I guess autoscaling should always take avhlines into account.

fig, ax = plt.subplots()
ax.axvline(0.5, color='b', transform=ax.transAxes)

I got: ValueError: 'transform' is not allowed as a kwarg;axvline generates its own transform.

but the doc suggested that transform is supported.

Am I missing something? If I wanna plot a line in the middle of a Axes, how should I do it for now rather than set xlim?

@StevenLi-DS axvline is a high-level convenience function for the special, but frequently used case where you want a vertical line spanning (part of the axes) positionned at a certain x coordinate in data units. The documentation does not suggest that transform is supported, in fact it clearly states

Valid kwargs are Line2D properties, with the exception of 'transform':

You can always achieve the same with a usual line and the respective transform, hence ax.axvline(0.5) is equivalent to ax.plot([.5,.5], [0,1], transform=ax.get_xaxis_transform()).

If you want your line to be completely in axes coordinates, it would be ax.plot([.5,.5], [0,1], transform=ax.transAxes).

All of this is unrelated to the original issue here. It would help to use a new issue in such cases in order to not disturb the thread of arguments in existing ones.

Back to the original thread. I think axvline should not change the axes limits. Ever. It’s a way of annotating data, not data. If someone wants to plot a vertical line then that’s what plot is for.

I think I would personally agree with "axvline should not change the axes limits".

Related to this: #7742, which already did undergo two unsuccessful attempts to fix.

So should it also not change limits even if plotting outside of the current limits? (I'm fine with that solution too.)

That would be what I would expect. But maybe other users not? In particular, I think it would be a significant change to how it is handled now.

In an ideal world it would not auto scale at all and explicitly say that in the docs

It's probably not too hard to change that behavior with a deprecation warning (check whether the limits would change and if so warn), the only question is whether it's worth the trouble.

axvline should not change the axes limits

not sure I agree with that. If the user plots this line, there's some intent behind it and it's most likely undesired if the line does not show up.

Example: Plot some relative values. Values larger than 1 represent data that got larger, values below 1 got smaller. It can be reasonable to axhline(1) to define the different semantics. It might be an essential information. If my data happen to be all smaller than 1, not taking into account the line would shift it off-screen, which would remove an important aspect of the graph.

I'd argue in that case the user can specify those important limits manually. My conceptualization of axvline is that its a special grid line.

I appreciate other folks might conceptualize it differently, so we should decide if it autoscales (in x) or not, but not do something that is half way.

If I plot a vertical line at x=0.5, I'd expect 0.5 to be considered when automatically determining X plot limits. I would not expect to need to manually tell matplotlib to include this.

... and would you expect an auto scaling margin around the line? If so, we should remove the funky logic and just autoscale.

I would expect a margin, for the same reason we have margins at all. Removing funky logic sounds good 😄.

@dopplershift If you plot a vertical line at x=0.99 while the current limits (computed on the rest of the data) is (0, 1), do you expect the 5% margin to go to 1.05(-epsilon)?
Just trying to clarify what everyone's opinion is...

@ImportanceOfBeingErnest

The documentation does not suggest that transform is supported

The reason why I was confused is that transform is listed in Other Parameters:

https://matplotlib.org/3.1.1/api/_as_gen/matplotlib.axes.Axes.axvline.html

Yeah, thats an unfortunate artifact of just repeating the Line2D kwargs. The caveat at the beginning tries to warn about that, but its kinda hard to catch if you are looking at a list of 30 properties.

@anntzer Personally, I expect the margin to increase as well. It may "just" be a notational line, but the whole point of margins is better aesthetics around things on the plot, keeping all items (dataspace wise) from being right on the edge.

so in #14651 (comment) you think we should increase the margins in the lower left plot?

Yes.

Do we understand why the complicated logic was put in, and does removing it fail any tests?

This came in in d4965f (2008); all the commit message says is "fixed a problem with axh and vline scaling". Previously it was never scaling, which is the behavior you (@jklymak) are advocating.
So I guess the answer to your question is "to fix a problem" :)

I feel strongly that the behaviour should be simple and consistent. I feel mildly that these should not participate in auto scaling, but don’t mind the opposite behaviour

I agree about simple and consistent. I can say that my reaction, if I call axvline or axhline, and the line does not appear in my plot due to unadjusted data bounds, will be WTF matplotlib?

pathces.Rectangle has the non-auto-scaling behavior, which is problematic if some things remain where they are and others are scaled. I agree the behavior should be consistent.