Adding _update_collection_limits could be a reasonable intermediate step to make handling more consistent. But overall, we likely want to move a way from _update_* methods, see #15595 (comment).

Edit: It may be much more complicated than this. Since we have the option add_collection(..., autolim=False) we must not always update. If a user/function has decided not to update the limits, we must not push updates through the backdoor of relim(). This means the bug cannot be fixed properly without addressing the general set_in_autoscale concept (#15595) - or hacking in a speical solution for collections, which is most likely not the way to go.

@Archiljain thanks for the contribution so far. As the topic is more involved, we cannot do a standalone solution without tackling a mechanism that makes inclusion of Artists in autoscaling optional.

Following up on my previous note regarding entanglement of Collection limits considered in relim() and optional inclusion in autoscaling, I think the way forward is this:

1. PR1: Implement a private version of set_in_autoscale()
2. PR2: Use that version to correctly handle conditional autoscaling of collecitons depending on the autolim paramter in add_collection(..., autolim=...). Thereby fix this topic here, also adding _update_collection_limits()
3. PR3: Consider rearchitecting limits handling further as discussed in Add set_in_autoscale() method  #15595 (comment). This may include replacing the _update_*_limits() functions by retrieveing limits from the Artists.

Thanks for the clarification.

I'm happy to work on PR1 (implementing the private set_in_autoscale mechanism) as a first step if that helps move this forward. Please let me know if that’s the preferred direction.

@Archiljain you are welcome to work on this. Please read the full discussion in #15595 and use for context, in particular #15595 (comment).

Since this is internal for now, naming is not too critical. I would stick to the in_autoscale terminology for now. But we may come up with a better name.

I propose introducing a private Artist-level _in_autoscale flag (mirroring _in_layout), checked centrally in Axes.relim() to gate whether an artist contributes to data limits. No behavior change yet.

Thanks for the guidance. I see that this needs a more foundational change.
I’ll put together a separate PR for the private Artist-level autoscale flag (PR1), and then rework this PR on top of that as PR2.

This should be re-done in a new PR; #31166 contained these commits so GitHub marked it as merged, but it also reverted all these changes, so they haven't actually landed in main.

I'm planning to apply for GSoC this year and would love to continue contributing to autoscaling improvements. Are there related issues where help would be valuable?