Bug summary

As seen in #27491, #27488, #27494, #27487

It appears that the tri* methods do not actually properly handle units.

Code for reproduction

fig, ax = plt.subplots()
np.random.seed(19680801)
limit_value = 30

startdate = np.datetime64("2023-12-01")

x_dates = startdate + np.random.randint(0, 15, size=(limit_value-1,)).astype("timedelta64[D]")
y_dates = startdate + np.random.randint(0, 15,size=(limit_value-1,)).astype("timedelta64[D]")

x_dates_converted = mpl.dates.date2num(x_dates)

y_data = np.random.rand(limit_value - 1)

z_data = np.sin(x_dates_converted) + np.cos(y_data)

ax.triplot(x_dates, y_dates)
ax.tripcolor(x_dates, y_dates, z_data)
#ax.tricontour(x_dates, y_dates, z_data)
#ax.tricontourf(x_dates, y_dates, z_data)

Actual outcome

Prior to (probably, though I didn't actually bisect it) #24522, the code would actually error, as it attempted to cast the x/y arrays to float64 arrays.

Now it will actually work to plot (seemingly even getting the actual values converted, though now that I think about it not sure where... my guess is that it is actually coincidental that they are datetime64[D] dtype which happens to be the same scale/epoch as mpl, but it is not actually being properly converted) however it does not set the Formatters/etc.

Expected outcome

Units are set properly (including axisinfo such as formatters).

Additional information

No response

Operating system

No response

Matplotlib Version

3.9.0.dev821+gfa6be74127 (~main currently)

Matplotlib Backend

No response

Python version

No response

Jupyter version

No response

Installation

git checkout