Bug: `formatter.log=True` not respected in `ax.format`

[syrte]

Description
When uplt.rc["formatter.log"] = True, log formatting is correctly applied via ax.loglog(), ax.semilogx() etc, but ignored when setting the scale via ax.format().

Minimal Reproducible Example

import ultraplot as uplt
import numpy as np

uplt.rc["formatter.log"] = True
fig, ax = uplt.subplots()

# Applied correctly
ax.semilogx(np.linspace(1, 100)) 

# Applied incorrectly (remains scalar)
ax.format(yscale='log') 

uplt: 1.66.0
mpl: 3.10.8

Expected Behavior
Both axes should display labels in log format.

Actual Behavior
Only the axis initialized via semilogx uses log formatting. The axis set via ax.format(yscale='log') uses standard scalar labels.

[cvanelteren]

Are you referring to the log formatter or the ticker? The ticker seems to work normally but the y formatter is showing linear labels (e.g. not 10^x).

[syrte]

Are you referring to the log formatter or the ticker? The ticker seems to work normally but the y formatter is showing linear labels (e.g. not 10^x).

I mean the formatter of tickers. Note the difference in the x-axis and y-axis. Naively, I'd expect they have the same formatter (sci-notations).

[cvanelteren]

I think the confusing part is that we use [x/y]formatter for this stuff, e.g.

uplt.rc["formatter.log"] = True
fig, ax = uplt.subplots()

# Applied correctly
ax.semilogy(np.linspace(1, 1000))

uplt.show(block=1)

will work accordingly. The [x/y]scale functions are directly parsed through matplotlib and caused this discrepancy.

[cvanelteren]

I think perhaps the formatters are reset through matplotlib -- we could cache the formatter and restore it after the matplotlib setter

[cvanelteren]

Fixed on main.