issue_comments: 843291675

Tables and views get "stream all rows" at the moment, so one workaround is to define a SQL view for your query - this only works for queries that don't take any parameters though (although you may be able to define a view and then pass it extra fields using the Datasette table interface, like on https://latest.datasette.io/fixtures/paginated_view?content_extra__contains=9)

I've explored this problem in a bit more detail in https://githu.com/simonw/django-sql-dashboard and I think I have a pattern that could work.

For your canned query, you could implement the pattern yourself by setting up two canned queries that look something like this:

https://github-to-sqlite.dogsheep.net/github?sql=select+rowid%2C+sha%2C+author_date+from+commits+order+by+rowid+limit+1000

sql select rowid, sha, author_date from commits order by rowid limit 1000 That gets you the first set of 1,000 results. The important thing here is to order by a unique column, in this case rowid - because then subsequent pages can be loaded by a separate canned query that looks like this: sql select rowid, sha, author_date from commits where rowid > :after order by rowid limit 1000 https://github-to-sqlite.dogsheep.net/github?sql=select+rowid%2C+sha%2C+author_date+from+commits+where+rowid+%3E+%3Aafter+order+by+rowid+limit+1000&after=1000

You then need to write code which knows how to generate these queries - start with the first query with no where clause (or if you are using rowid you can just use the second query and pass it ?after=0 for the first call) - then keep calling the query passing in the last rowid you recieved as the after parameter.

Basically this is an implementation of keyset pagination with a smart client. When Datasette grows the ability to do this itself it will work by executing this mechanism inside the Python code, which is how the "stream all rows" option for tables works at the moment.