Hi @StefRe , I came across this PR, thank you. Previously I just manually adjusted the position of the title.

I think it will be great to also consider the x position of yaxis. With your current fix, the title located centered is also moved, as shown below. While in matlab, it seems that the title only get moved upwards if there will be overlap.

fig, ax = plt.subplots()
ax.plot([1e8, 2e8])
ax.set_title('Left', loc='left')
ax.set_title('Center')
ax.set_title('Right', loc='right')
plt.show()

@yuanx749 Thanks for your feedback. I don't think, however, we should strive to cover all possible cases: moving the title up to not overlap with top ticks (which is current behavior) also sets the title position for all titles, not just for the ones that would overlap:

import matplotlib.pyplot as plt
fig, ax = plt.subplots(layout='constrained')
ax.plot([1e8, 2e8], [1, 2])
ax.set_title('Left', loc='left')
ax.set_title('Center')
ax.set_title('Right', loc='right')
ax.xaxis.tick_top()
plt.show()

I'm 50/50 on this. I'm not sure we want titles in the centre to be so high so maybe only do this if the left title is not empty?

If we have multiple titles as in the example (which is, admittedly, seldom in practice), the different hights of the different titles would look awkward.

I agree. I just meant that we only move all three of the left title is occupied.

I guess a counter argument is that some centred titles can be very long, but I think most of the time folks are going to complain about all the wasted vertical space.

Maybe we can consider this as a separate issue and only change one thing at a time, just to not overcomplicate things:

• right now the title gets moved above the highest top x axis tick label and x offset text
• this PR proposes the same for the case that there is an y axis offset text
• neither the current code not this PR check if there's indeed an overlap between the title at its given position and any axis decorations
• if this is desired (to save space), it could be discussed as a separate issue / PR (I'm not overly enthusiastic about it); right now there's just one y position for all three locations ('center', 'left', 'right'), so I guess we'd need three y positions then?

I guess I'm not being clear. My proposal is you only change the y position to account for the y offset if there is a string in the left title. I'm not proposing 2 or more separate positions.

@jklymak I think I understood you correctly - the only thing I wanted to say is that if we move the left title only (to not overlap with an y axis offset text), then we should also change the existing code so that center and left titles are moved up just above the top x tick labels, not above the x offset text (in case there is an x offset).

Right now all titles are treated the same way:

Further, all titles are currently lined up at one level:

This is why I suggested to deal with these things in a separate issue "Update title positions per title to move them just the necessary amount", if this is really needed.

@StefRe, I am not sure you did understand correctly, or I'm not understanding you. I am agreeing with you that the vertical level of all three titles should always be the same. And in general I am agreeing with fixing the overlap between a left-side title and the y-axis offset text.

However, I disagree that we should always displace all the titles if there is y-axis offset text. That is a breaking change that will affect many people, adding undesirable new vertical space to 99.9% of the users who just have a centered title.

You mention the precedent of the x-axis being taken into account, in particular when it is at the top of the axes. That is, in fact, the whole point of the automatic repositioning feature in the first place. Having the extra space was deemed useful because the top of the axes was already full of x-tick labels, and usually the xlabel itself. I don't think that applies to an offset text on the left side and most centred titles.

Just to re-iterate, I think your PR should go in as-is, except the logic should only be called if the left title string is non-empty. (Or the right title string, if the y axis is on the right, I suppose). If that is satisfied, I agree all three titles should move up together. But I don't think we should do this if the left title string is empty.

Updated version which only moves the title if it indeed would overlap and aligns them if there are multiple titles:

• a) move left title above offset
• b) don't move title if it doesn't overlap
• c) + d) same for right offset
• e) align multiple titles at the highest one
• f) works for any title

Just for the records: this PR doesn't change title movement in case of top x axis (as discussed above):

• g) right title is above (right) top x axis offset
• h) center title is moved too, although it wouldn't overlap

Demo code:

import matplotlib.pyplot as plt

fig, axs = plt.subplots(3, 2, layout='constrained', figsize=(6,8))
for ax, letter in zip(axs.flat, 'abcdef'):
    ax.text(.05, .9, letter + ')', transform=ax.transAxes)
    ax.set_ylim(1e8)

axs[0,0].set_title('Left', loc='left')

axs[0,1].set_title('Center')

axs[1,0].yaxis.tick_right()
axs[1,0].set_title('Right', loc='right')

axs[1,1].yaxis.tick_right()
axs[1,1].set_title('Center')

axs[2,0].set_title('Left', loc='left')
axs[2,0].set_title('Center')

axs[2,1].set_title('A very long center title')

fig, axs = plt.subplots(1, 2, layout='constrained', figsize=(6,3))
for ax, letter in zip(axs, 'gh'):
    ax.text(.05, .9, letter + ')', transform=ax.transAxes)
    ax.set_xlim(1e8)
    ax.xaxis.tick_top()
    
axs[0].set_title('Right', loc='right')
axs[1].set_title('Center')

Hm, I'm a bit at a loss: there are 7 failing tests

test_axes.py::test_errorbar[png], [svg], [pdf]
test_streamplot.py::test_direction[png]
test_axisartist_grid_helper_curvelinear.py::test_axis_direction[png]
test_mplot3d.py::test_trisurf3d[png]
test_mplot3d.py::test_stem3d[png]

for Python 3.8 + 3.9 on Ubuntu (they pass on 3.7 and 3.10 as well as for all versions on Windows and macOS) and I don't see how the failures are related to this PR:

• in test_errorbar the changed code section is not called at all (see debug message "_base.py:2963 title position was updated manually, not adjusting", and we have differing view limits for the log axis
  

• in the remaining 4 failing tests we have some displacements of the drawn objects:
  axis_direction      axis_direction-expected      axis_direction-failed-diff
  stem3d      stem3d-expected      stem3d-failed-diff
  streamplot_direction      streamplot_direction-expected      streamplot_direction-failed-diff
  trisurf3d      trisurf3d-expected      trisurf3d-failed-diff

I can't debug as I only have a Python 3.10.1 Arch and a Python 3.8 Windows computer and all tests pass here. Do you have any idea or hint how to proceed from here?

We are not sure what is causing the tests to fail, but they are failing on doc PRs so something has gone wrong in our library chain.

I guess this needs an API change note; put in draft until its added, definitely feel free to pop back....