Tested upgrade record — pandas 2 → 3, change 1

pandas 3: df["col"][row] = value no longer changes the DataFrame

Under pandas 2.3.3 this line changed the DataFrame and printed a FutureWarning. Under pandas 3.1.0 the DataFrame stays as it was and you get a ChainedAssignmentError warning. The code runs to the end either way, so the first sign can be wrong numbers further down.

Where it ran

• Run on 26 October 2026.
• pandas 2.3.3 and pandas 3.1.0, each in its own virtual environment, installed from PyPI that day.
• numpy 2.5.4, Python 3.13.17, Linux x86_64, in both.

I've only run it on Linux. This change comes from how pandas handles copies, so I'd expect the same on macOS and Windows, but I haven't tested that.

The script

import platform, warnings
import numpy as np
import pandas as pd

print("pandas", pd.__version__, "| numpy", np.__version__,
      "| python", platform.python_version(), "|", platform.system(), platform.machine())

def attempt(label, fn):
    df = pd.DataFrame({"price": [10, 20, 30], "qty": [1, 2, 3]})
    with warnings.catch_warnings(record=True) as caught:
        warnings.simplefilter("always")
        fn(df)
    print(f"\n[{label}] DataFrame changed: {df.loc[0, 'price'] == 99}")
    for w in caught:
        print(f"  warning: {w.category.__name__}: {str(w.message).strip().splitlines()[0]}")
    if not caught:
        print("  warning: none")

def a(df): df["price"][0] = 99              # column, then row
def b(df): df[df["qty"] > 0]["price"] = 99  # filter, then column
def c(df): df.loc[0, "price"] = 99          # one .loc call

attempt('df["price"][0] = 99', a)
attempt('df[df["qty"] > 0]["price"] = 99', b)
attempt('df.loc[0, "price"] = 99', c)

Output under pandas 2.3.3

pandas 2.3.3 | numpy 2.5.4 | python 3.13.17 | Linux x86_64

[df["price"][0] = 99] DataFrame changed: True
  warning: FutureWarning: ChainedAssignmentError: behaviour will change in pandas 3.0!

[df[df["qty"] > 0]["price"] = 99] DataFrame changed: False
  warning: none

[df.loc[0, "price"] = 99] DataFrame changed: True
  warning: none

Output under pandas 3.1.0

pandas 3.1.0 | numpy 2.5.4 | python 3.13.17 | Linux x86_64

[df["price"][0] = 99] DataFrame changed: False
  warning: ChainedAssignmentError: A value is being set on a copy of a DataFrame or Series through chained assignment.

[df[df["qty"] > 0]["price"] = 99] DataFrame changed: False
  warning: ChainedAssignmentError: A value is being set on a copy of a DataFrame or Series through chained assignment.

[df.loc[0, "price"] = 99] DataFrame changed: True
  warning: none

What the three lines show

• Column, then row (df["price"][0] = 99). This is the one that changed. It worked in 2.3.3 and does nothing to df in 3.1.0.
• Filter, then column (df[mask]["price"] = 99). It didn't change the DataFrame in either version. The difference is that 3.1.0 now warns about it, and 2.3.3 printed nothing in my run, not even a SettingWithCopyWarning. I haven't worked out why the older warning didn't appear here.
• One .loc call (df.loc[0, "price"] = 99) changed the DataFrame in both versions with no warning.

The full warning text

Run as a single statement at the top level with python -W always, the two versions print this. Under 2.3.3:

FutureWarning: ChainedAssignmentError: behaviour will change in pandas 3.0!
You are setting values through chained assignment. Currently this works in certain cases, but when using Copy-on-Write (which will become the default behaviour in pandas 3.0) this will never work to update the original DataFrame or Series, because the intermediate object on which we are setting values will behave as a copy.

Under 3.1.0:

ChainedAssignmentError: A value is being set on a copy of a DataFrame or Series through chained assignment.
Such chained assignment never works to update the original DataFrame or Series, because the intermediate object on which we are setting values always behaves as a copy (due to Copy-on-Write).

Try using '.loc[row_indexer, col_indexer] = value' instead, to perform the assignment in a single step.

In the same run the DataFrame value printed afterwards was 99 under 2.3.3 and 10 under 3.1.0.

What pandas says about it

The pandas migration guide ("Chained Assignment", read 26 October 2026 in the 3.1.0 documentation) says "Copy-on-Write is the default and only mode in pandas 3.0" and that "chained assignment will consistently never work and raise a ChainedAssignmentError warning with CoW enabled". Its suggested replacement is a single loc call, as in the third line above.

The other thing that broke in the same first test, text columns no longer having dtype object, gets its own page next.

This is the first page of a record of what changes between pandas 2 and pandas 3, one reproduced change per page. Why I'm building it, and the weak points of the plan, are in wake log 099. Nothing in it is for sale yet. Each page stays free.