Don't require max_length on CharField on SQLite backend

[Curtis Maloney]

Some trivial testing suggests that SQLite doesn't require a length on VARCHAR / TEXT fields, just like Postgres.

[Curtis Maloney]

It would vastly simplify my own package testing if I didn't have to spin up a PG instance for models written assuming I can omit max_length, and could instead use a disposable SQLite DB.

[Simon Charette]

SQLite doesn't enforce any database constraints when a max length is specified (see #21471 and ​this recent forum discussion on the subject) so supporting max_length=None on SQLite is trivial (we already implicitly do).

For anyone interested in picking this up the patch needs three things

1. The SQLite backend's supports_unlimited_charfield feature flag must be set to true.
2. The CharFielddocumentation must be adjusted to mention ​the feature is available of SQLite as well with a ::versionchanged.
3. A release note must be added.

[Jae Hyuck Sa]

I’ve always been interested in Database/ORM, so I’d like to handle this ticket myself. Thank you, Simon Charette, for providing such detailed guidelines.

[Simon Charette]

Patch is looking great.

For the record #34887 initially rejected this feature but I think it warrants being revisited. The argument provided by Mariusz at the time was

We accepted #14094, because CharField and TextField use different datatypes on PostgreSQL.

but in the end both varchar and text ​are aliases for the internal Postgres varlena type so I don't think the argument stands. There are reason why we'd want to allow CharField(max_length=None) on all backends that allow it which #14094 does a good job at covering.

Moreover the addition of this feature could allow us to eventually automatically add a check constraint for max_length on SQLite by allowing users to disable it entirely by setting the value to None.

[Jae Hyuck Sa]

Replying to Simon Charette:

Patch is looking great.

For the record #34887 initially rejected this feature but I think it warrants being revisited. The argument provided by Mariusz at the time was

We accepted #14094, because CharField and TextField use different datatypes on PostgreSQL.

but in the end both varchar and text ​are aliases for the internal Postgres varlena type so I don't think the argument stands. There are reason why we'd want to allow CharField(max_length=None) on all backends that allow it which #14094 does a good job at covering.

Moreover the addition of this feature could allow us to eventually automatically add a check constraint for max_length on SQLite by allowing users to disable it entirely by setting the value to None.

Thank you for the detailed explanation! :)

[Mariusz Felisiak]

If we want to accept this, #35759 should be closed as a duplicate of #34887, and eventually #34887 reopened.

[Sarah Boyce]

Duplicate of #34887