dreaming ideas for teaching / IPython / Thonny

[villares]

I was playing with IPython's "REPL", which is so nice, and thinking, how nice would it be if we could have IPython at Thonny's lower window slot? Would it be possible?

My harder to achieve dream, I'm afraid it might not be possible, would be to have non-blocking imported mode sketches running on Thonny and being able to interact with them in the console/REPL.

[hx2A]

I was playing with IPython's "REPL", which is so nice, and thinking, how nice would it be if we could have IPython at Thonny's lower window slot? Would it be possible?

My harder to achieve dream, I'm afraid it might not be possible, would be to have non-blocking imported mode sketches running on Thonny and being able to interact with them in the console/REPL.

I don't know how to do that in Thonny, but you can use IPython with py5 imported mode. The command is jupyter console, which is basically the same as (but not identical to) what you get with the ipython command. Then add the --kernel=py5 param to specify what kernel you want:

% jupyter console --kernel=py5                          
Jupyter console 6.6.3

Python 3.8.17 | packaged by conda-forge | (default, Jun 16 2023, 07:11:34) 
Type 'copyright', 'credits' or 'license' for more information
IPython 8.12.2 -- An enhanced Interactive Python. Type '?' for help.

py5 0.9.2.dev0 | py5 kernel 0.1.3a0 | A Python Jupyter kernel plus py5 in imported mode
In [1]

[villares]

this looks very promissing!
I might have to rebuild my conda env for py5...

[hx2A]

I don't know why that didn't run, but I would try rebuilding the conda environment and maybe install java with installjdk.

[villares]

UPDATE: ignore this... I forgot about python -m py5jupyter.kernels.py5.install --sys-prefix ... the error message could be more helpful I guess... No module named 'py5jupyter' makes one think one needs to install py5jupyter that is already installed...

Quick question, as I created a new conda env using conda env create -n py5coding -f http://py5coding.org/files/install/py5_environment.yml

Per this:

Should it install py5jupyter? It didn't work for me :(

[villares@L340 ~]$ conda activate py5coding
(py5coding) [villares@L340 ~]$ jupyter console --kernel=py5
/usr/bin/python: Error while finding module specification for 'py5jupyter.kernels.py5' (ModuleNotFoundError: No module named 'py5jupyter')

After pip install py5jupyter told me all requirements already satisfied, but still no luck.

[villares]

OHHH! I forgot about python -m py5jupyter.kernels.py5.install --sys-prefix

[villares]

This is about dreaming right? It`s about thinking of the most magic & wonderful tools possible without prejudice?

So, after installing py5jupyter could we type py5jupyter on the command line and have it install py5 kernels if needed (or ask to install the kernels if they are not installed) and then open the jupyter console --kernel=py5 for us?

Maybe better if it checked everything and launched jupyter lab instead... because there you can choose a notebook or a console session with the py5 kernel.

[villares]

how to check installed kernels https://stackoverflow.com/a/49299687/19771

[hx2A]

This is an interesting idea. Installing py5jupyter could create a command line utility called something like py5console which would install the py5 kernels (in the default locations) if they have not been installed already and then run jupyter console --kernel=py5. One problem (I think) is that jupyter console --kernel=py5 doesn't work at all on OSX. At least that is my recollection? I think it cannot open a sketch window. Conveniently, it works just fine on OSX.

One difference between ipython and jupyter console is how it handles print statements. A running Sketch with print statements (either the builtin print or py5.println) will display output immediately with ipython but will wait for you to enter a command with jupyter console.

[villares]

Oh, does the ipython console use the same jupyter kernels?

Could we have a py5lab sister command that opens jupyter lab checking if the py5 and py5bot kernels were installed?

No hurry/pressure... I'm still getting the hang of the interactive workflows.

[hx2A]

Oh, does the ipython console use the same jupyter kernels?

I suspect that ipython and jupyter were developed separately, and then the jupyter console is a fork of ipython that was adapted to fit in the jupyter framework. The ipython command has no ability (that I know of) to use a different kernel besides the default Python kernel. I wouldn't be surprised if one day ipython was deprecated in favor of jupyter console.

(this is a guess, but I have spent a lot of time reading the source code for jupyter and ipython)

Could we have a py5lab sister command that opens jupyter lab checking if the py5 and py5bot kernels were installed?

Why not? If py5console works, py5lab should work as well.

No hurry/pressure... I'm still getting the hang of the interactive workflows.

This won't be in the next release...right now I going to work on the existing open issues and then I have a lot of documentation to write. Between the color stuff and the shapely / trimesh features, this is already a release with a lot of stuff in it! Also, this change would have to go in py5jupyter, not py5. It is a separate Python library.

[hx2A]

About this, being careful of what you do outside the animation thread, I was very surprised today to learn that I can't call py5.save_frame(...) from my key_pressed() event function when using P3D (an OpenGL renderer)! Was that always so? Or is is it something new to Processing 4? Also I would say it is not intuitive to think of saving an image as a "drawing method"...

Hmmm, that error message is a bit confusing.

Here's what's happening:

py5's save_frame() is different from Processing's saveFrame(). py5 uses the PIL image library to save the image, letting you save the image in many formats and all the other nice features PIL has. py5 also lets you save to a BytesIO object. For these work, it needs to access the frame's pixels, calling Processing's updatePixels() method. That call must be triggering that error.

Processing does let you call saveFrame() from the mousePressed() event with the P3D renderer. I just tested it because I wasn't sure what would happen.

I don't think I can get py5 to allow saving from the mousePressed() event but might be able to come up with something. I do know I can improve the error message to be something that isn't confusing. If you want, open an issue and I'll think about this and see what I can come up with. I guess at a minimum py5 can check to see if save_frame() is being called from draw() or not and fall back to Processing's method instead. Or at least, that's the idea that I thought of while writing this paragraph. :) Yes, so open an issue if you think this is something that is worth working on. I won't do it for this release (this weekend) but perhaps for the next release.

[villares]

Let's think about this for the future... no hurry. But intercepting this message and providing a custom py5 one would be nice.

[hx2A]

In the meantime you can use py5_tools.screenshot(), which would work for you here. It returns PIL Image object, which you can then save.

https://py5coding.org/reference/py5tools_screenshot.html

py5 could use this when save_frame() is not called from draw() to get the frame, then save it.

I'm going to open an issue for this because I'll forget about the issue and my ideas for fixing it.

[villares]

I'll have a look at the screenshot tool! My naive solution would be just to make a global flag variable and check it to save the frame from inside draw().

[hx2A]

It happens that's exactly what the screenshot tool does. It adds a hook to your draw() function, attaching itself to it for one execution of draw() and then saves the frame.

[villares]

Another thing for people doing "interactive live coding" is somehow to simplify the sketch hot_reload_draw feature
https://py5coding.org/reference/sketch_hot_reload_draw.html?highlight=run_sketch
I'm not sure how it could be done (maybe a @hot_reload decorator for when you define a new draw?), but it is something I was thinking about and I'd like to leave here so I don't forget about it.

[hx2A]

Right now the signature for the hot_reload_draw() method is:

hot_reload_draw(
    draw: Callable,  # function to replace existing draw function
) -> None

I could easily make it so you don't have to pass the function. Instead it could also grab a function named draw() in the local namespace (which is how run_sketch() works...observe you don't have to explicitly pass setup() and draw() to run_sketch().) Your decorator idea is interesting also. In what environment would you use this? Jupyter?

[villares]

could be Jupyter, IPython, or even the Thonny console... but I haven't played enough with them yet....