Currently, pysqlite_statement_create() approx. looks like this:

1. some sanity checks (type, sql lenght, etc.)
2. allocate (PyObject_GC_New)
3. initialise members
4. determine if statement is a DML statement
5. create the statement (sqlite3_prepare_v2)
6. PyObject_GC_Track
7. check statement return value
8. more sanity checking
9. done!


Suggesting to refactor as this:
1. all sanity checks => early exit on failure
2. create the statement and validate return value
3. determine if statement is a DML statement => no need to do this if statement creation failed
4. allocate
5. initialise members
5. return


This will avoid unneeded allocations/GC tracking, it will avoid unneeded statement creation, and it will be more readable/maintainable.

>  will avoid unneeded allocations/GC tracking,

I am not sure what you mean, in the happy path you still need to GC track and allocate.

> I am not sure what you mean, in the happy path you still need to GC track and allocate.

Currently, we allocate the object, then try to create the statement using the SQLite API. If we create the statement first, we can do the sanity check on the return value and just return NULL; then we can allocate the Py object and initialise the members.

... to expand: so currently, if statement creation fails, we must deallocate the PyObject.

Ah, I see I formulated myself a bit unclear:
Yes, we need to allocate/track every time. I just propose to do so as late as possible, in order to avoid allocating a PyObject before we know if it is possible to actually create the statement.

> Yes, we need to allocate/track every time. I just propose to do so as late as possible, in order to avoid allocating a PyObject before we know if it is possible to actually create the statement.

That describes much better the intent :)

New changeset 1c02655fb08043b3027748ca1179c416c21a4277 by Erlend Egeberg Aasland in branch 'main':
bpo-44329: Refactor sqlite3 statement creation (GH-26566)
https://github.com/python/cpython/commit/1c02655fb08043b3027748ca1179c416c21a4277