GeezerVison

agammann left a comment

With set_theta_direction(-1) and an explicit rlabel position on a wedge,
the rlabel pad shift pointed to the clockwise side of the radial line
while RadialTick.update_position aligns the text on the counterclockwise
side, so the labels straddled the gridline. Shift toward the side the
text is on.

Extend test_polar_rlabel_position_wedge to both theta directions and
assert the anchor, the pad shift and that the whole label stays on the
padded side of the radial line.

Keep only the rlabel angle, None until set_rlabel_position is called,
in place of the Affine2D and the flag that recorded whether the position
was set. PolarAxes._get_rlabel_theta decides in one place where the
labels go: the set angle, else the default angle on a full circle and
thetamin on a partial wedge. The lazily evaluated _RLabelTransform
builds the label translation from it, and label1, tick1, the 'rlabel'
pad and RadialTick.update_position all use it, so the "full or manual"
special cases are gone from draw(), get_yaxis_text1_transform and
update_position.

With a set position on a wedge, this also fixes the padding when the
position changes after a draw or when the ticks existed before it was
set, keeps the end-edge labels and tick marks (label2, tick2) on their
edge where they used to jump to the set angle plus one radian, and
turns the start tick marks perpendicular to the label line as on the
default edge. With inverted theta limits, the default labels now follow
thetamin, which their alignment and padding already assumed.

Add tests for setting the position over existing ticks and for a
position at thetamin giving the default wedge layout.