sqlite3 Blob objects leave behind weakrefs which are never cleaned up

[takluyver]

Bug report

Bug description:

The sqlite3 Blob class registers a weakref to itself on the Connection it came from. These are added to a list and never cleaned up, so if you have a long-lived connection and call .blobopen() repeatedly, you gradually accumulate ever more dead weakrefs.

It looks like the blob weakrefs were added to match a similar system for weakrefs (removed in #144378), but the code to periodically clear out dead weakrefs was missed.

import gc
import sqlite3

conn = sqlite3.connect(":memory:")
conn.execute("CREATE TABLE foo(a INTEGER PRIMARY KEY, b BLOB)")
conn.execute("INSERT INTO foo VALUES(NULL, ?)", (b'abcdefpokpokpko',))
conn.commit()

# Create & destroy many Blob objects
for _ in range(10_000):
    conn.blobopen("foo", "b", 1)

# Get a reference to the internal list of blob weakrefs - you may need to play with the index here
blob_list = gc.get_referents(conn)[-3]
print(len(blob_list))

CPython versions tested on:

3.14

Operating systems tested on:

Linux

Linked PRs

• gh-144424: Clean up dead weakrefs to sqlite3 Blob objects #144435

[vstinner]

What is the purpose of this list of blob weak references? It seems to be unused. Why not just removing it instead?

[vstinner]

What is the purpose of this list of blob weak references? It seems to be unused. Why not just removing it instead?

Ah! It's used by pysqlite_close_all_blobs() function which is called by _sqlite3.Connection.close(). The function was added by the commit ee47543.

[erlend-aasland]

The list of blob weakrefs is used to close blobs on connection exit. Seems to me we can close this issue?

[takluyver]

This issue is that the weakrefs themselves are not cleaned up, so if you repeatedly access blobs on a long-lived connection, the list of weakrefs grows indefinitely. It's not questioning whether those weakrefs are used. #144435 is my attempt to fix this.