Search by partial package name

[edsko]

For example, searching for diagrams- finds nothing, while it should find diagrams-pdf, diagrams-postscript, etc.

[dcoutts]

Searching for diagrams works of course.

The problem is that integrating substring search into a keyword-based search is rather tricky and ad-hoc, how are we to score a substring hit?

We can ad-hoc strip '-' off the search term of course :-)

[edsko]

Searching for diagrams works only by accident (because the various diagrams packages have diagrams in the description) and gives too many results. I would say that a package with a name that has the search term as a substring should be scored highly.

[dcoutts]

So honestly I'm not sure how to integrate substring search into a keyword-based search in a principled way.

[gbaz]

The correct thing seems to me to be to consider any word with a matching substring a "keyword" with a score proportional to the ratio substring/string?

[ghost]

@danchoi What has your experience with indexing documents lead you to conclude about this?

[gbaz]

"Searching for diagrams works only by accident (because the various diagrams packages have diagrams in the description)" -- this is not true (at least anymore). It works because we split diagrams-x package names and we have exact matches on the diagrams component. This seems fine to me?