Colorbar positioning is problematic in subplot mode

[seisman]

Initially found in #2354 (comment).

Here is a minimal example to reproduce the issue:

import pygmt
fig = pygmt.Figure()
with fig.subplot(nrows=1, ncols=2, figsize=(10, 5)):
    fig.basemap(region=[0, 10, 0, 10], projection="X?", frame=True, panel=True)
    fig.basemap(region=[0, 10, 0, 10], projection="X?", frame=True, panel=True)
fig.colorbar(position="JBC+w5c+h", cmap="turbo", frame=True)
fig.show()

The "equivalent" GMT bash script is:

gmt begin
    gmt figure map
	gmt subplot begin 1x2 -Ff10c/5c
		gmt basemap -R0/10/0/10 -JX? -B -c
		gmt basemap -R0/10/0/10 -JX? -B -c
	gmt subplot end
	gmt colorbar -DJBC+w5c+h -Cturbo -B
gmt end show

PyGMT and GMT CLI produce different outputs, and the GMT CLI one is correct:

PyGMT	GMT CLI

[seisman]

After adding verbose="d" to the Figure.colobar() call in the PyGMT script, I see debug messages like:

pygmt-session [DEBUG]: Use PS filename /Users/seisman/.gmt/sessions/gmt_session.6653/gmt_1.ps-
pygmt-session [DEBUG]: gmtinit_get_current_panel: No current panel selected so not in subplot mode
pygmt-session [DEBUG]: Modern mode: Added -R to options since history is available [0/3.93701/0/1.9685].
pygmt-session [DEBUG]: Modern: Adding -JX15c/15c to options since there is no history available.
pygmt-session [DEBUG]: Revised options: -B -Cturbo -DJBC+w5c+h -Vd -R -JX15c/15c
colorbar [DEBUG]: History: Process -R
colorbar [DEBUG]: History: Process -JX15c/15c

For comparison, adding -Vd to the colorbar command in the bash script gives:

gmt [DEBUG]: Use PS filename /Users/seisman/.gmt/sessions/gmt_session.6943/gmt_0.ps-
gmt [DEBUG]: gmtinit_get_current_panel: No current panel selected so not in subplot mode
gmt [DEBUG]: Modern mode: Added -R to options since history is available [0/3.93701/0/1.9685].
gmt [DEBUG]: Modern: Adding -J to options since history is available [x1i].
gmt [DEBUG]: Revised options: -DJBC+w5c+h -Cturbo -B -Vd -R -J
colorbar [DEBUG]: History: Process -R
colorbar [DEBUG]: History: Process -J

So, for unknown reasons, PyGMT fails to get the history information for -J option and use the default -JX15c/15c instead, which explains the incorrect positioning of colorbars.

[seisman]

Ping @PaulWessel. Need your help here.

[PaulWessel]

Hm, got to be something different in the wrapper when it calls GMT compared to the CLI. I suggest you first do this

1. gmt clear sessions
2. Run each of the CLI commands and examine what is created in ~/.gmt/sessions/gmt_session.* Take a note of files being created and destroyed after each command
3. Repeat 1
4. Do the same on command at the time via PyGMT and compare to see what and when things go wrong.

[seisman]

Hm, got to be something different in the wrapper when it calls GMT compared to the CLI. I suggest you first do this

1. gmt clear sessions
2. Run each of the CLI commands and examine what is created in ~/.gmt/sessions/gmt_session.* Take a note of files being created and destroyed after each command
3. Repeat 1
4. Do the same on command at the time via PyGMT and compare to see what and when things go wrong.

Thanks. It helps me find the culprit. It's NOT a GMT bug. It's a bug on the PyGMT side.