Use -> operator to implement KeyTransform on SQLite 3.38+

[Sage Abdullah]

SQLite added support for -> and ->> operators as of version 3.38.0. From the ​docs:

Both the -> and ->> operators select the same subcomponent of the JSON to their left. The difference is that -> always returns a JSON representation of that subcomponent and the ->> operator always returns an SQL representation of that subcomponent. Thus, these operators are subtly different from a two-argument json_extract() function call. A call to json_extract() with two arguments will return a JSON representation of the subcomponent if and only if the subcomponent is a JSON array or object, and will return an SQL representation of the subcomponent if the subcomponent is a JSON null, string, or numeric value.

It would be nice to have this as a cleanup/optimization, e.g. for the workaround introduced in ​this PR. Still, we need to check whether the SQLite bundled in the user's Python installation supports it though, so we can't just replace the current implementation.

[Mariusz Felisiak]

Thanks for the ticket. Tentatively accepted, it can be tricky to implement.

[bcail]

For what it's worth, the SQLite tests pass on SQLite 3.39.4 if you replace JSON_EXTRACT with ->>, like this:

diff --git a/django/db/models/fields/json.py b/django/db/models/fields/json.py
index 7296fe42bc..accc6ae61a 100644
--- a/django/db/models/fields/json.py
+++ b/django/db/models/fields/json.py
@@ -361,7 +361,7 @@ class KeyTransform(Transform):
         )
         return (
             "(CASE WHEN JSON_TYPE(%s, %%s) IN (%s) "
-            "THEN JSON_TYPE(%s, %%s) ELSE JSON_EXTRACT(%s, %%s) END)"
+            "THEN JSON_TYPE(%s, %%s) ELSE (%s ->> %%s) END)"
         ) % (lhs, datatype_values, lhs, lhs), (tuple(params) + (json_path,)) * 3
 

[rajdesai24]

I am quite new to this topic but based on what I can understand how about we put a simple version check and return the value accordingly
something like the following which checks the version and returns the value accordingly.

        if connection.sqlite_version >= '3.38':
                    return (
                    "(CASE WHEN JSON_TYPE(%s, %%s) IN (%s) "
                    "THEN JSON_TYPE(%s, %%s) ELSE (%s ->> %%s) END)"
                    ) % (lhs, datatype_values, lhs, lhs), (tuple(params) + (json_path,)) * 3
 
        else:
            return (
            "(CASE WHEN JSON_TYPE(%s, %%s) IN (%s) "
            "THEN JSON_TYPE(%s, %%s) ELSE JSON_EXTRACT(%s, %%s) END)"
            ) % (lhs, datatype_values, lhs, lhs), (tuple(params) + (json_path,)) * 3

This might make it redundant but please let me know of any suggestions. Till then I will dive deeper

[Paolo Melchiorre]

This change would be an improvement. +1

[bcail]

@rajdesai24, are you planning to open a PR for this?

[Claude Paroz]

13 months after an assignee last commented, I guess we can assume it is open to anyone willing to work on it.

[Anže Pečar]

I have found this ticket during DjangoCon Europe sprints and opened a PR for it: ​https://github.com/django/django/pull/18241

I didn't add the version check discussed above because Python 3.10.9 updated SQLite to 3.39 so the operator should be supported on all Python versions that Django currently supports.

[Paolo Melchiorre]

As we discussed during the DjangoCon Europe 2024 sprints, we can't be sure of the SQLite version in some cases, so a version check can be useful.

[Anže Pečar]

I have added the version check to my PR, thank you Paolo, for the quick PR review!

[Paolo Melchiorre]

The PR looks better now. Thanks for update it.