Optimizing startup time

[me21]

It would be nice to conditionally omit imports of LinePlot and Plot classes. Doing so brought the startup time on Beaglebone Black single board computer from 20+ to 8 seconds.

Another similar thing is justpy importing its chartcomponents.py inside, but perhaps this is an issue for their own repository.

[falkoschindler]

I'm having trouble reproducing your result. The following code measures the time from starting the script until the UI is running. Independent of whether LinePlot and Plot are imported in ui.py, the time is always around 1.4s on my MacBook. Maybe it's an issue with Matplotlib on different processor architectures. Or I misunderstood something.

import time
t = time.time()

from nicegui import ui

ui.label('Hello world!')

ui.on_startup(lambda: print(f'{time.time() - t:.3f}'))

ui.run(port=1234)

[me21]

I'll try to run your code, but I think MacBook is much more powerful than Beaglebone Black, so maybe it's not noticeable there.
I'll update this issue later.

[me21]

With commented imports:

debian@beaglebone:~$ web_env/bin/python nicegui_test.py
/home/debian/web_env/lib/python3.7/site-packages/justpy
Module directory: /home/debian/web_env/lib/python3.7/site-packages/justpy, Application directory: /home/debian
JustPy ready to go on http://0.0.0.0:1234
14.747

With uncommented imports:

debian@beaglebone:~$ web_env/bin/python nicegui_test.py
/home/debian/web_env/lib/python3.7/site-packages/justpy
Module directory: /home/debian/web_env/lib/python3.7/site-packages/justpy, Application directory: /home/debian
JustPy ready to go on http://0.0.0.0:1234
37.066

In my research, I ran python interpreter with -X importtime flag to generate import statements timings. And viewed it with tuna tool. But the results are the same, nicegui is much slower on Beaglebone Black with plot imports.

[falkoschindler]

Ok, we checked the timing on a Jetson Nano and got 4.0 seconds each. The large difference might be caused by the little RAM you have available on the Beaglebone Black.

Nevertheless, we will shortly implement feature flags for things like Matplotlib, Highcharts, AG Grid and Three.js.

[falkoschindler]

Ok, it is finally released as v0.7.29