Repository navigation
Description
Describe the bug
Cursor.executemany gives each parameter column one SQL type, and it takes that type from one sample value. Cursor._compute_column_type picks the longest string of the column as the sample. Cursor._map_sql_type declares a str as SQL_VARCHAR when Cursor._is_unicode_string is false, which means when the sample is ASCII. So if the longest string of a column is ASCII, the whole column binds as varchar. The server then converts every value of that column to the code page of the database collation, and each character outside that code page is replaced:
- U+2212 MINUS SIGN becomes
- ŘbecomesR- Thai text becomes
????
No exception is raised, so the data changes silently. A plain execute types each value by itself, and it stores the same values exactly.
The page Handle strings and Unicode says: "The mssql-python driver sends every str parameter as UTF-16LE and declares it as nvarchar by default." It also says: "Because the driver declares string parameters as nvarchar, characters reach the server intact regardless of the database collation." executemany does not do that.
The code is in mssql_python/cursor.py at v1.18.0: _compute_column_type ("For string/binary objects, prefer the longer one"), _map_sql_type and _is_unicode_string.
To reproduce
Set MSSQL_CONNECTION_STRING to any database whose collation uses code page 1252, such as SQL_Latin1_General_CP1_CI_AS.
import os
import platform
import mssql_python
conn = mssql_python.connect(os.environ["MSSQL_CONNECTION_STRING"])
cursor = conn.cursor()
cursor.execute("CREATE TABLE #t (id int, txt nvarchar(100))")
rows = [
(1, "plain ASCII text, the longest value in the column"),
(2, "a \u2212 b"), # U+2212 MINUS SIGN
(3, "\u0158\u00edm"), # Czech
(4, "\u0e23\u0e49\u0e32\u0e19"), # Thai
]
cursor.executemany("INSERT INTO #t (id, txt) VALUES (?, ?)", rows)
for row in rows[1:]:
cursor.execute("INSERT INTO #t (id, txt) VALUES (?, ?)", (row[0] + 10, row[1]))
cursor.execute("SELECT id, txt FROM #t ORDER BY id")
stored = {r[0]: r[1] for r in cursor.fetchall()}
for row_id, text in rows[1:]:
print(f"sent {text!r}: executemany stored {stored[row_id]!r}, execute stored {stored[row_id + 10]!r}")
cursor.execute("SELECT CAST(SERVERPROPERTY('ProductVersion') AS nvarchar(32)), CAST(DATABASEPROPERTYEX(DB_NAME(), 'Collation') AS nvarchar(128))")
print("mssql-python", mssql_python.__version__, "| Python", platform.python_version(), "|", platform.platform(), "| server", tuple(cursor.fetchone()))
conn.close()Output:
sent 'a − b': executemany stored 'a - b', execute stored 'a − b'
sent 'Řím': executemany stored 'Rím', execute stored 'Řím'
sent 'ร้าน': executemany stored '????', execute stored 'ร้าน'
Expected behavior
executemany stores every value as sent, as execute does and as the documentation describes. A str column binds as SQL_WVARCHAR, at least when any value of the column is not ASCII.
Further technical details
Python version: 3.14.9
mssql-python version: 1.18.0, and 1.17.0 behaves the same
SQL Server version: Azure SQL Database, ProductVersion 12.0.2000.8, collation SQL_Latin1_General_CP1_CI_AS
Operating system: macOS 27.2.1, arm64
Additional context
- My workaround is
Cursor.setinputsizesbefore eachexecutemany, withSQL_WVARCHARfor every text column. - Separate Type Contructors for (var)char and n(var)char #174 asks for type constructors for
varcharandnvarchar. That would also give a caller a way to choose. - setencoding(ctype=SQL_CHAR) is silently ignored when binding text parameters #825 shows that
Connection.setencodingdoes not change how a text parameter binds, so it is no workaround.
Activity
github-actions commentedon Nov 18, 2026
Hi J Rob Gant (@rgant), thank you for opening this issue!
Our team will review it shortly. We aim to triage all new issues within 24-48 hours and get back to you.
If you have additional information to share, please feel free to update the issue.
Thank you for your patience!