Thanks for all the suggestions! I hope the documentation related comments will come out OK.

The last commit changes this:

Into this:

Does that change the behavior when the user does not opt-in to using offsets?

OK so after thinking about it once more, I realized that what I implemented in the demo at #28463 may be too complex - there is no need to show an order of magnitude of the data at the offset text, if we already have the ENG_PREFIXES that can shorten the strings of the ticks. Hence I managed to further simplify the logic in EngFormatter such that self.orderOfMagnitude is always 0, and the offset text only shows offset prefixed with +, and not showing the oom of the data after offset subtraction.

The only debatable drawback of this behavior, is that a 0 tick will always appear without a unit prefix. It sort of makes sense because 0 is 0 no matter what is the data's general oom. I got convinced after playing a bit, that making sure that all ticks have the same unit prefix is making the class so much more complex that it doesn't worth it. The current behavior is consistent and well tested.

The last force push simplifies as described above, and improves the test I added for this feature.

OK so after testing many more edge cases, I revised the behavior of this I changed my opinion once more. The tests now are much more extensive and I finally managed to reach the behavior I was aiming for.

The git history got too messy and I had a few reverts locally so I forced push.

Here are a few screenshots:

+/-0.5e6

+/- 3.1e3

+/- 2.1e5

Also it is automatically updated when zooming in

Offset enabled in principal, but data doesn't qualify to get an offset:

(Also when started like that, offset is computed automatically and presented if needed)

I almost lost in the rebase the fixes to the comments mentioned above. They are fixed now. Hopefully docs will build fine now 🙏.

I'm having some trouble building the documentation locally, so I have to rely upon CI. Last force push fixed a few small issues with it.

Thank you @QuLogic for the detailed review! I don't have time now to respond to all of your suggestions, so I started with the easier parts, documentation and comments.

OK so last night I struggled a bit running the tests locally, so I hoped my changes would just work. Now I finally checked locally before pushing. I also responded to all other comments, and the topics left to discuss are the local variables helping to debug with pytest --showlocals, and meaning of oom_ variables in the test_engformatter_offset_oom, and:

`matplotlib.ticker.EngFormatter`

v.s

:class:`matplotlib.ticker.EngFormatter`

In the docs (not in the titles of documents - that's understandable).

I don't understand the circle CI errors... They don't seem related to my changes. Same with the errors on Windows and MacOS on Azure.

I don't understand the circle CI errors... They don't seem related to my changes. Same with the errors on Windows and MacOS on Azure.

It appears you have rebased against a very old commit instead of the current main.

Rebased.

Less errors now, but I see now:

Path does not exist: /Users/runner/work/1/s/result_images

The mypy CI error doesn't seem related to the PR...

Mypy passes locally, CI failure was fixed by #29014, would need a rebase to resolve on CI, butr since it passes locally and the failure is understood, I'm satisfied.

Thank you for your corporation :).