Possibly functionalize the plotting functions?

[daquinteroflex]

Break this up:

• Restructuring the plotting functions into viz
• On the fly import of matplotlib
• Eventually restructure the same functions to be composed using the (fig, axs) mechanism.
• Possibly introduce standard object.plot functions using the viz module as an internal function call onto self.

Based on a conversation I had with Marc and my own personal experience:

Maybe it'd be nicer in terms of extensibility if we had purely functional plotting? Something on the lines of:

fig, axs = my_plotting_function(fig, axs, ...)
fig, axs = my_second_plotting_function(fig, axs, ...)

This can make it easier to represent information as just a set of operations between fig, axs and kwargs. We can move all the plotting functionality outside the classmethods and maybe make it more reusable? This way we avoid

def current_plotting_function():
    fig, ax = self.another_function(...)
    ax.do_something()
    self.another_function_in_this_class(fig, ax) # can mutate the fig, ax object directly and sometimes not accessible at a higher level
    return fig, axs

[momchil-flex]

I'm in favor of this, only curious about the exact scope especially from user perspective. If it's about moving all plotting functionality in a viz module while still keeping the same user-facing methods, then I think it sounds great. It's also in line with what we discussed about "properly" eliminating matplotlib as a core dependence.

[daquinteroflex]

Yeah that'd be great actually, and change the functional definitions accordingly when we do this change.

my_pseudo_static_classes/
viz/
   all_my_internal_functions/

and then refactor those functions in that module to be used in:

class MyClass:

   my_plotting_function -> (fig, ax) = viz.my_function() # just like before but also 

In principle, for backwards compatibility we can make my_plotting_function -> (fig, ax) return the same data as before with the new structure.

[momchil-flex]

Yeah, what I like about this apart from backwards compatibility is that otherwise it seems to me that we have two usage options, either:

viz.plot(object: Union[Simulation, Source, Monitor, ...])
# ugly-ish internal dispatch based on ``isinstance(object)``

or

viz.plot_simulation(simulation)
viz.plot_source(source)
viz.plot_monitor(monitor)
# a bit annoying to use

and both seem inferior to just object.plot().

[tylerflex]

side note: should we look into using plotly instead of matplotlib? the approach to plotting seems much more in line with the overall style of the frontend. every plotlly plot is backed by a json representation whereas it's state based in matplotlib. so we could in principle have properties or methods to attach plotly plots to our Tidy3D components, or define what these plotting components look like and assemble the total plot at the higher level viz.py interface.