I continue to be surprised every time this happens to me! Facts show I must grudgingly accept it.

Even if changes were all in summer I still might not see them until mid-semester when I reran a particular bit of code the one time I needed it. Long-term stability is a bigger help than specific timing.

It’s vexing w.r.t. devs wanting to move things forward vs. users’ intermittency.

Adding my own ux reaction/fears here, the challenge in tracking this down is that there are about 20 versions of the underlying library in between, it’s a low-level change affecting a core operation, and, practically, it’s unknown if I need to spend 10 min or a day to figure it out.

Example: this worked in xarray 2023.4.2 with no warnings but in 2025.06.0: “IndexError: arrays used as indices must be of integer (or boolean) type.” No idea what changed.

ix = {'track': [slice(1, 2, None), slice(4, 5, None)], 'cell': [np.int64(1), np.int64(4), np.int64(49),]}
ds = ds[ix]

My re-teach cycle is about 2 yrs., so a bit longer than that would be ideal. Feels like I see some new decision/persnicketiness on that cycle with some regularity. Really breaks flow of lecture prep when you have to switch to debugging coder brain to get the red (warnings) out.

Thomas Kinkade bad CGI front exterior

Nope!

Yeah, that’s more than enough. I think a fair share is 3x the amount of your first and student authored papers each year. I bet you’ve exceeded that?

Yeah, that’s mostly my experience.

The tip of the pyramid seems linked to language from empirical-statistical modeling where disciplines can’t draw on ~universally agreed, more general, physical-mathematical theory. But we can as meteorologists. Those are our hard, but most compelling, studies.

This is @persistent.info’s post with some more technical details – and some further twists like using an LLM to a control an old computer!

blog.persistent.info/2025/07/infi...