issue_comments

Another demo: https://fivethirtyeight.datasettes.com/fivethirtyeight-2628db9/congress-age%2Fcongress-terms

This is now released in Datasette 0.15 https://github.com/simonw/datasette/releases/tag/0.15

I've merged this into master.

Here's a demo of the new clickable column headers: https://datasette-issue-189-demo-3.now.sh/salaries-7859114-7859114/2017+Maryland+state+salaries?_search=university&_sort_desc=last_name

Actually I think I always want nulls last when ordering asc, nulls first when ordering desc.

I'm going to split the following out into separate tickets:

• Ability to sort by multiple columns e.g. ?_sort=name&sort_desc=age&_sort=height
• Ability to specify nulls last e.g. ?_sort_desc_nulls_last=age

Small bug: "201 rows where sorted by sortable_with_nulls" shouldn't have the word "where" in it.

Another demo: https://datasette-issue-189-demo-2.now.sh/salaries-7859114-7859114/2017+Maryland+state+salaries?_search=university&_sort_desc=annual_salary

https://datasette-issue-189-demo-2.now.sh/salaries-7859114-7859114/2017+Maryland+state+salaries?_search=university&last_name__exact=JOHNSON&_sort_desc=annual_salary

Demo: senator tweets ordered by number of replies:

https://datasette-issue-189-demo.now.sh/fivethirtyeight-2628db9/twitter-ratio%2Fsenators?_sort_desc=replies

Page 2 (note that since Senators retweet things there are tweets with the same text/number-of-replies but retweeted by different senators that span the page break): https://datasette-issue-189-demo.now.sh/fivethirtyeight-2628db9/twitter-ratio%2Fsenators?_next=8556%2C121799&_sort_desc=replies

Actually next page SQL when sorting looks more like this:

select rowid, * from [alcohol-consumption/drinks] where "country" like :p0 and ( beer_servings > 111 or (beer_servings = 111 and rowid > 190) ) order by beer_servings, rowid limit 101

The next page after row 190 with sortable value 111 should show either records that are greater than 111 or records that match 111 but have a greater primary key than the last one seen.

https://fivethirtyeight.datasettes.com/fivethirtyeight-2628db9?sql=select+rowid%2C+*+from+%5Balcohol-consumption%2Fdrinks%5D%0D%0Awhere+%22country%22+like+%3Ap0%0D%0Aand+%28%0D%0A++++beer_servings+%3E+111%0D%0A++++or+%28beer_servings+%3D+111+and+rowid+%3E+190%29%0D%0A%29%0D%0Aorder+by+beer_servings%2C+rowid+limit+101&p0=%25a%25

To break this up into smaller units, the first implementation of this will only support a single _sort or _sort_desc querystring parameter.

A note about views: a view cannot be paginated using keyset pagination because records returned from a view don't have a primary key - so there's no way to reliably distinguish between _next= records when the sorted column has duplicates with the same value.

Datasette already takes this into account: views are paginated using offset/limit instead. We can continue to do that even for views that have been sorted using a _sort parameter.

A common problem with keyset pagination is that it can distort the "total number of rows" logic - every time you navigate to a further page the total rows count can decrease due to the extra arguments in the where clause. The filtered_table_rows value (see #194) calculated using count_sql currently has this problem.

I'm going to combine the code for explicit sorting with the existing code for _next= pagination - so even tables without an explicit sort order will run through the same code since they are ordered and paginated by primary key.

I think this is the right incantation for a "next" link: https://datasette-column-sort-wip.now.sh/sortable-4bbaa6f?sql=select+*+from+sortable%0D%0Awhere+sortable+%3C%3D+94%0D%0Aand+%28%0D%0A++%28pk1+%3E+%27d%27%29%0D%0A++or%0D%0A++%28pk1+%3D+%27d%27+and+pk2+%3E+%27w%27%29%0D%0A%29%0D%0Aorder+by+sortable+desc%2C+pk1%2C+pk2%0D%0Alimit+7

Pushed some work-in-progress with failing unit tests here: https://github.com/simonw/datasette/commit/2f8359c6f25768805431c80c74e5ec4213c2b2a6

Here's a demo: https://datasette-column-sort-wip.now.sh/sortable-4bbaa6f/sortable?_sort=sortable - note that the _sort_desc and _sort_nulls_last options aren't done yet, plus it doesn't correctly paginate (the _next tokens do not yet take sorting into account).

Maybe the answer here is that anything that's encoded in the next token is treated as >= with the exception of columns known to be primary keys, which are treated as >

The problem is that our _next= pagination currently works based on a > - but for this case a >= for the value is needed combined with a > on the tie-breaker (which would be the rowid column).

So I think this is the right SQL:

select rowid, * from [nfl-wide-receivers/advanced-historical] where career_ranypa >= -6.331167749 and rowid > 2736 order by case when career_ranypa is null then 1 else 0 end, career_ranypa, rowid limit 11

https://fivethirtyeight.datasettes.com/fivethirtyeight-2628db9?sql=select+rowid%2C+*+from+%5Bnfl-wide-receivers%2Fadvanced-historical%5D%0D%0Awhere+career_ranypa+%3E%3D+-6.331167749+and+rowid+%3E+2736%0D%0Aorder+by+case+when+career_ranypa+is+null+then+1+else+0+end%2C+career_ranypa%2C+rowid+limit+11

But how do I encode a _next token that means ">= X and > Y"?

I'm not entirely sure how to get _next= pagination working against sorted collections when a tie-breaker is needed.

Consider this data:

https://fivethirtyeight.datasettes.com/fivethirtyeight-2628db9?sql=select+rowid%2C+*+from+%5Bnfl-wide-receivers%2Fadvanced-historical%5D%0D%0Aorder+by+case+when+career_ranypa+is+null+then+1+else+0+end%2C+career_ranypa%2C+rowid+limit+11

If the page size was set to 9 rather than 11, the page divide would be between those two rows with the same value in the career_ranypa column. What would the ?_next= token look like such that the correct row would be returned?

Alternative idea: by default enable all sorting in the UI.

If a table has more than 100,000 rows disable sorting UI except for columns that have an index.

Allow this to be overridden in metadata.json

There is one other interesting option for auto-enabling/disabling sort: the inspect command could include data about column index presence and whether or not a column has any null values in it.

This would allow us to dynamically include a "nulls last" option but only for columns that contain at least one null.

It's quite a lot of additional engineering for a very minor feature though, so I think I'll punt on that for the moment.

We may find that the _group_count feature can benefit from column value statistics later on though.

I'm tempted to put these verbose sorting options inline in the page HTML but have them in the table footer so they don't clog up the top half of the page with uninteresting links - then use JavaScript to hoik them out into a dropdown menu attached to each column header.

I think there are actually four kinds of sort order we need to support;

• ascending
• descending
• ascending, nulls last
• descending, nulls last

It looks like [-blah] is a valid SQLite table name, so mark I descending with a hyphen prefix isn't good.

Instead, maybe this:

?_sort_asc=col1&_sort_desc_nulls_last=col2

In terms of user interface: the obvious place to put this is as a drop down menu on the column headers.

This also means the UI can support combined sort orders. Assuming you are already sorted by county descending and you select the candidate column header, the options could be:

• sort all by candidate
• sort all by candidate, descending
• sort by county descending, then by candidate
• sort by county descending, then by candidate descending

I'd like to continue to support _next=token pagination even for custom sort orders.

To do that I should include rowid (or general primary key) as the tie breaker on all sorts so I can incorporate that it into the _next= token.

This is a better pattern as you don't have to pick a minimum value:

ORDER BY CASE WHEN SOMECOL IS NULL THEN 1 ELSE 0 END, SOMECOL

Might have to do something special to get sort-by-nulls-last: https://stackoverflow.com/questions/12503120/how-to-do-nulls-last-in-sqlite

order by ifnull(column_name, -999999)

Would need to figure out a smart way to get the default value - maybe by running a min() or max() against the column first?

I think this can work with a ?_sort=xxx parameter - and ?_sort=-xxx to sort in the opposite direction.

I'd like to support "sort by X descending, then by Y ascending if there are dupes for X" as well. Two ways that could work:

?_sort=-xxx,yyy

Or...

?_sort=-xxx&_sort=yyy

The second option is probably better in that it makes it easier for columns to have a comma in their name.

Is it possible for a SQLite column to start with a - character?

Given how unlikely it is that this will pose a real problem I think I like option 1: enable sort-by-column by default for all tables, then allow power users to instead switch to explicit enabling of the functionality in their metadata.json if they know their data is too big.