Search
Search covers the text of every page listed in every table of contents, across all versions of all products.
Indexing happens in the background
The server does not wait for the index before it starts serving. Opening an archive reads its metadata and table of contents only, which is a few kilobytes; the index is built afterwards, on a background worker.
Until it is ready the search page says so, and reports its progress, rather than returning an empty result set that looks like an answer.
When archives are added or removed the index is rebuilt and replaced in one step, so a search never sees a half-built index.
How queries are matched
- All terms must appear
-
A query of
stylesheet customizationreturns only pages containing both. Documentation search is a lookup rather than a discovery engine, and partial matches mostly produce noise. - Titles count for more
-
A term in a page title weighs more heavily than the same term in the body.
- Results are ranked
-
Ranking uses term frequency against how common the term is across the corpus, so a word that appears everywhere contributes little.
Filtering
Restrict a search to one product or one version with query parameters:
/search?q=install&slug=handbook
/search?q=install&slug=handbook&version=1.0.0Searching from within a product carries the product filter automatically.