Hello,

in 15569ea9 we downgraded pandas version to 0.24.2 which is the last version which works for python2. In ca5ba490 we further downgraded the pandas version to the old version 0.19.2, because ERP5 raised an exception with a newer version.

This exception can easily be solved and I think instead of downgrading the pandas version, we should instead fix ERP5. This is done with levin.zimmermann/erp5@11b22568 (MR is pending).

Sorry @tomo that I didn't saw 15569ea9 (comment 174707) earlier!

Some background context:

I think pandas 0.19.2 has a bug, which can be reproduced with:

import pandas as pd
from datetime import datetime

s = pd.Series(
	[
		datetime(2020, 1, 1),
		datetime(2020, 1, 2),
		datetime(2020, 1, 3),
	]
)

print s.diff().dt.seconds

We run into this bug in the WWM/wind project (see https://lab.nexedi.com/nexedi/woelfel-web-monitoring/commit/48b4034b745badc48e052f60c2971f6a76ef8631).

I think the most straightforward way to fix this bug is to simply use a newer pandas version (where this bug can't be reproduced, at least I tried it with 0.20.x for now).

• SlapOS integration tests (pending): https://erp5js.nexedi.net/#/test_result_module/20230215-C14A0965
• ERP5 unit tests (with old pandas): https://erp5js.nexedi.net/#/test_result_module/20230215-107194B0D
• ERP5 coding style tests (with old pandas): https://erp5js.nexedi.net/#/test_result_module/20230215-BEC1A1A1
• ERP5 unit tests (with new pandas): https://erp5js.nexedi.net/#/test_result_module/20230215-82915E6

Maybe we should also run this change against Wendelin unit tests?

Is there any other reason why we prefer 0.19.x over 0.24.x?

Best, Levin

/cc @klaus @tomo @tatuya @jerome