The Percy diffs are because this code generates x domains that range from 0-1 instead of master which goes from 0-0.98. 0-1 seems "more correct" to me... @jonmmease any insight into why the previous version does what it does?

Ah I found it: https://github.com/plotly/plotly.py/blob/master/packages/python/plotly/plotly/subplots.py#L529

Basically I'm now passing a length-0 element to row_titles in single-facet cases, not [None] and so it skips this line.

OK so I'm fine with these Percy diffs. Ready for review!

I guess I need some tests.

It was not introduced by this PR, but the behaviour when there is a very large number of cols/rows is weird. For example

px.scatter(gapminder, x="pop", y="lifeExp", facet_row="country")

or

px.scatter(gapminder, x="pop", y="lifeExp", facet_col="country", facet_col_wrap=8)

give errors like

~/code/plotly.py/packages/python/plotly/_plotly_utils/basevalidators.py in raise_invalid_val(self, v, inds)
 281 typ=type_str(v),
 282 v=repr(v),
--> 283 valid_clr_desc=self.description(),
 284 )
 285 )
ValueError: 
 Invalid value of type 'builtins.float' received for the 'domain[1]' property of layout.yaxis
 Received value: -0.010555555555555565
 The 'domain[1]' property is a number and may be specified as:
 - An int or float in the interval [0, 1]

which is not very helpful as an error message. Should we either introduce a limit in the number of cols/rows, or try to catch this and output a better message?

Thank you for fixing the bug of #1836.

ok, I just left a doc suggestion which you might or might not implement. My comment about the large number of cols/rows is not specific to this PR and can go to an issue. 💃

Nice catch on the "many columns" thing, here's the issue I broke it off into: #1839

OK I've added percy tests for the two tricky cases of facet wrapping, and a test locking down the fix for #1836

The plotlyjs_dev_build test failure looks like something which will go away the next time there's a master commit on plotly.js