Implement support for reactive generator functions

[philippjfr]

In looking at #5016 I realized that the only way to write reactive expressions with multiple steps in a natural way would be using generators. So I had a go at implementing support for them and now I'm in love, here's a simplified version of the app in #5016:

Asynchronous Generator

import asyncio
import panel as pn

pn.extension(sizing_mode="stretch_width", template='bootstrap')

run = pn.widgets.Button(name="Press to run calculation")

async def runner(running):
    if not running:
        yield "Calculation did not run yet"
        return
    yield pn.Column(
        pn.indicators.LoadingSpinner(value=True, width=25, height=25),
        "Running...Please wait!",
    )
    time_start = time.perf_counter()
    await asyncio.sleep(1.5)
    time_end = time.perf_counter()
    yield f"""Done!
    The function took {time_end - time_start:1.1f} seconds to complete"""

pn.Column(
    "# Calculation Runner", run, pn.bind(runner, run)
).servable()

Classic Generator

import time
import panel as pn

pn.extension(sizing_mode="stretch_width", template='bootstrap')

run = pn.widgets.Button(name="Press to run calculation")

def runner(running):
    if not running:
        yield "Calculation did not run yet"
        return
    yield pn.Column(
        pn.indicators.LoadingSpinner(value=True, width=25, height=25),
        "Running...Please wait!",
    )
    time_start = time.perf_counter()
    time.sleep(1.5)
    time_end = time.perf_counter()
    yield f"""Done!
    The function took {time_end - time_start:1.1f} seconds to complete"""

pn.Column(
    "# Calculation Runner", run, pn.bind(runner, run)
).servable()

Streaming example

import asyncio
import panel as pn
import random

pn.extension()

@pn.depends()
async def stream():
    while True:
        await asyncio.sleep(0.1)
        yield random.randint(0, 100)
        
pn.indicators.Number(value=stream, name='Random')

[codecov]

Codecov Report

Merging #5019 (089f664) into main (480e6b0) will increase coverage by 10.17%.
The diff coverage is 91.37%.

@@             Coverage Diff             @@
##             main    #5019       +/-   ##
===========================================
+ Coverage   73.40%   83.57%   +10.17%     
===========================================
  Files         271      271               
  Lines       38361    38586      +225     
===========================================
+ Hits        28157    32247     +4090     
+ Misses      10204     6339     -3865     

Flag	Coverage Δ
ui-tests	40.72% <20.00%> (?)
unitexamples-tests	73.54% <90.98%> (+0.14%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
panel/depends.py	75.67% <76.19%> (+0.93%)	⬆️
panel/reactive.py	80.98% <81.69%> (+0.87%)	⬆️
panel/param.py	87.57% <81.81%> (+0.81%)	⬆️
panel/io/convert.py	72.18% <100.00%> (+46.99%)	⬆️
panel/layout/base.py	91.84% <100.00%> (+0.01%)	⬆️
panel/tests/test_param.py	99.73% <100.00%> (+0.01%)	⬆️
panel/tests/test_reactive.py	99.32% <100.00%> (+0.10%)	⬆️

... and 59 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[MarcSkovMadsen]

The streaming example is extremely powerful. The Async and Classic generators too. ❤️

The downside is that we have to teach our users yield. 🤔 I think its pythonic though and where Python is heading.

[MarcSkovMadsen]

Questions

• Does pn.bind(runner, run) run twice when I click the run button? Has it always done that?
• Does this work with other widgets like an IntSlider?
• Does the streaming example perform nicely if you have 100 streaming generators in one app and 2-100 simultaneous users? I'm asking because I've found that its not performant to setup 100 period callbacks - its better to update 100 times in one. It would be so great not to have to couple the streaming components.

[MarcSkovMadsen]

Other use cases for generator functions

I added another use case for generator functions in Panel that I believe should be supported as well. Please consider this in the context of this PR. Thanks. #5020

[philippjfr]

Does pn.bind(runner, run) run twice when I click the run button? Has it always done that?

It runs once but if you bind a Button to a function it'll run the first time with a value of False and on every click it'll run with a value of True.

Does this work with other widgets like an IntSlider?

Absolutely, it works just like a regular function with bind.

Does the streaming example perform nicely if you have 100 streaming generators? I'm asking because I've found that its not performant to setup 100 period callbacks - its better to update 100 times in one. It would be so great not to have to couple the streaming components.

Absolutely, this should be significantly more lightweight than periodic callbacks, here's 100 instances of the streaming generator

[philippjfr]

Okay, feature-wise this is now done. Last item was implementing synchronization primitives to ensure the async generators were updated when a dependency changes, e.g. we make the timeout dynamic:

import asyncio
import panel as pn
import random

pn.extension()

timeout = pn.widgets.FloatSlider(start=0.1, end=1.0, value=1, name='Timeout')

@pn.depends(timeout)
async def stream(timeout):
    while True:
        await asyncio.sleep(timeout)
        yield random.randint(0, 100)

pn.Column(
    timeout,
    pn.FlexBox(*(pn.indicators.Number(value=stream, name='Random', font_size='12pt') for _ in range(200)))
).servable()