sqlite3: Blob crashes when using a negative-step slice

What happened?

sqlite3.Blob crashes when you use a negative step in a slice — both for reading and writing.
The sequence operations docs say s[i:j:k] with negative k is valid, and bytearray handles it fine.

import sqlite3

con = sqlite3.connect(":memory:")
con.execute("CREATE TABLE t(b BLOB)")
data = b"this blob data string is exactly fifty bytes long!"
con.execute("INSERT INTO t(b) VALUES (?)", (data,))

# bytearray works
ba = bytearray(data)
ba[9:0:-2] = b"12345"
print(ba)
# b't5i4 3l2b1data string is exactly fifty bytes long!'

# Blob crashes
blob = con.blobopen("t", "b", 1)
blob[9:0:-2] = b"12345"
blob.close()

>>> blob[9:0:-2] = b"12345"
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
SystemError: Negative size passed to PyBytes_FromStringAndSize

Reading also crashes: blob[9:0:-2] raises the same error.

On Python 3.11.6 and 3.14.5 the error is SystemError: Negative size passed to PyBytes_FromStringAndSize.
On the current main branch read_multiple was changed to use PyBytesWriter_Create, so the message is now ValueError: size must be >= 0 — but the root cause is the same.

CPython versions tested on:

3.11.6, 3.14.5, main(629da5c914b4407e01c1dc06cbcbd8dce825fef3)

Operating systems tested on:

macOS - 26.3.1 (a)(25D771280a)

Related

• gh-142787 / gh-142824 — different bug (assertion failure on empty slice, already fixed)
• https://docs.python.org/3/library/stdtypes.html#typesseq-common (s[i:j:k], negative k)

I'd like to work on a fix for this. I'll submit a PR.

Linked PRs

• gh-150450
• gh-155773

Commencez par l’implémentation du slicing de sqlite3.Blob et reproduisez les cas signalés avec un pas négatif, aussi bien en lecture qu’en écriture, en utilisant l’exemple de l’issue. Le travail est considéré comme terminé lorsque les slices avec un pas négatif se comportent comme les opérations de séquence documentées sans lever d’erreur liée à la taille ; les PRs associés gh-150450 et gh-155773 montrent que ce travail est déjà en cours.