Provide app.storage.client as a location to store volatile data which only matters for the current connection

[Alyxion]

As discussed in pull request #2811 and the discussion Add app.storage.tab or similar #1308 I added a feature which allows the storage of data per Client connection.

This - in practice - and when either implementing an own single page application approach, e.g. based upon the example/single_page_app or when #2811 is merged enables developers to create per-browser tab user data instances, for example with different logins and/or settings with the same comfort of as currently when using app.storage.user - so without handing over another per-page instance from event handler to handler.

[rodja]

Could you replace the double quotes with single quotes? See https://github.com/zauberzeug/nicegui/blob/main/CONTRIBUTING.md#single-vs-double-quotes

[rodja]

I just started a first proof-of-conecept for app.storage.tab in #2837.

[Alyxion]

Could you replace the double quotes with single quotes? See https://github.com/zauberzeug/nicegui/blob/main/CONTRIBUTING.md#single-vs-double-quotes

Done. Have read the CONTRIBUTING.md of course, after so many years of C++ and double quotes for everything except single character it though "hurts" as much as I would have to use a knife in my left and fork in my right hand ;-)... fixed the other branch/PR as well.

[Alyxion]

Added the missing checks + unit tests, removed Client.current_client and added the sections to the documentation page based upon your examples in the .tab variant.

[Alyxion]

Updated. I already referred #2837 in the documentation to explain when to use tab and when to use client. The count in It features four built-in storage types: needs to be adjusted accordingly when both PRs should be united in the main branch.

[Alyxion]

Resolved the conflicts with the new main branch and adjusted the doc accordingly.

[rodja]

I've just made minimal changes to the documentation. In markdown we like start each sentence in a new line.
@falkoschindler it would be great if you can have a look, too.

[falkoschindler]

I just reviewed the code and made a few minor changes. Most prominently I moved the tests into test_storage.py. Even though this file is already pretty long, we tend to move tests for file x.py into a file test_x.py. So all storage tests are in one place.

I've got three questions remaining:

1. • Is client.state a good name for this kind of storage? What do you think about client.storage? I'm undecided though. Since it is very volatile, it's somehow less than real storage. But it is implemented and documented as "storage".
2. • Why do we need to call get_slot_stack() in clear()? I guess it's for some important but slightly unintuitive reason.
3. • Why are there two buttons and two labels in test_client_storage?

[Alyxion]

Thanks @falkoschindler, regarding your questions:

1. Is client.state a good name for this kind of storage? What do you think about client.storage? I'm undecided though. Since it is very volatile, it's somehow less than real storage. But it is implemented and documented as "storage".

The reason why I chose state was actually 1:1 taken from Streamlit st.session_state which has exactly the same behaviour - "keep the state" as long as the user is connected. I think it's a bit philosophical what one interpretes into the words "state" and "storage", after all its also called Solid State Drive and not Solid Storage Drive, right? ;-)

2. Why do we need to call get_slot_stack() in clear()? I guess it's for some important but slightly unintuitive reason.

Otherwise app.reset crashes during the tests because it is also called w/o an available slot and then context.get_client would crash.

3. Why are there two buttons and two labels in test_client_storage?

Fixed, I assume through a git merge which duplicated those lines.

[falkoschindler]

2. Why do we need to call get_slot_stack() in clear()? I guess it's for some important but slightly unintuitive reason.

Otherwise app.reset crashes during the tests because it is also called w/o an available slot and then context.get_client would crash.

Ok, I've rewritten this part a little:

• I explicitly get the client and catch any runtime errors.
• I don't check for auto-index clients, because we wan't to clear all storage.
• We don't need getting the slot stack anymore, which was a bit confusing in my opinion.

[falkoschindler]

Code for testing all our storages:

@ui.page('/')
async def index():
    if 'time' not in app.storage.browser:
        app.storage.browser['time'] = time.time()
    ui.label(f'{app.storage.browser["time"]=}')

    if 'time' not in app.storage.client:
        app.storage.client['time'] = time.time()
    ui.label(f'{app.storage.client["time"]=}')

    if 'time' not in app.storage.user:
        app.storage.user['time'] = time.time()
    ui.label(f'{app.storage.user["time"]=}')

    if 'time' not in app.storage.general:
        app.storage.general['time'] = time.time()
    ui.label(f'{app.storage.general["time"]=}')

    await context.get_client().connected()

    if 'time' not in app.storage.tab:
        app.storage.tab['time'] = time.time()
    ui.label(f'{app.storage.tab["time"]=}')

[falkoschindler]

I think this is ready to merge now. Thanks everyone! 🙂