Can't generate multipanel plot with PhasePlot

[gravertino]

Bug report

Bug summary

A multipanel plot with PhasePlot can not be generated normally following the guidance in the cookbook unless using annotate_title to add a title.

Code for reproduction

A minimum reproducible example is listed below. This example can not generate a mutlipanel phase plot normally (shown in Actual outcome ). However, if the line phs.annotate_title(title) hasn't been commented out, the plot will be created correctly (shown in Expected outcome ).

import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1 import AxesGrid
from yt.testing import fake_amr_ds
import yt

fig = plt.figure()
grid = AxesGrid(
fig,
111,
nrows_ncols=(2, 2),
axes_pad=0,
label_mode="L",
share_all=True,
cbar_location="right",
cbar_mode="single",
cbar_size="3%",
cbar_pad="0%",
aspect=True )
for i,title in enumerate(["995","1133","37373","37548"]):
    ds = fake_amr_ds()
    phs = yt.PhasePlot(ds,("gas","x"),("gas","y"),("gas","volume"),weight_field=None)
    # phs.annotate_title(title) # the multipanel plot can't be generated if this line has been commented out.
    plot = phs.plots[("gas","volume")]
    plot.figure = fig
    plot.axes = grid[i].axes
    if i == 0:
        plot.cax = grid.cbar_axes[i]
    phs.render()
plt.savefig("test.jpg")

Actual outcome

Expected outcome

Version Information

• Operating System: WSL2 (Ubuntu-20.04)
• Python Version: 3.8.10
• yt version: 4.3.dev0
• Other Libraries (if applicable): matplotlib: 3.6.2

[welcome]

Hi, and welcome to yt! Thanks for opening your first issue. We have an issue template that helps us to gather relevant information to help diagnosing and fixing the issue.

[neutrinoceros]

wow, that's a weird one !
An interesting consequence of the current behaviour is that the title-less version of the multi-panel figure may be obtained by calling phs.annotate_title("").

Thank you for reporting and providing a nice, self-contained and minimal reproducer ! I'll look into it now.

[neutrinoceros]

So, it's actually not something weird going on with annotate_title specifically: any plot annotation has a similar effect here. What they all have in common is that they invalidate the plot object: they force a new rendering next time the figure needs to be displayed.
In fact, replacing psh.annotate_title("") with phs._plot_valid = False also produces the correct result.
I also note that if I replace the phase plot with a slice plot, _plot_valid is False from the start (no annotation needed), so the end result is correct.
This seems to indicate that the attribute _plot_valid is improperly initialised to True in PhasePlot.

[neutrinoceros]

Got a minimal fix in #4491

[gravertino]

Thank you for looking into this problem, and this explanation sounds very reasonable! But I still have a little question, why phs.render() didn't redraw itself as expected?

[neutrinoceros]

That's a great question !
The render method calls a private, equivalent method _setup_plots. Looking how PhasePlot._setup_plots is implemented, we see that it intentionally does nothing when the _plot_valid attribute is True

yt/yt/visualization/profile_plotter.py

Lines 369 to 371 in ec1e375

	def _setup_plots(self):
	if self._plot_valid:
	return

In fact, all other PlotContainer classes (like SlicePlot, ProjectionPlot ...) follow this pattern too.

You are right that it's also a bug that explicitly calling render() isn't doing anything. In my opinion it should never be no-op. I'll open another PR to fix that too

[gravertino]

Thank you for your clear reply and now I understand what's happening here! Many thanks!