Antony Lee

There's two separate issues at hand here: color and flags (which are really ligatures). For the DE->flag problem, this arises specifically because emoji fonts rely on ligatures to do the...

I really don't think asking users to manually copy NullLocators or FuncFormatters makes sense from a user-friendliness PoV.

I think (consistently with above) the correct solution would be to have locators not know about the axis except at call time (`locator(axis)` instead of `locator()`), but that runs into...

@timhoffm Perhaps you can have a look at #28429? It's not the most elegant solution, but it effectively works around the existing design issues while keeping backcompat.

Sure, the proposed behavior seems reasonable (in fact, likely better). I guess the current strategy in Image.save (stash the params into encoderinfo and later let the plugin's _save_all parse that)...

I don't think we can really change this long-standing behavior due to backcompat concerns, but I agree that we should document this better (perhaps the error message "latex was not...

In general, I like the idea. Would it be possible to make this work with arbitrary underlying scales (e.g. log); e.g. ``` ax.set_xscale("broken", breaks=..., base_scale="log") ``` (possibly base_kwargs too for...

Indeed, the idea would be to make your scale a wrapper around an inner, already existing scale. I haven't thought at all about the implementation, I'm just throwing some ideas...

re: breaking the plot itself: you probably can get away with inserting a point with value NaN.

I didn't check, but did all the baselines get updated because they were in fact already slightly invalid but the errors were hidden by the low resolution? If so, does...