Closed
Description
Bug report
blobopen
internally uses int
to hold the requested rowid ([1], [2]), but SQLite rowids are actually 64-bit integers:
int sqlite3_blob_open(
sqlite3*,
const char *zDb,
const char *zTable,
const char *zColumn,
sqlite3_int64 iRow, // <-- the rowid parameter
int flags,
sqlite3_blob **ppBlob
);
This makes an attempt to open a blob with large rowid raise an OverflowError
when Python is compiled with 32-bit int
, which is the case on Windows even when compiling as 64-bit application.
This might seem like an edge case, but any INTEGER PRIMARY KEY
in a rowid table aliases rowid, which means any application that uses non-autoincrement primary keys (e.g. timestamps, checksums) is likely to hit this very trivially -- I know I did on basically the first insert. You don't need to have more than 2**32 rows or anything like that for this to happen.
100% reproducible with:
import sqlite3
con = sqlite3.connect(':memory:')
rowid = 2**32
con.execute("create table t(t blob)")
con.execute("insert into t(rowid, t) values (?, zeroblob(1))", (rowid,))
con.blobopen('t', 't', rowid)
Expected: nothing (i.e. successful call)
Instead:
Traceback (most recent call last):
File "E:\Temp\blob.py", line 10, in <module>
con.blobopen('t', 't', rowid)
OverflowError: Python int too large to convert to C int
Your environment
- CPython versions tested on: 3.11.0 (main, Oct 24 2022, 18:26:48) [MSC v.1933 64 bit (AMD64)], 3.12.0a3+ (heads/main-dirty:cb60b6131b, Dec 20 2022, 14:37:41) [MSC v.1934 64 bit (AMD64)]
- Operating system and architecture: Windows x64
Linked PRs
Metadata
Metadata
Assignees
Projects
Status
Done