Repository navigation
PandasCursor drops the sign of negative years and reads years after 9999 as text #1215
Closed as not planned
Description
Part of #1212.
Summary
PandasCursor reads a negative year as a positive one, and reads a column with a year after 9999 as text.
Facts
Measured on master b2d7aa2 (pandas 3.0.6), the same for whole and chunked reads and for managed results:
| Athena text | Type | DataFrame |
|---|---|---|
-0001-01-02 03:04:05.123 |
timestamp |
datetime64[us], Timestamp('1-01-02 03:04:05.123000') |
-0001-01-02 |
date |
datetime64[us], Timestamp('1-01-02 00:00:00') |
+11020-01-02 03:04:05.123 |
timestamp |
str column |
+11020-01-02 |
date |
str column |
With UNLOAD, date values outside 1 to 9999 raise OperationalError instead.
Expected behavior
Following #1212, years outside 1 to 9999 raise DataError, and the column keeps its temporal dtype. pandas keeps the sign when it parses with format="ISO8601"; the fix must be measured for elapsed time and peak RSS on ordinary data.
Activity
laughingman7743 commentedon Oct 17, 2026
Closing: years outside 1 to 9999 are out of scope (#1212). PyAthena does not detect them; what pandas returns or raises for them is pandas' behavior.