issue_comments

Most complicated example of a paginated query: https://latest.datasette.io/fixtures?sql=select%0D%0A++pk1%2C%0D%0A++pk2%2C%0D%0A++content%2C%0D%0A++sortable%2C%0D%0A++sortable_with_nulls%2C%0D%0A++sortable_with_nulls_2%2C%0D%0A++text%0D%0Afrom%0D%0A++sortable%0D%0Awhere%0D%0A++(%0D%0A++++sortable_with_nulls+is+null%0D%0A++++and+(%0D%0A++++++(pk1+%3E+%3Ap0)%0D%0A++++++or+(%0D%0A++++++++pk1+%3D+%3Ap0%0D%0A++++++++and+pk2+%3E+%3Ap1%0D%0A++++++)%0D%0A++++)%0D%0A++)%0D%0Aorder+by%0D%0A++sortable_with_nulls+desc%2C%0D%0A++pk1%2C%0D%0A++pk2%0D%0Alimit%0D%0A++101&p0=h&p1=r

sql select pk1, pk2, content, sortable, sortable_with_nulls, sortable_with_nulls_2, text from sortable where ( sortable_with_nulls is null and ( (pk1 > :p0) or ( pk1 = :p0 and pk2 > :p1 ) ) ) order by sortable_with_nulls desc, pk1, pk2 limit 101 Generated by this page: https://latest.datasette.io/fixtures/sortable?_next=%24null%2Ch%2Cr&_sort_desc=sortable_with_nulls

The _next= parameter there decodes as $null,h,r - and those components are tilde-encoded, so this can be distinguished from an actual $null value which would be represented as ~24null.

Hey, isn't this essentially the same issue as #448 ?

I'm going to build completely separate tests for this in test_pagination.py.

lets users make schema changes, so it's important to me that the tool work in a non-surprising way -- if you ask for a column of type X, you should get type X. If the column or table previously had CHECK constraints, they shouldn't be silently removed

I've got your concern. Let's see if we will be replied on it and i'll close the issue some later.

Ha, yes, I might end up making something very niche. That's OK.

I'm building a UI for Datasette that lets users make schema changes, so it's important to me that the tool work in a non-surprising way -- if you ask for a column of type X, you should get type X. If the column or table previously had CHECK constraints, they shouldn't be silently removed. And so on. I had hoped that I could just lean on sqlite-utils, but I think it's a little too surprising.

Ah, it looks like that is controlled by this dict: https://github.com/simonw/sqlite-utils/blob/main/sqlite_utils/db.py#L178

I suspect you could overwrite the datetime entry to achieve what you want

And thank you for pointing me to it. At least, i can make a monkey patch for my need...

Isn't your suggestion too fundamental for the utility?

The bigger flexibility, the bigger complexity. Your idea make sense defenitely, but how often do you make schema changes? And how many people could benefit from it, what do you think?

I think the bigger issue is that sqlite-utils mixes mechanism (it implements the 12-step way to alter SQLite tables) and policy (it has an opinionated stance on what column types should be used).

That might be a design choice to make it accessible to users by providing a reasonable set of defaults, but it doesn't quite fit my use case.

It might make sense to extract a separate library that provides just the mechanisms, and then sqlite-utils would sit on top of that library with its opinionated set of policies.

That would be a very big change, though.

I might take a stab at extracting the library, but just for the table schema migration piece, not all the other features that sqlite-utils supports. I wouldn't expect sqlite-utils to depend on it.

Part of my motivation is that I want to provide some other abilities, too, like support for CHECK constraints. I see that the issue in this repo (https://github.com/simonw/sqlite-utils/issues/358) proposes a bunch of short-hand constraints, which I wouldn't want to accidentally expose to people -- I want a layer that is a 1:1 mapping to SQLite.

Do you see a way to enable it without affecting existing users or bumping the major version number?

I don't see a clean solution, only extending code with a side variable that tells us we want to apply advanced types instead of basic.

it could be a similiar command like tranform-v2 --type column DATETIME or a cli option transform --adv-type column DATETIME along with a dict that contains the advanced types. Then with knowledge that we run an advanced command we take that dictionary somehow, we can wrap the current and new dictionaries by a superdict and work with it everywhere according to the knowledge. This way shouldn't affect users who are using the previous lib versions and it have to be merged in the next major one.

But this way looks a bad design, too messy.

I'd support that, but I'm not the author of this library.

One challenge is that would be a breaking change. Do you see a way to enable it without affecting existing users or bumping the major version number?

I could, of course.

Doest it worth bringing such the improvement to the library?

Is this feature covered by any more recent updates to datasette, or via any plugins that you're aware of?

Ah, it looks like that is controlled by this dict: https://github.com/simonw/sqlite-utils/blob/main/sqlite_utils/db.py#L178

I suspect you could overwrite the datetime entry to achieve what you want

That said, it looks like the check is only enforced at the CLI level. If you use the API directly, I think it'll work.

It works, but a column becomes TEXT

```python In [1]: import sqlite_utils In [2]: db = sqlite_utils.Database('events.sqlite') In [3]: table = db['cards.chunk.get'] In [4]: table.columns_dict Out[4]: {'id': int, 'timestamp': float, 'data_chunk_number': int, 'user_id': str, 'meta_duplication_source_id': int, 'context_sort_attribute': str, 'context_sort_order': str}

In [5]: from datetime import datetime In [7]: table.transform(types={'timestamp': datetime}) In [8]: table.columns_dict Out[8]: {'id': int, 'timestamp': str, 'data_chunk_number': int, 'user_id': str, 'meta_duplication_source_id': int, 'context_sort_attribute': str, 'context_sort_order': str} ```

bash ❯ sqlite-utils schema events.sqlite cards.chunk.get CREATE TABLE "cards.chunk.get" ( [id] INTEGER PRIMARY KEY NOT NULL, [timestamp] TEXT, ...

Doing this as a class makes sense to me. There are a few steps:

• Instantiate the class with the information it needs, which includes sort order, page size, tiebreaker columns and SQL query and parameters
• Generate the new SQL query that will actually be executed - maybe this takes the optional _next parameter? This returns the SQL and params that should be executed, where the SQL now includes pagination logic plus order by and limit
• The calling code then gets to execute the SQL query to fetch the rows
• Last step: those rows are passed to a paginator method which returns (rows, next) - where rows is the rows truncated to the correct length (really just with the last one cut off if it's too long for the length) and next is either None or a token, depending on if there should be a next page.

So I think I can write an abstraction that applies keyset pagination to ANY arbitrary SQL query provided it is given the query, the existing params (so it can pick names for the new params that won't overlap with them), the desired sort order, any existing _next token AND the columns that should be used to tie-break any duplicates.

Those tie breakers will be either the primary key(s) or rowid if none are provided.

What about the case of SQL views, where offset/limit should be used instead? I'm inclined to have that as a separate pagination abstraction entirely, with the calling code deciding which pagination helper to use based on if keyset pagination makes sense or not.

Might be easier to design a class structure for this starting with OffsetPaginator, then using that to inform the design of KeysetPaginator.

Might put these in datasette.utils.pagination to start off with, then maybe extract them out to sqlite-utils later once they've proven themselves.

Even more complicated: https://latest.datasette.io/fixtures/sortable?sortable_with_nulls__notnull=1&_next=0~2E692704598586882%2Ce%2Cr&_sort=sortable_with_nulls_2

The rewritten SQL for that is:

sql select * from (select pk1, pk2, content, sortable, sortable_with_nulls, sortable_with_nulls_2, text from sortable where "sortable_with_nulls" is not null) where (sortable_with_nulls_2 > :p2 or (sortable_with_nulls_2 = :p2 and ((pk1 > :p0) or (pk1 = :p0 and pk2 > :p1)))) order by sortable_with_nulls_2, pk1, pk2 limit 101 And it still has the same number of explain steps as the current SQL witohut the subselect.

A more complex example: https://latest.datasette.io/fixtures/sortable?_next=0~2E2650566289400591%2Ca%2Cu&_sort=sortable_with_nulls_2

SQL:

sql select pk1, pk2, content, sortable, sortable_with_nulls, sortable_with_nulls_2, text from sortable where (sortable_with_nulls_2 > :p2 or (sortable_with_nulls_2 = :p2 and ((pk1 > :p0) or (pk1 = :p0 and pk2 > :p1)))) order by sortable_with_nulls_2, pk1, pk2 limit 101

https://latest.datasette.io/fixtures?sql=select+pk1%2C+pk2%2C+content%2C+sortable%2C+sortable_with_nulls%2C+sortable_with_nulls_2%2C+text+from+sortable+where+%28sortable_with_nulls_2+%3E+%3Ap2+or+%28sortable_with_nulls_2+%3D+%3Ap2+and+%28%28pk1+%3E+%3Ap0%29%0A++or%0A%28pk1+%3D+%3Ap0+and+pk2+%3E+%3Ap1%29%29%29%29+order+by+sortable_with_nulls_2%2C+pk1%2C+pk2+limit+101&p0=a&p1=u&p2=0.2650566289400591

Here's the explain: 49 steps long https://latest.datasette.io/fixtures?sql=explain+select+pk1%2C+pk2%2C+content%2C+sortable%2C+sortable_with_nulls%2C+sortable_with_nulls_2%2C+text+from+sortable+where+%28sortable_with_nulls_2+%3E+%3Ap2+or+%28sortable_with_nulls_2+%3D+%3Ap2+and+%28%28pk1+%3E+%3Ap0%29%0D%0A++or%0D%0A%28pk1+%3D+%3Ap0+and+pk2+%3E+%3Ap1%29%29%29%29+order+by+sortable_with_nulls_2%2C+pk1%2C+pk2+limit+101&p2=0.2650566289400591&p0=a&p1=u

Rewritten with a subselect:

sql select * from ( select pk1, pk2, content, sortable, sortable_with_nulls, sortable_with_nulls_2, text from sortable ) where (sortable_with_nulls_2 > :p2 or (sortable_with_nulls_2 = :p2 and ((pk1 > :p0) or (pk1 = :p0 and pk2 > :p1)))) order by sortable_with_nulls_2, pk1, pk2 limit 101 https://latest.datasette.io/fixtures?sql=select+*+from+(%0D%0A++select+pk1%2C+pk2%2C+content%2C+sortable%2C+sortable_with_nulls%2C+sortable_with_nulls_2%2C+text+from+sortable%0D%0A)%0D%0Awhere+(sortable_with_nulls_2+%3E+%3Ap2+or+(sortable_with_nulls_2+%3D+%3Ap2+and+((pk1+%3E+%3Ap0)%0D%0A++or%0D%0A(pk1+%3D+%3Ap0+and+pk2+%3E+%3Ap1))))+order+by+sortable_with_nulls_2%2C+pk1%2C+pk2+limit+101&p2=0.2650566289400591&p0=a&p1=u

And here's the explain for that - also 49 steps: https://latest.datasette.io/fixtures?sql=explain+select+*+from+%28%0D%0A++select+pk1%2C+pk2%2C+content%2C+sortable%2C+sortable_with_nulls%2C+sortable_with_nulls_2%2C+text+from+sortable%0D%0A%29%0D%0Awhere+%28sortable_with_nulls_2+%3E+%3Ap2+or+%28sortable_with_nulls_2+%3D+%3Ap2+and+%28%28pk1+%3E+%3Ap0%29%0D%0A++or%0D%0A%28pk1+%3D+%3Ap0+and+pk2+%3E+%3Ap1%29%29%29%29+order+by+sortable_with_nulls_2%2C+pk1%2C+pk2+limit+101&p2=0.2650566289400591&p0=a&p1=u

For the SQL underlying this page (the second page in that compound primary key paginated sequence): https://latest.datasette.io/fixtures/compound_three_primary_keys?_next=a%2Cd%2Cv

The explain for the default query: https://latest.datasette.io/fixtures?sql=explain+select%0D%0A++pk1%2C%0D%0A++pk2%2C%0D%0A++pk3%2C%0D%0A++content%0D%0Afrom%0D%0A++compound_three_primary_keys%0D%0Awhere%0D%0A++%28%0D%0A++++%28pk1+%3E+%3Ap0%29%0D%0A++++or+%28%0D%0A++++++pk1+%3D+%3Ap0%0D%0A++++++and+pk2+%3E+%3Ap1%0D%0A++++%29%0D%0A++++or+%28%0D%0A++++++pk1+%3D+%3Ap0%0D%0A++++++and+pk2+%3D+%3Ap1%0D%0A++++++and+pk3+%3E+%3Ap2%0D%0A++++%29%0D%0A++%29%0D%0Aorder+by%0D%0A++pk1%2C%0D%0A++pk2%2C%0D%0A++pk3%0D%0Alimit%0D%0A++101&p0=a&p1=d&p2=v

The explain for that query rewritten as this:

sql explain select * from ( select pk1, pk2, pk3, content from compound_three_primary_keys ) where ( (pk1 > :p0) or ( pk1 = :p0 and pk2 > :p1 ) or ( pk1 = :p0 and pk2 = :p1 and pk3 > :p2 ) ) order by pk1, pk2, pk3 limit 101 https://latest.datasette.io/fixtures?sql=explain+select+*+from+%28select+%0D%0A++pk1%2C%0D%0A++pk2%2C%0D%0A++pk3%2C%0D%0A++content%0D%0Afrom%0D%0A++compound_three_primary_keys%0D%0A%29%0D%0A++where%0D%0A++%28%0D%0A++++%28pk1+%3E+%3Ap0%29%0D%0A++++or+%28%0D%0A++++++pk1+%3D+%3Ap0%0D%0A++++++and+pk2+%3E+%3Ap1%0D%0A++++%29%0D%0A++++or+%28%0D%0A++++++pk1+%3D+%3Ap0%0D%0A++++++and+pk2+%3D+%3Ap1%0D%0A++++++and+pk3+%3E+%3Ap2%0D%0A++++%29%0D%0A++%29%0D%0Aorder+by%0D%0A++pk1%2C%0D%0A++pk2%2C%0D%0A++pk3%0D%0Alimit%0D%0A++101&p0=a&p1=d&p2=v

Both explains have 31 steps and look pretty much identical.

Maybe the correct level of abstraction here is that pagination is something that happens to a SQL query that is defined as SQL and params, without an order by or limit. That's then wrapped in a sub-select and those things are added to it, plus the necessary where clauses depending on the page.

Need to check that the query plan for pagination of a subquery isn't slower than the plan for pagination as it works today.