Update `test_zoneinfo` data to 2025b

[StanFromIreland]

Feature or enhancement

Currently, the extract is compiled from 2020a. Updating the extract results in ~94 errors (without tzdata installed). I therefore propose updating it and our tests.

Linked PRs

• gh-139546: Update test_zoneinfo data to 2025c #139547

[zware]

I'm not clear on the issue here. What and where are the errors? This is just test data, so does it really need to be up to date? Or is it that (correctly) updating the test data results in errors of the library that need to be fixed?

[StanFromIreland]

Updating the test data results in errors as we make incorrect assumptions in our test data (you can see them in the PR), I think we should periodically regenerate to ensure compatibility with the data, and secondly, to avoid the spurious errors for those testing with tzdata installed (for a long time, a tzdata installed test failed on Windows but no one noticed, possibly because it would be in the midst of the spurious errors).

cc @pganssle

[vstinner]

test_zoneinfo.py uses zoneinfo_data.json. Tests are written for this test data (2020a). I'm not convinced that it's useful to update the test data.

to avoid the spurious errors for those testing with tzdata installed (for a long time, a tzdata installed test failed on Windows but no one noticed, possibly because it would be in the midst of the spurious errors).

Do you mean that test_zoneinfo.py doesn't only use zoneinfo_data.json?

I'm sorry, I don't understand well which problem is being solved here :-(

[StanFromIreland]

Do you mean that test_zoneinfo.py doesn't only use zoneinfo_data.json?

Yes, TZDataTests only run if the tzdata package is available (i.e. not the extract).

The assumptions in the tests are for the 2020a data, which as a result breaks the TZDataTests if run with a newer package installed.

[vstinner]

I ran the following commands on Python 3.15:

./python -m venv env
env/bin/python -m pip install tzdata
env/bin/python -m test -v test_zoneinfo

Many tests of test_zoneinfo are failing.

If I uninstall tzdata, test_zoneinfo does pass, but multiple tests are skipped.

Ok, now I understand the issue.

It's a bit unfortunate that test_zoneinfo depends on a 3rd party package (tzdata) and requires a specific package version (2020a?).

[zware]

So instead of updating the test data which we will never keep up to date, we should fix the tests to avoid reliance on the exact data in tzdata.

[pganssle]

So instead of updating the test data which we will never keep up to date, we should fix the tests to avoid reliance on the exact data in tzdata.

It is not that simple, really. You can't count on much of anything in tzdata staying exactly how it was, because we sometimes get new historical information or bugfixes that changes the data for the past, and this is doubly true for the edge cases, which tend to be in old zones or weird in other ways subject to change. Some of this is handled by hand crafting fake zones that exhibit the weird behavior, but we also want to test against real data.

You also don't want the data to stay stale forever because then the tests become increasingly unlike the actual deployed TZif files (there have been changes where not doing so might cause problems).

The right thing to do is probably to pin an exact version of tzdata so that we know exactly what data we are testing against but also update that pin (and the fixed subset we vendor in the json file) from time to time.

[pganssle]

It's a bit unfortunate that test_zoneinfo depends on a 3rd party package (tzdata) and requires a specific package version (2020a?).

tzdata isn't a third party package, it is effectively part of CPython, it's just distributed on pypi for reasons of cadence (see PEP 615)