The subslice optimizer does not handle the case of fully nan inputs, so
the relevant check in recache() is not whether x is sorted but really
whether x is sorted and not fully nan.  Fix that.

Previously, `plt.plot([np.nan] * 2000, range(2000))` would crash.

meeseeksmachine pushed a commit to meeseeksmachine/matplotlib that referenced this pull request

oscargus added a commit that referenced this pull request

…978-on-v3.7.x

Backport PR #25978 on branch v3.7.x (Fix subslice optimization for long, fully nan lines.)