-
Notifications
You must be signed in to change notification settings - Fork 8
[WIP] proof of concept: HTMLCanvas bitmap context in pyodide #115
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Conversation
feel like some stuff breaks due to the python name mangling when using double underscores in combination with the subclass... which happens quite a bit - so I am sorta surprised it still works this far. But I can't find any such references for Pyodide which would be odd if that is a known limitation. |
that wasn't the case. I actually fell through the line here rendercanvas/rendercanvas/base.py Lines 450 to 451 in fa7defc
I also registered the auto backend successfully - meaning if you build the wheel and then load it statically. The examples noise.py and snake.py work out of the box (although not events yet). But the weekend has a few more days :)
auto_demo.mp4E: turns out that wasn't true either and I am using the existing |
singular keydown event works... so more events and other types shouldn't be impossible. However I will get to that another day. snake_events.mp4 |
tried all day to make it work for the docs ... but either the .whl don't get included as static files or pyodide has trouble importing the wheel. I also wanted to automate the iframe inclusion with sphinx-gallery but seems like you need to either modify the classic "works on my machine", so have a video of what could have been instead: doc_embed.mp4 |
Even so it's awesome how far you managed to take this! |
pythonCode = ` | ||
# Use python script as normally | ||
from rendercanvas.auto import RenderCanvas, loop | ||
canvas = RenderCanvas(title="Example") |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I would argue that providing a constructor like this would be more conventional for the web:
canvas = RenderCanvas(title="Example") | |
canvas_el = document.getElementById("canvas") | |
canvas = RenderCanvas(canvas_el, title="Example") |
Since often there are multiple canvas elements on the page, users should be able to control which is used.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
One of the goal was to keep the python code portable between auto backends. So passing a string to __init__()
would work even on backends where this kwarg isn't used like glfw
and the user doesn't need to use any pyodide specific code in python. (Once we have a wgpu-py
version for browser, most examples should just work without changes to shadertoy, pygfx or fastplotlib etc).
I also losely followed the idea of https://pyodide.org/en/stable/usage/sdl.html#setting-canvas where they provide a specific API to accessing the canvas, although I not using it.
Maybe I can write a little multi canvas example to see if my approach works.
I have zero webdev experience, so my design decisions are directed to the python devs wanting to write their python code (like myself).
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have zero webdev experience, so my design decisions are directed to the python devs wanting to write their python code (like myself).
I hear you, but just because you can use python the language, doesn't mean you can "ignore" the environment it's running in! I don't mind what kind of API you choose (I value portability as well) as long as the user can control which <canvas>
is used.
I imagine python devs turning to browsers will often do so because they want to use the browser's capabilities to build the UI they have in mind. It's easy to envision applications with multiple canvases embedded in a richer UI.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
it shouldn't be impossible to support both.
canvas_el: [str|HTMLCanvasElement] = "canvas"
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
would work even on backends where this kwarg isn't used like
glfw
apparently kwargs don't get ignored when a different auto backend is selected because the base class calls super.__init__(*args, *kwargs)
. We could use the title arg as I am not sure if that has a use in the browser, but that seems janky.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
For resizing, have a look at window.ResizeObserve
: I use this code in another project to get the physical size (it's PScript, so you need to convert to Python/JS depending on where it runs):
# Inside the ResizeObserve callback ...
entry = entries.find(lambda entry: entry.target is self.node)
if entry.devicePixelContentBoxSize:
# Best if we have the physical pixels ...
psize = [
entry.devicePixelContentBoxSize[0].inlineSize,
entry.devicePixelContentBoxSize[0].blockSize,
]
else:
# ... but not all browsers support that (see issue #423) ...
if entry.contentBoxSize:
lsize = [
entry.contentBoxSize[0].inlineSize,
entry.contentBoxSize[0].blockSize,
]
else: # even more backward compat
lsize = [entry.contentRect.width, entry.contentRect.height]
ratio = get_pixel_ratio()
psize = Math.floor(lsize[0] * ratio), Math.floor(lsize[1] * ratio)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
In general the standard reference material for browser APIs is on MDN, in this case see this page for example usage: https://developer.mozilla.org/en-US/docs/Web/API/ResizeObserver
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
finally found a bit of time to implement this - also tried to make the demo page react and it seems to work. Altough I am not sure if this matches real webframeworks
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm pretty sure we can hook things up in a clean way, but I will need to sit down and play a bit to get it right.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
feel free to commit into this branch or similar if you have the capacity. I am very much out of my depth with the web stuff and also don't have too much time currently (settling into new job and new university) to sit down for some long evenings and figure it out.
rendercanvas/html.py
Outdated
from pyodide.ffi import run_sync, create_proxy | ||
from js import document, ImageData, Uint8ClampedArray, window | ||
|
||
# TODO event loop for js? https://rendercanvas.readthedocs.io/stable/backendapi.html#rendercanvas.stub.StubLoop |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think asyncio can be used in Pyodide, right?
edit: should have read more before reacting. So is pyodide.webloop a sort of more native loop implementation?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
yes, they overwrite a couple of functions.
But I should try to use the async loop class and see if this whole class can be avoided.
edit: you can't find much on it in the docs so I looked at the source and went from there. Perhaps other prominent pyodide packages provide some specific insight. I pretty much learned all my async and js knowledge from attempting this. Surprised myself it sorta works. But this obviously means I might be doing something wrong.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can I ask why a loop is needed at all? Why not just implement call_later with setTimeout and leave everything else to the browser's native loop?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Loop is made to fit into the existing rendercanvas implementation, although it can likely be simpler with more changes. Very much out of my depth here.
The docstrings here says as much, although I am not sure if that's happening all the way down. https://github.com/pyodide/pyodide/blob/bcd0235bff5351d1dc383e4f3e34b9917fdf3281/src/py/pyodide/webloop.py#L280L294
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The point is that in the browser there already is an event loop driving the page, and you can not get a reference to it. So I am just wondering what the point of having an event loop abstraction is inside a browser, apart from compatibility.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
okay, so I tried with the asyncio loop and it works just fine. Also not calling loop.run()
at the end of an example still works.
so should I drop my loop class and just import the asyncio loop instead?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
so should I drop my loop class and just import the asyncio loop instead?
Yes, let's just do that. It's the simplest to do, and makes sure all works together if the user uses asyncio for other stuff.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
one more consideration that might be complex to implement: use the browsers requestAnimationFrame
with a callback to run the loop? I spotted something like this here: https://github.com/pyglet/pyglet/blob/5d1fe49db83bd2c060d97226e1334f06a706b1fc/pyglet/app/async_app.py#L64
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Using requestAnimationFrame
is the standard approach for rendering, since it automatically pauses when the tab becomes hidden for example.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
yeah, with some of the wgpu examples I am having issues where the animation stops randomly or when you change tab etc... it just suspends and doesn't recover. So we might need a webloop or whole sheduler... the rendercanvas bitmap examples seems to survive tab switching
return res | ||
|
||
|
||
self._pointer_inside = False # keep track for the pointer_move event |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
You might want to look into:
- Pointer capturing
- Mouse leave and enter events
Maybe you already have and this comment is entirely moot
# char ... it's not this | ||
# def _html_char(proxy_args): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Have a look at the implementation of char events in jupyter_rfb vispy/jupyter_rfb#119
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I looked at that initially, I tried to avoid actually changing the webpage, but maybe there is no way around it. Which also opens new questions like should the python code spawn new canvases?
Also all the global events might mess up when there are multiple canvases...
#38
two evenings of tinkering but I can feel a bit of progress. Rendercanvas looks very web inspired, so I am reading a lot of things between the docs, pyodide docs, pyodide source that sound close but are ever so slightly different. Plus I don't have any webdev experience, it's more like a learning opportunity.
for those that want to give it a try - you can essentially just load the .html as a static page with the python script inserted. But you will need to build wheels and load them locally. Installing from pypi via micropip doesn't include these changes.
some todos:
message
events from the browser (for example to close?)