[Feature]: Elaborate "native" support for Plotly charts in Shiny Express (@render.express)

[pfalcon]

Category

Integrations

Scope

Minor Enhancement

Problem

As a new user, reading documentation on how to render Plotly charts, it suggests to use additional and separate shinywidgets package: https://shiny.posit.co/py/components/outputs/plot-plotly/ . However, it is a matter of fact that Plotly charts can be rendered without additional dependencies, and in the most natural way for the Shiny Express dialect: by simply "mentioning" the figure object:

# This is going to be text output
"Description of chart"

fig = px.line(...)
# This is going to be chart output
fig

However, how that works currently is flaky at best. But again, due to utter naturalness of syntax like above (for Shiny Express), I would kindly suggest that it's worth effort to make it work reliably and present to user among (the first) choices of rendering Plotly diagrams (would be naturally extended to other types charting packages, but I didn't try them yet and don't know the situation there).

This ticker is intended to serve as an "umbrella" issue, a statement of direction (make it all work out of the box), with specific issue to tackle linked against it.

Solution

I would like to suggest that the following code should work robustly:

import plotly.express as px
from shiny.express import input, ui


ui.input_slider("n", "Number of sth", 1, 100, 20)

fig = px.line(x=[1, 2, 3], y=[4, 1, 2])
fig

Likewise, I'd like to suggest that the following code also should work reliably:

import plotly.express as px
from shiny.express import input, ui, render


ui.input_slider("n", "Number of sth", 1, 100, 20)

@render.express
def render_func():
    _ = input.n()
    ui.h1("Heading1")
    fig = px.line(x=[1, 2, 3], y=[4, 1, 2])
    fig

Alternatives (Optional)

An alternative is of course to use shinywidgets, but how it all started for me is apparent inability to do following with it:

@render.express
def my_render():
    ui.h2("Chart 1")
    fig1 = px.line(...)
    # Wanna chart to be rendered here    

    ui.h2("Chart 2")
    fig2 = px.line(...)
    # Wanna chart to be rendered here    

Example (Optional)

Impact (Optional)

Would make Shiny Express mode much more natural. I'm a Streamlit refugee who would like to write dashboard apps in a similarly easy manner (but without straitjacket Streamlit enforces).

Contribution? (Optional)

Yes, I can implement (or help).

[pfalcon]

import plotly.express as px
frsom shiny.express import input, ui


ui.input_slider("n", "Number of sth", 1, 100, 20)

fig = px.line(x=[1, 2, 3], y=[4, 1, 2])
fig

So, the people who are lucky to bypass dependency minefield (or debug it, like myself), can get that working: posit-dev/py-shinywidgets#201 .

But then the situation is that following the official docs (https://shiny.posit.co/py/components/outputs/plot-plotly/) and installing the recommended sister package (shinywidgets) causes native Shiny Express support not to work. This is rather unideal situation, if you agree.

Some ideas how to deal with it:

1. Umm, fix it. But given that the culprit is the upstream project, it requires coordination/cooperation across projects. Solving that may take time.
2. But then as long as we agree that doing fig is the most natural way to render it for Shiny Express, then the official docs should at least mention it. And the mention that either there're known issues with it (so keep using shinywidgets if you can), or maybe provide some workaround(s) ahead of upstream fixture.

Note that my process was:

1. Follow the official docs and trying shinywidgets example.
2. "But how do I render 2 charts in row, in my existing @shiny.express function?"
3. Let's try just fig. No output (not even an error, though Shiny Express docs warned that there would be errors on unknown output types).
4. (Hacks to make shinywidgets's render_widget() to be usable not only as a decorator, but as an immediate function call).
5. (Morally prepares to submit tickets to the affected projects, so browses existing issues first. Then suddenly finds that shinywidgets maintainer suggests that Plotly charts are supposed to be renderable without dependency on shinywidgets: Animated Plotly Graphs Not Rendering py-shinywidgets#156 (comment))
6. Proceeds to investigate what's the (dependency) hell is happening there.

[pfalcon]

Likewise, I'd like to suggest that the following code also should work reliably:
@render.express

Issue with this code was posted as #2008

[cpsievert]

For this particular problem, I think you're looking for the ui.hold context manager? This way, you can still use the (more official) @render_plotly, but locate the UI elements into a larger "reactive express" context. Like so,

with ui.hold() as plot1:
    @render_plotly
    def plot_1():
        return ...

with ui.hold() as plot2:
    @render_plotly
    def plot_2():
        return ...


@render.express
def _():
    plot1
    "Something else"
    plot2