AFAICS, the docs for sqlite3_column_name() says that if it returns NULL, we are out of memory, thus we should call PyErr_NoMemory() and bail.

FYI, there are the calls to sqlite3_column_name() in Modules/_sqlite//cursor.c

Ref:
- http://sqlite.org/c3ref/column_name.html
- http://sqlite.org/c3ref/free.html

Well, it returns NULL in case of out of memory, but is it the only cause? Can NULL be returned for other reasons?

> Well, it returns NULL in case of out of memory, but is it the only cause? Can NULL be returned for other reasons?

According to the SQLite docs, no. Looking at the source code, we see that it also returns NULL if the second parameter (column index) is out of range, but we already check that. AFAICS, the only reason for a NULL in our case is OOM.

It also returns NULL if the column name was not set (is it even possible?).

That's inside sqlite3_value_text() and friends, then? Let's investigate further before concluding.

Investigate where column names are initialized and whether is it possible to get them NULLs. The array of column names is filled with NULLs when initialized and later it is filled with references to strings. I am wondering whether there is a gap between this which lefts some column names as NULLs.

"SELECT 1" yields "1" as column name.
"SELECT <...> AS N" yields "N" as column name.
"SELECT NULL" yields "NULL" as column name.

We can't set the column name via the API, so that's pretty much it (implicit name or explicit through the AS keyword). As long as sqlite3_prepare_v2() returned SQLITE_OK, it seems that SQLite assures we've got a valid column name.

As far as I can sqlite3_column_name() returns NULL if:
1) The first parameter (stmt) is NULL
2) The second parameter is out of bounds
3) Memory error
4) No column name is set (SQLite bug)

We've got 1) and 2) covered. SQLite has a pretty good code coverage, so I'd say 4) is unlikely.

What do you think?

Please go ahead!

Will do. Thanks for pushing the investigation.

After discussing the matter briefly on the SQLite forum, I'm no longer 100% certain about this. There seems to be uncertainty about which other conditions can produce NULL, although memory error seems to be the most probable. I consider closing both this issue and the PR and just leave the code as it is.

Could you please give a link to the discussion?

Sure! Here it is: https://sqlite.org/forum/forumpost/574c6bc66c

Larry Brasfield's comment https://sqlite.org/forum/forumpost/6430fc589d?t=h aligns with https://bugs.python.org/issue43251#msg387428

I'll think twice before posting there again, though.

I believe we can proceed with this as planned. Serhiy, do you have additional comments or change requests?

New changeset 2183d06bc8a481098d62a4ebed8d6982b3d1602a by Erlend Egeberg Aasland in branch 'master':
bpo-43251: sqlite3_column_name() failures now raise MemoryError (GH-24609)
https://github.com/python/cpython/commit/2183d06bc8a481098d62a4ebed8d6982b3d1602a