TypeError when trying to unpickle results with equal column names

[tweakimp]

Environment

• Python: 3.7.4
• pyodbc: 4.0.27
• OS: win10 64bit
• DB: SQLITE and MSSQL
• driver: sqliteodbc_w64.exe from here

Issue

When trying to unpickle a result from a query with some equal column names, a TypeError is raised: TypeError: cannot create 'pyodbc.Row' instances

Code to reproduce

I noticed that at work where we have a MS SQL Server, but it is possible to reproduce the error with sqlite.

import sqlite3
import pickle
import pyodbc

# create a database
conn = sqlite3.connect("db.db")
c = conn.cursor()
# create table and fill it
c.execute("""CREATE TABLE IF NOT EXISTS test (value1 int, value2 int,value3 int)""")
c.execute("INSERT INTO test VALUES (1,1,1),(2,2,2),(3,3,3)")
conn.commit()
conn.close()

# create pyodbc connection
cnxn = pyodbc.connect("Driver=SQLite3 ODBC Driver;Database=db.db")

cursor = cnxn.cursor()

# select unique column names
cursor.execute("SELECT value1, value2, value3 from test")
result = cursor.fetchall()

# we get the actual values
print(f"actual result1 {result}")

# pickle the results
with open("result_ok.pkl", "wb") as file:
    pickle.dump(result, file, protocol=pickle.HIGHEST_PROTOCOL)

# unpickle them
with open("result_ok.pkl", "rb") as file:
    unpickled = pickle.load(file)

# we can load the pickled values
print(f"unpickled result1 {unpickled}")

# same thing again, but select a column twice
cursor = cnxn.cursor()
cursor.execute("SELECT value1, value2, value3, value1 from test")
result = cursor.fetchall()

# we get the actual values again, everything seems fine
print(f"actual result2 {result}")
# pickle the results
with open("result_error.pkl", "wb") as file:
    pickle.dump(result, file, protocol=pickle.HIGHEST_PROTOCOL)

# unpickle them
with open("result_error.pkl", "rb") as file:
    unpickled = pickle.load(file)  # code breaks here

print(f"unpickled result2 {unpickled}")

[v-makouz]

I will try to reproduce this, just to confirm, this happens on SQL Server as well, with "ODBC Driver 17 for SQL Server"?

[tweakimp]

Yes.

[v-chojas]

Possibly related: https://github.com/mkleehammer/pyodbc/issues/446

[v-makouz]

@tweakimp I am able to repro the issue, and I will into it over the next few days

[tweakimp]

Great, thank you for your time

[v-makouz]

I've located the code that throws that error, but I still don't quite understand the logic behind it, in the file row.cpp around line 110, there's this:

    if (PyDict_Size(map) != cols || PyTuple_GET_SIZE(args) - 2 != cols)
        return 0;

Which will cause the early return if you select different number of columns, because then PyDict_Size(map) != cols will be true. I tried running the repro with that check removed and I got correct looking results back after unpickling (with 4 columns), but that check was put there for a reason, I'm just not sure what it is @mkleehammer?

[piskvorky]

Also bitten by this on SQL Server 2014 + msodbcsql17; @v-makouz did you manage to find a work around?

[v-makouz]

@piskvorky I'm afraid I don't really know anything new, other then if the above code is changed to

    if (PyTuple_GET_SIZE(args) - 2 != cols)
        return 0;

The above repro will work correctly, but I'm not sure if that'll cause issues for anything else. I'm just not sure why that check is there.