add_shape, add_layout_image, add_annotation for multiple facets at once

[nicolaskruchten]

Right now if I want to add the same shape (i.e. a horizontal line) to every facet in a PX figure, I have to loop through and do it myself. It seems like having a way of saying row="*" or something might be handy.

[emmanuelle]

row='all'? As a linux user I love the * but it's a bit too geeky for most users.

[nicolaskruchten]

Once we do the easy version of #2141 then this should apply to add_hline and add_vline as well.

[nicholas-esterer]

It looks like the change can just be added by iterating over rows and columns in packages/python/plotly/plotly/basedatatypes.py:BaseFigure._add_annotation_like because, although no copy of the new_obj argument is made, calling id on the objects added to self.layout[prop_plural] shows each is stored in unique memory. I assume this uniqueness is to be maintained in future versions?

[nicholas-esterer]

For the specification of multiple rows and columns, I propose the following options

row	col	The rows and columns addressed
"all"	"all"	Every combination of row and column (the cartesian product)
"all"	list	For each column specified, every row
"all"	int	For the column specified, every row
list	"all"	For each row specified, every column
list	list	If len(row)==len(col) The row and column pairs formed by calling zip on both lists, otherwise all combinations of specified rows and columns
list	int	The rows in the specified column
int	"all"	All the columns in the specified row
int	list	The specified columns in the specified row
int	int	The specified row and column

[nicolaskruchten]

Thanks for starting to flesh this out! One thing I'd like to think about is whether we bolt this on to e.g. add_shape or whether we add a new plural method add_shapes analogous to the add_trace/add_traces pair we have. I'm reluctant to add new methods to go.Figure because we have so many but we should at least consider it.

If we add these new plural methods, then we could consider row=None to mean row="all", say.

[nicholas-esterer]

I've also added the possibility of passing the tuple (list,'all') for rows or columns. In this case, all the combinations of rows and columns are used to look up the subplot, even if rows / columns are the same length.

[nicholas-esterer]

Finally, range(...) can now also be passed (it simply gets converted to a list), but maybe this is overkill (the change to support this is very small, though).

[nicholas-esterer]

Thanks for starting to flesh this out! One thing I'd like to think about is whether we bolt this on to e.g. add_shape or whether we add a new plural method add_shapes analogous to the add_trace/add_traces pair we have. I'm reluctant to add new methods to go.Figure because we have so many but we should at least consider it.

If we add these new plural methods, then we could consider row=None to mean row="all", say.

I've currently implemented in change in the BaseFigure._add_annotation_like of packages/python/plotly/plotly/basedatatypes.py. This way it should work for all the other go.Figure.add_* functions. I haven't looked, but would we win this "polymorphism" if we added a new method? Howver, I'm also happy to move it to its own function, if that's preferable.

[nicolaskruchten]

Oh I hadn't realized you were already implementing :) Would you mind opening up a draft PR so I can see how you're approaching this?