issue_comments

Rather than trying to run that monstrosity in a single union all query, a better approach may be to use fetch() requests as seen in https://datasette.io/plugins/datasette-search-all

Figured out a query for searching across every column in every table! https://til.simonwillison.net/datasette/search-all-columns-trick#user-content-same-trick-for-the-entire-database

sql with tables as ( select name as table_name from sqlite_master where type = 'table' ), queries as ( select 'select ''' || tables.table_name || ''' as _table, rowid from "' || tables.table_name || '" where ' || group_concat( '"' || name || '" like ''%'' || :search || ''%''', ' or ' ) as query from pragma_table_info(tables.table_name), tables group by tables.table_name ) select group_concat(query, ' union all ') from queries The SQL query this generates for larger databases is extremely long - but it does seem to work for smaller databases.

A related but potentially even more useful ability would be running a search across every column of every table in a whole database. For anything less than a few 100MB this could be incredibly useful.

The bigger problem here is UI design. This feels like a pretty niche requirement to me, so adding a prominent search box to the table page (which already has the filters interface, plus the full-text search box for tables that have FTS configured) feels untidy.

I could tuck it away in the table cog menu, but that's a weird place for something like this to live.

Maybe add it as a new type of filter? Filters apply to specific columns though, so this would be the first filter that applied to all columns - which doesn't really fit the existing filter interface very well.

The downside of the current implementation of this trick is that it only works for exact LIKE partial matches in a specific table - if you search for dog cat and dog appears in title but cat appears in description you won't get back that result.

I think that's fine though. If you want more advanced search there are other mechanisms you can use. This is meant to be a very quick and dirty starting point for exploring a brand new table.