gh-152817: Prevent deletion of sqlite3 cursor.row_factory attr, missed from: gh-149738

[stestagg]

gh-149738 prevents deletion of row_factory and text_factory attributes on a sqlite3 connection, this PR applies that fix to cursor objects too, to avoid a similar segfault:

import sqlite3

cur = sqlite3.connect(":memory:").cursor()
del cur.row_factory
cur.execute("select 1").fetchone()

I've piggy-backed off the existing docs change and blurb as that seems to cover this already

• Issue: segfault after deleting row_factory attr of sqlite3 cursor #152817

[stestagg]

@sepehr-rs - You were the author of the original, are you able to take a look? Thanks in advance!

[sepehr-rs]

Hi @stestagg, thank you for reaching out to me on this!
I think your change is good, but I believe you should also add a versionchanged:: next note under the Cursor.row_factory attribute docs, mirroring what was added for Connection.row_factory/text_factory. Besides that, +1 from me.

[stestagg]

Hi @stestagg, thank you for reaching out to me on this! I think your change is good, but I believe you should also add a versionchanged:: next note under the Cursor.row_factory attribute docs, mirroring what was added for Connection.row_factory/text_factory. Besides that, +1 from me.

As I understand it, gh-149738 already altered the doc for the Cursor.row_factory attr (as well as for the Connection.row_factory), just not the code:

cpython/Doc/library/sqlite3.rst

Lines 1718 to 1719 in f3bf8ab

	.. versionchanged:: next
	Deleting the ``row_factory`` attribute is no longer allowed.

It's likely I've misunderstood/missed something here!

[sepehr-rs]

As I understand it, gh-149738 already altered the doc for the Cursor.row_factory attr (as well as for the Connection.row_factory), just not the code:

Oh, you're right. I must have forgotten about that. This is a real gap, thanks for checking this! :)

[stestagg]

There's a related issue that I uncovered above, if you create a Connection or Cursor without calling __init__ on it (why would you do this?) then row_factory/text_factory fields are never initialized, so then you get a segfault anyway.

To me, there's a few options:

1. Make the users of row_factory/text_factory handle NULL like None, it's a cheap check, then having a NULL in this fields doesn't matter, we can remove the 'not allowed to del the attribute' code which makes things a bit simpler.
2. Move the current field initialisation into tp_new (currently we don't have one), so that we never have an object in python space with row_factory=NULL, we'd have to keep the delattr guards, but that's fine, why would you delattr this anyway
3. Just ignore the above issue, get this PR in, and raise a new issue to fix the new/init problem.

I'm not sure of the best approach tbh.

[sobolevn]

I propose to split this into 2 PRs: one with the deletion, one with the assignement. Let's keep this PR focused on one thing :)

[stestagg]

@sobolevn I propose to split this into 2 PRs: one with the deletion, one with the assignement. Let's keep this PR focused on one thing :)

fair point, reverted, I'll raise the other part separately