would be similar to Github's search bar.

I would be happy to implement this proposal, but just want to make sure there is interest before doing so.

[bkcsoft]

Fire at will 😄

[tboerger]

Makes sense for sure, but I'm also sure we need some kind of indexer or cache layer.

[lunny]

@tboerger, indexer could be put in 1.x.x, we could implement it use database like as a short term solution.

[tboerger]

@lunny the only think in the database are issues. For choose search you are forced to search within the repo.

[ethantkoenig]

I looked into using a full-text-index to speed up searches, specifically for Postgres. Postgres has support for full-text-indices (docs), but they do not seem to interact with xorm well.

On calls to Sync2(..), xorm tries to delete any index that is not listed in the corresponding struct's definition (session.go, line 4091). The full-text-index is not listed in the struct's definition, so xorm tries to delete it.

Is there some way to get xorm to be okay with the full-text-index, particularly a way that will work for all supported RDBMSs? Any help would be appreciated 😃

[lunny]

But maybe not all the database support full-text-index? Another way is to store index ourself, and we can have a task called index which will index issue titles and contents, wiki title and contents and codes?

[lunny]

Maybe we can use http://www.blevesearch.com/ to do that.

[tboerger]

I would also suggest a bleeve based index.

[ethantkoenig]

I've begun looking into bleve. Just to double-check, the idea is to

• maintain one bleve index per repository, and
• during an issue search, issue both a SQL query (for other criteria) and a search on the index (for keyword), and then intersect the two results?

Let know me if I'm missing something.

[lunny]

If we use bleve, SQL query will be no need. I think one repo one index file is not good idea. I prefer one function one index. So that issue/pr index, repo index, wiki index. And in fact this work's emphasis is splitting words in different languages or mixed languages.

[ethantkoenig]

If we don't have a SQL query, how can we support existing filters like tags, or issues created by a particular user?