lazy imports using lazy-imports library

[evnchn]

Motivation

Observing how we basically re-import NiceGUI again if we want to do run.cpu_bound, lazy imports which defer heavy imports until actually needed is a promising approach to make it faster (ref: #5684)

Implementation Strategy

It began with my rapid iteration with Manus to use https://github.com/bachorp/lazy-imports, and we iterate until from nicegui import ui works.

Implementation (this part by Copilot)

Refactoring and performance improvements:

Clipboard API modernization:

• Refactored the clipboard helper functions into a Clipboard class with async read, write, and read_image methods, and provided a singleton clipboard instance for use throughout the codebase. [1] [2] [3] [4]
• My note: Check lazy imports using lazy-imports library #5303 (comment) for additional context.

UI lazy loading:

• Introduced lazy imports for all UI elements in ui.py using the lazy_imports library, significantly improving startup performance by loading modules only when accessed. Registered the lazy module in sys.modules for proper caching. [1] [2] [3] [4]
• Added lazy-imports as a required dependency in pyproject.toml.

Import handling fixes:

• Eagerly imported all importmap-based elements in client.py to resolve issues with lazy imports and ensure they are available when needed.
• My note: Check lazy imports using lazy-imports library #5303 (comment) for additional context.

Progress

• I chose a meaningful title that completes the sentence: "If applied, this PR will..."
• The implementation is complete.
• Pytests pending: Not sure if Pytest test isolation ensure each test start off fresh, or if some components are already lazily-loaded. Either way we want to test both if possible
• Documentation is still good-to-have since this lazy-import behaviour could throw off some people, for example those who expect the library to currently function in a certain way.

[evnchn]

f415575 fixed the double-access on ui.label breaking.

Now the basic code works:

from nicegui import ui

ui.label('First label')
ui.label('Second label. Did not break!')

ui.run()

As always pipeline will be our guiding star. Let's see tomorrow.

[evnchn]

Pipeline's less dramatic than I think. 2 things are broken, mainly:

1. ui.clipboard totally unavailable (I think this is truly what the docs warned us)
2. importmap is empty from the start, and so ui.mermaid and ui.echarts does not work.

Actually point 2 can theoretically affect someone who's

• creating custom elements,
• expect to use importmap for importing your stuff,
• and 🤡 enough to dynamically define the element's class.

Gist is: with lazy-imports what we call 🤡 before is actually the 🧠 move empowering lazy import's efficiency.

[evnchn]

Well, as of 425cb4f the only failing tests is the ui.clipboard one.

@falkoschindler This is getting somewhere, somewhere farther than anywhere we had been. Consider closing:

• Test environment for faster startup #3365
• Profile imports #4429
• Create a proof-of-concept for fast cpu_bound with lazy import #4471

[evnchn]

If in #4564 we swapped a function for a Class, may we in this case swap a module for a Class and then we can solve issue 1.

[falkoschindler]

@evnchn I added the other PRs to the description of this PR. If this PR converges and is merged eventually, we can close them.

[evnchn]

Actually this PR is more mergable if we incorporate https://github.com/evnchn/nicegui/tree/feature/lazy-imports-clipboard-class @falkoschindler can you review the changes there and then if it looks good we can merge that into this PR

[evnchn]

Crazily efficient 💯

Running:

from nicegui import ui


@ui.page('/')
def page():
    ui.highchart({'series': {'data': [1.0]}}, type='stockChart', extras=['stock'])


ui.run(show=False, reload=False)

[falkoschindler]

@evnchn What's the status of this PR?
As I just described in #5684, it's currently our best bet to improve the initialization of multiprocesses.

[evnchn]

@falkoschindler I chose to rebase because the merge conflict is unreal. Also nobody remember what happened those 98 days ago, so rebase over merge is more similar to how our memory works 🧠

Nevertheless, I want to report: Despite what the original description says, this PR is working!

test.py (frankly you can put whatever)

from nicegui import ui

@ui.page('/')
def page():
    ui.json_editor({})

ui.run(show=False, reload=False)

time_script.sh

#!/bin/bash

source /workspaces/nicegui/.venv/bin/activate

start=$(date +%s.%N)

coproc python test.py

while read -r line <&${COPROC[0]}; do
    echo "$line"
    if [[ "$line" == *"NiceGUI ready to go on"* ]]; then
        end=$(date +%s.%N)
        elapsed=$(awk "BEGIN {print $end - $start}")
        echo "Time taken: $elapsed seconds"
        kill $COPROC_PID
        break
    fi
done

• Before the PR: 0.6 seconds
• After this PR: 0.35 seconds

[evnchn]

Would consider that this is not draft. If you agree, you can mark it ready for review and review it.