#circuitpython-dev
manic glacierBOT
runic loom
Here is a detailed writeup of what I did : https://ashishware.com/2026/01/10/cnn_circuitpython/
Here is link to repository containing required files: https://github.com/code2k13/neural_network_cp
The model is tiny:
Total params: 3,116 (12.17 KB)
Here is link to Kaggle notebook used to train and export model to onnx:
https://www.kaggle.com/code/finalepoch/digit-classifier-to-onnx
It is the same model which I handcoded using ulab and CircuitPython in my below experiment:
https://ashishware.com/2024/06/29/pipico_digit_classification_cnn/
tulip sleet
The model is tiny:
Total params: 3,116 (12.17 KB)
Here is link to Kaggle noteb...
a model will have to be compiled in with circuitpython source code.
So the model could be loaded into RAM at runtime, it sounds like.
slender iron
The model is tiny:
Total params: 3,116 (12.17 KB)
Here is link to Kaggle noteb...
Thanks! Want to send these links to cpnews@adafruit.com or should I?
I have zephyr setup to rotate through four smaller buffers. The filling thread will wait for the next free buffer once all four are full.
tulip sleet
I have zephyr setup to rotate through four smaller buffers. The filling thread w...
did you see that zephyr has support in I2S for ping-pong buffers? https://zephyr-docs.listenai.com/reference/audio/i2s.html#c.I2S_OPT_PINGPONG
2.7.0-rc.2
slender iron
did you see that zephyr has support in I2S for ping-pong buffers? <https://zephy...
I saw that but if you search for it, then you'll see many implementations don't support it
the memory slab 4 buffer thing will be ok
as long as I can get nrfx to not try and free an old buffer
runic loom
Thanks! Want to send these links to cpnews@adafruit.com or should I?
Feel free to send them
runic loom
> a model will have to be compiled in with circuitpython source code.
So the mod...
Thats right. Model is loaded into RAM. Its part of CircuitPython
tulip sleet
Thats right. Model is loaded into RAM. Its part of CircuitPython
is the model loaded from a file, or is it compiled into the firmware image?
runic loom
is the model loaded from a file, or is it compiled into the firmware image?
The model is compiled in firmware image.
tulip sleet
you could load it from a file instead, then, so you could switch models?
load it into an array.array
runic loom
You mean the weights of model or the model itself. I mean there are two parts to this. The model structure, that dictates what operations have to be perfromed and the weights used for performing these operations. Weights can be loaded from a file, thats lot easier to achieve
tulip sleet
I assumed it was all data (weights), but you're saying there's custom compiled code for the model?
or is it pointers?
pointers could be converted to relative indices
runic loom
yes, the compiled code contains everything like a tar ball, the structure of model (like how many layers, what type of layers) and functions that support inference. A different model will have totally different compiled code.
Weights are also in the compiled code , like constants
tulip sleet
i see. If the compiled code (functions) are position-independent, then for any particular architecture (M0, M33, M4) they could be loaded at runtime. But that's a whole 'nother structural element to implement
runic loom
They could be made position-independent say if we were to use the same model (Structure). Like in my case the model say has 2 hidden convolution layers and input is 30x30 image and output is array of 10 probabilities. A different model will have different structure, hence computations will be totally different (not just the numbers). However, if let us say we decide to have a generic, well known model like yolo8 for example as part of CircuitPython, then I feel it would make sense to enable loading weights into a fixed model structure (computations and their sequence remains unchanged)
tulip sleet
got it, thank you, I thought it was more parameterized (at run-time) than it is.
runic loom
I wish it was, we could have add a module called neural network in CircuitPython and load the model and weights using some kind of JSON. This can be done to some extend for 'dense neural newtworks' because underlying computational pattern is repeatitive and easy. But for some more exotic types of models like CNN or RNN, this get's tricky. Not saying it cant be done, but quite a bit of work and the layer that does runtime parsing might actually be heavy.
Before I log off, you people are doing awesome work. Thank you for your work on CircuitPython
tulip sleet
you're welcome, and thanks for your more thorough knowledge than my casual surmises.
manic glacierBOT
The back of the box has GT911 too. I'm trying to determine it without opening the box so I can return it.
The box says GT911, but the back of the system says ST7123.
tiny peak
@slender iron For the Zephyr port, where are things at with how you're implementing USB support? Are you still using TinyUSB, or has that changed? Based on your comments over on the baochip server's not-rust channel, I got the impression that when you mentioned implementing USB support, you meant for Zephyr APIs, not for TinyUSB.
Context for the question being, I could potentially start looking at that, but I'd want to make sure I wasn't implementing the wrong API.
slender iron
<@252717193496756235> For the Zephyr port, where are things at with how you're i...
correct. the zephyr port uses the zephyr usb device stack now
that way it just works
we only need usb going after getting the cpu and uart going though
usb can wait
tiny peak
we only need usb going after getting the cpu and uart going though
ah, okay. Supposing it's possible to wrap Rust no_std device drivers from the baochip bootloader in a C FFI, then connect those up to the relevant Zephyr APIs with a C shim layer, would you have any objection in principle to adding a dependency on Rust tooling? I'm not certain that's actually possible, but it might be.
Otherwise, it would be a matter of translating the drivers to C, I guess.
slender iron
ah, okay. Supposing it's possible to wrap Rust no_std device drivers from the ba...
that's a question for zephyr, not me
I'd just do C
tiny peak
Are you assuming that Adafruit will end up attempting to upstream your work or probably just maintain your own fork? Wondering how important it would be to follow all their policies and such. That could be a bit burdensome.
tiny peak
Ah, okay. That does complicate things a bit. They have a strict policy on commit signoffs with real names and live emails. among other things
I was thinking that with the way they have the build system set up, it might be pretty easy to just keep some board definitions in your own repo if you wanted, then do builds in a workspace with all the regular upstream drivers.
Then you'd only need to do the PR bureaucracy shepherding thing if you wanted to upstream driver patches. From reading the discussions in their PRs, getting code merged is typically a serious ordeal. I don't know how people would be willing to do that if they weren't getting a salary for it.
Worst case, I suppose I could just figure out techniques that work and somebody could redo it, or we could transfer copyright, if you wanted to upstream that stuff.
For the moment, I could probably start working on a stand-alone repo with a board def in the style of the board defs I made last year for some of the Feather S3 stuff. That would work with vanilla Zephyr, and it could probably also be used with your port.
tiny peak
oh, that's good news then!
slender iron
not sure about peripheral drivers though
tiny peak
I think all of that is set up to work optionally out of tree if you want it to be
I can definitely look into that
tiny peak
okay. sounds good. I just don't want to get going down a road that will end up making things more complicated for you rather than easier.
runic loom
got it, thank you, I thought it was more parameterized (at run-time) than it is.
@tulip sleet your advice got me thinking. I am trying to write a reusable neural network library entirely in C. The idea is to support a tiny subset of operations needed by dense neural networks and Convolutional Neural Networks. People should be able to create tiny neural networks in Tessorflow/Pytorch, use a custom tool to convert them to *.bin format, and the library would load them. I will try to see how far I go with this idea. Will try this on Pi Pico and ESP S3 boards and come back . Thanks !
tulip sleet
<@329766224093249548> your advice got me thinking. I am trying to write a reusa...
Interesting idea! I was wondering if there was an issue about the model not fitting into RAM but it could fit into flash. I think you would need it to be contiguous so you use the flash mapping into the address space and treat it as just read-only memory., as you are now.
MicroPython created a special kind of filesystem to allow loading things into contiguous regions of flash. The FAT filesystem we are using does not guarantee that, and because the microcontrollers do not have page tables to allow us to map file pages into memory regions at low granularity, we can't do that like regular OS's (mmap, etc.)
on Espressif or other boards with large PSRAM (some RP2350 boards) it could be copied into contiguous PSRAM from the filesystem.
scarlet fjord
i know this is ancient, but im working again on a joystick project and wondering where the gamepad/joystick docs went? this link doesnt seem to work any more
or if theres anywhere else that has docs for joystick inputs
jaunty juniper
i know this is ancient, but im working again on a joystick project and wondering...
hi, the gamepad code was moved into the examples
https://docs.circuitpython.org/projects/hid/en/latest/examples.html#simple-gamepad
it is discussed here:
https://learn.adafruit.com/custom-hid-devices-in-circuitpython/report-descriptors
I made a separate library along with another example that has a hat, 2 joysticks and 32 buttons
https://github.com/Neradoc/circuitpython_hid_gamepad/
lone axle
<@&356864093652516868> The weekly meeting will be held here on Discord at the usual time, about 90 minutes from now at 2pm Eastern / 11am Pacific. Feel free to add hug reports and status updates to the notes doc ahead of time https://docs.google.com/document/d/1Kc6iC6-aOjE3SaAg9BfiK8xVNjm9YW0BBSvjk1jHqU8/edit?usp=sharing. We look forward to hearing from all that can attend.
slender iron
EPD and LSM6DS don't have semver version numbers. I tried rereleasing them but it doesn't look like the bundle picked up on it.
turbid radish
Thanks folks for the likes on my mini-IBM Fruit Jam project
slender iron
next week is on Tuesday
turbid radish
Lots in the newsletter due to the holiday break
tulip sleet
also group hug, checking mic
turbid radish
Cool work
slender iron
tidal kiln
👋
turbid radish
Great meeting @lone axle
lone axle
Thanks everyone, have a great week 👋
errant grail
Thanks for hosting, Tim!
lone axle
Here is the notes document for next Monday’s CircuitPython Weekly Meeting. It is at the normal time of 11am Pacific / 2pm US Eastern here on Discord. Add your hug reports and status updates to the document before the meeting. If you are unable to attend but would still like to contribute, feel free to add your notes and we’ll read them off during the meeting. Hope to see you there! <@&356864093652516868> https://docs.google.com/document/d/1knBYDjHI1NDehhhJfRufXqgP66RPJFA-LNYMTd1-dVA/edit?usp=sharing
candid sun
^^ meant to pr my fork, not main
slender iron
^^ meant to pr my fork, not main
my branch is here: https://github.com/tannewt/circuitpython/tree/yoto please PR to it. I haven't updated all of the pins
it has mmc working though
candid sun
my branch is here: https://github.com/tannewt/circuitpython/tree/yoto please PR ...
will do! yesterday i was just getting a build going for testing. now i will work from your branch
slender iron
great! I'm thinking about how to get the display going
manic glacierBOT
tender ginkgo
Out of curiosity, are there any plans to port Circuitpython to a TRMNL X (ESP32-S3 for 2.4Ghz, plus ESP32-C5 for 5Ghz, 10.3" e-ink display, high density, 1872x1404 resolution, with 16 grayscale. Full refresh <= 1.2s, partial refresh <=200ms) or TRMNL OG (ESP32-C3 based, smaller)
plucky tulip
Out of curiosity, are there any plans to port Circuitpython to a TRMNL X (ESP32-...
I may be wrong, but if it has a ESP32-S3 or C3, it should be possible to install CircuitPython on it. If I understand correctly, the port is needed for pin definitions etc. maybe for easy access to the e-ink display without manually defining it in code, but overall I think it should be possible to start tinkering with it even now.
stuck elbow
the drivers for the display may be the main problem
slender iron
Out of curiosity, are there any plans to port Circuitpython to a TRMNL X (ESP32-...
No plans. Happy to give tips on how to do it though. This is the place to start: learn.adafruit.com/building-circuitpython
we don't currently support external ESP-C5 afaik. that'd be trickier. the display is an unknown as well
manic glacierBOT
short tendon
Random question on preferences. I've been doing a lot of playing around with sensors, and noticed that sensors in a group (let's say pressure), don't always do things the same way, and there's a lot of duplicated code. If I were building these from scratch, I'd create base classes that did the common stuff and import from that. Obviously in CircuitPython that becomes another mpy file, but allows new features to be added globally (for example globally adding more accurate altitude when incorporating temperature).
Is this unified approach something Adafruit would be interested in?
slender iron
Random question on preferences. I've been doing a lot of playing around with sen...
I don't think I'd do it as a base class, instead I'd do composition and pass in the raw pressure sensor.
but it does depend on what is duplicated
short tendon
I don't think I'd do it as a base class, instead I'd do composition and pass in ...
That way would also work (ironically how I'm currently doing it since the sensors are already written). But it felt more complicated to update, because then in theory you would remove altitude from all of the libs and require a new one to get it.
The duplication is a little all over. Here's an example: https://github.com/search?q=org%3Aadafruit+0.190+language%3APython&type=code&l=Python
slender iron
meh, I don't really want to add a dep just to have it shared
short tendon
I've created a whole unified sensor lib, that takes in all the sensors I have and then can output all sorts of things.
short tendon
meh, I don't really want to add a dep just to have it shared
Fair
tidal kiln
lib bundle build seems broken
jan 16 zips both 9.x and 10.x have:
adafruit-circuitpython-bundle-10.x-mpy-20260116.zip
lone axle
jan 16 zips both 9.x and 10.x have:
It came without the lib/ folder?
tidal kiln
yep
lone axle
How big is the total zip size? I downloaded it just now from the buttons here: https://circuitpython.org/libraries and I seem to have all of th expected files.
tidal kiln
wait. it's something local.
looking...
ok. the zip content issue is something going on locally. so its me. ignore.
@lone axle can you check something else though - what version of Adafruit_CircuitPython_Display_Shapes are you seeing in latest bundle?
lone axle
<@382939733107408897> can you check something else though - what version of `Ada...
2.10.4
tidal kiln
thanks. gonna reboot. im stuck in the weirdness multiverse. brb.
tidal kiln
@lone axle sry, wrong lib, what version do u see for adafruit_display_text
lone axle
<@382939733107408897> sry, wrong lib, what version do u see for `adafruit_displa...
4.0.2. but bear in mind though it is frozen in for lots of devices so in order to use the bundle one instead of the frozen one you'd have to paste it in the root of CIRCUITPY instead of lib/
tidal kiln
ah. ok. thanks. let me check for that.
ok, looks frozen for magtag
empty lib folder
Adafruit CircuitPython 10.0.3 on 2025-10-17; Adafruit MagTag with ESP32S2
>>> import adafruit_display_text
>>> adafruit_display_text.__version__
'3.3.3'
>>>
>>> import sys
>>> sys.path
['', '/', '.frozen', '/lib']
frozen shadows lib
@lone axle how did you get around this for the mag lit clock?
https://learn.adafruit.com/eink-literary-quotes-clock-for-magtag/code
the color_palette kwarg is not in 3.3.3 for Label
lone axle
<@382939733107408897> how did you get around this for the mag lit clock?
<https:...
I had to paste the latest release of adafruit_display_text in the root of CIRCUITPY. I thought I had noted that in the guide but looking now I do not see it.
that project was the impetus for some new functionality and ultimately some refactoring to make BitmapLabel more flexible and reduce the need for having so many different types of labels.
tidal kiln
thanks. im looking at guide too...
yah, and not seeing anything.
but ok, that generally explains things. i think/hope 🙂
lone axle
The version of display text downloaded by the project bundle button should be the latest, but it does need to be placed in the root instead of lib where it is in the bundle zip.
I'll add an alert to the guide shortly.
tidal kiln
thanks. yep. that's the general issue.
gotta work around the frozen version
the screen cap also shows it sitting in lib
lone axle
Yeah, I'm going to have to tinker a little more to get a screenshot of it the right way. It's called out with alerts on on this page now https://learn.adafruit.com/eink-literary-quotes-clock-for-magtag/code
tidal kiln
cool. thanks! glad you were around for this 🙂
lone axle
okay, the screenshot is updated now too. I'll need to remember to swap that back to linking the one from github once the next release of cp core is made.
manic glacierBOT
main furnace
I'm working on an MCP to allow Claude Desktop or any other agent with MCP support to read and write code on an attached CIRCUITPY drive. It also supports web workflow.
lone sandalBOT
lone sandalBOT
vague thicket
Hello, I´m trying to update the definition of my board (es3ink) as I want to use a different ESP32-S3 module (N4R2 instead of N16R8). I´m building CPY fw locally in my VM, and I have 2 issues with it.
I can build UF2 file, but when I put my board in bootloader mode, and I copy the fw file to the board, when it finishes uploading, it does not auto-restart as it should. Then, when I restart it manually, it jumps to the reboot loop, and the CPY drive does not show up.
I´m attaching the log file from this loop. I tried to upload the UF2 file with the latest fw from adafruit_feather_esp32s3_4mbflash_2mbpsram, which has the same module, and this fw is working fine. I tried to find a solution to my problem, but I did not find any here or online. Any help is really appreciated.
tulip sleet
Hello, I´m trying to update the definition of my board (es3ink) as I want to use...
have you compared the board files for adafruit_feather_esp32s3_4mbflash_2mbpsram with the changes you made locally for the es3ink?
small cypress
@slender iron Hey, hope you don't mind me asking here, but I was wondering if this being closed means the M5Stack Tab5 is fully supported in CP now?
lone sandalBOT
manic glacierBOT
manic glacierBOT
manic glacierBOT
tulip sleet
<@&356864093652516868> in about 1-1/4 hours, at 2pm US ET and 11 am US PT, we'll have our weekly live CircuitPython meeting (delayed by one day due to a US holiday). See the top pinned message in this channel. We'd love to have you add our notes! https://docs.google.com/document/d/1knBYDjHI1NDehhhJfRufXqgP66RPJFA-LNYMTd1-dVA/edit?usp=sharing
thorny jay
<@&356864093652516868> in about 1-1/4 hours, at 2pm US ET and 11 am US PT, we'll...
I was thinking that I missed yesterday's meeting... and now I am going to miss the real date. 🙁
tulip sleet
I was thinking that I missed yesterday's meeting... and now I am going to miss t...
you can always watch the recording 🙂
slender iron
@tulip sleet I'm home with Rynn. I'm going to try to make the meeting but if I don't respond then just read stuff off
tulip sleet
<@329766224093249548> I'm home with Rynn. I'm going to try to make the meeting b...
sure, np
thorny jay
you can always watch the recording 🙂
Wrote something quick from the phone...
Still searching for the miss PacMan repo and trick to enable, as seen on show and tell.
lone axle
cookiecutter I think.
slender iron
thanks all!
lone axle
Thanks for hosting Dan, have a great week everyone.
tulip sleet
Here is the notes document for next Monday’s CircuitPython Weekly Meeting. It is at the normal time of 11am Pacific / 2pm US Eastern here on Discord. Add your hug reports and status updates to the document before the meeting. If you are unable to attend but would still like to contribute, feel free to add your notes and we’ll read them off during the meeting. Hope to see you there! <@&356864093652516868> https://docs.google.com/document/d/1Db6h50frFZKWA3bCOdiDdZEkdtajotgL1SWobppy938/edit?usp=sharing
slender iron
<@252717193496756235> Hey, hope you don't mind me asking here, but I was wonderi...
I wouldn't say it is fully supported. I got the old display working but don't have the new one to test with. The core should be able to support everything. I haven't written drivers for everything though.
slender iron
Hello, I´m trying to update the definition of my board (es3ink) as I want to use...
what bootloader are you using?
slender iron
I'm working on an MCP to allow Claude Desktop or any other agent with MCP suppor...
interesting! I think there may be a command line tool you could use too if you want.
slender iron
@tulip sleet how are you capturing the additional socket requirements you are finding? could we add them as tests?
tulip sleet
I was planning to write some tests. Right now they are error conditions to be detected.
There is not necessarily consistency between what the POSIX socket API does and what CPython does, and it varies across OS's also.
I should see whether the POSIX socket spec describes this kind of thing
the Python doc does, to some extent (you can't bind() twice, for instance), but it's kind of vague in other ways. And since sockets cover a lot of protocols, it's not that specific
slender iron
It'd be nice to have it for CP anyway to make it consistent across impls
vague thicket
what bootloader are you using?
@tulip sleet @slender iron thank you for your help, but I did find out that I forget to make "make clean BOARD=" when I made changes, so my issue was that I was still using same config for the board from old config. This is solved.
tulip sleet
<@329766224093249548> <@252717193496756235> thank you for your help, but I did f...
it's happened to all of us 🙂
manic glacierBOT
manic glacierBOT
lone sandalBOT
manic glacierBOT
manic glacierBOT
manic glacierBOT
candid sun
<@&356864093652516868> We'll have our weekly meeting in about 90 minutes from now in this text channel and in the circuitpython voice channel. Please take the time to add your notes in advance to the document: https://docs.google.com/document/d/1Db6h50frFZKWA3bCOdiDdZEkdtajotgL1SWobppy938/edit?tab=t.0 -- I look forward to everyone's updates!
turbid radish
Folks, I'm looking for ways I can increase the number of subscribers for the Python on Microcontrollers Newsletter.
tulip sleet
just read mine
wraith crow
Thanks all 👋
lone axle
Thanks for hosting Liz, have a great week everyone 👋
slender iron
Thanks Liz!
candid sun
thanks folks!
Here is the notes document for next Monday’s CircuitPython Weekly Meeting. It is at the normal time of 11am Pacific / 2pm US Eastern here on Discord. Add your hug reports and status updates to the document before the meeting. If you are unable to attend but would still like to contribute, feel free to add your notes and we’ll read them off during the meeting. Hope to see you there! <@&356864093652516868>
https://docs.google.com/document/d/1wMpXc_QzzngY0Nk-AnDDyDzIqjy_pXXW76yKbKys838/edit?tab=t.0
manic glacierBOT
serene token
In Framebuffer_RP2350.c, there is the following line variable declaration:
int8_t pins[8] = {
clk_dp->number, clk_dn->number,
red_dp->number, red_dn->number,
green_dp->number, green_dn->number,
blue_dp->number, blue_dn->number
};
The various structs referenced here are passed into the function (common_hal_picodvi_framebuffer_construct()). I need to figure out what the actual pin numbers are (in Pico-SDK terms), not Arduino or CP).
Oh, I almost forgot: For the Fruit Jam specifically.
serene token
they are passed as arguments https://github.com/adafruit/circuitpython/blob/8c0d...
I got that part. I'm trying to figure out what those arguments are for the Fruit Jam. Each step I try to go back is just more symbols with their actual definitions somewhere else I can't find.
stuck elbow
for the fruit jam, there are default values: https://github.com/adafruit/circuitpython/blob/8c0d59693ede2ef8c144a5a1d54f1727433f4040/ports/raspberrypi/boards/adafruit_fruit_jam/mpconfigboard.h#L27-L34
serene token
That's definitely closer! Now, is "pin_GPIO12" pin 12 in Pico SDK terms or in CP terms?
Wait, got it. Pins 12 through 19 are the pin numbers for HSTX in Pico SDK terms, so these must be in Pico SDK terms. Awesome. That should do it. Thanks!
stuck elbow
GPIO12 is the pin labeled "12" on the pi pico
serene token
Right. My problem was figuring out the order, because the HSTX peripheral uses a different pin order on some boards than others, and I couldn't find the Fruit Jam pin order. Thanks again!
inner orbit
I am compiling CP for Trinket M0, removing some built-in libs I don't need. What I want to do is include some CONST in the build to save RAM usage. Google AI says this and that, but not specifically where to add the definitions. Are there docs relating to this process anywhere on the 'net.
slender iron
I am compiling CP for Trinket M0, removing some built-in libs I don't need. Wha...
what are you trying to include into the flash? a python library?
inner orbit
what are you trying to include into the flash? a python library?
just some variables I can access later in the code.py like... homer = 100. or dict = {"foo": "bar"}
Which I guess could be a python lib. just not much going on in it except decarations.
I am trying to figure out where to put said declarations so they compile correctly.
slender iron
I am trying to figure out where to put said declarations so they compile correct...
why do you need that in the firmware instead of the filesystem?
inner orbit
Save RAM space. If I declare the variable in code.py, for example, it will push it into RAM. I would like to compile it into the firmware, as I have space there, to conserve the RAM. If it is not possible, I accept that. Kind of like how PROGMEM works in Arduino.
inner orbit
Thanks for the help on my question. I think I found out how to integrate my own modules into my CP build.
slender iron
Thanks for the help on my question. I think I found out how to integrate my own...
Nice! You'll want to "freeze" python into the build then
slender iron
heh, I mistype codespell in my commit message
manic glacierBOT
tulip sleet
There is GitHub CI build problem for Espressif builds which has reappeared as the result of a new version of esptool. https://github.com/espressif/esptool/issues/1107
manic glacierBOT
tulip sleet
@slender iron in the zephyr port, in build_circuitpyhon.py, these are defined but are not defined as 1 or as the value of some variable. Is that intentional?
(trying to fix some build errors in my airlift PR)
circuitpython_flags.append("-DINTERNAL_FLASH_FILESYSTEM")
circuitpython_flags.append("-DLONGINT_IMPL_MPZ")
circuitpython_flags.append("-DCIRCUITPY_SSL_MBEDTLS")
Nope, you can change them
tulip sleet
where is the master list of the .toml variables? That is what I think I am looking for?
slender iron
I'm not sure what you mean by toml variables. Those are ones that would traditionally be in mpconfig.mk
tulip sleet
right, so like how do it create ``./boards/nordic/nrf7002dk/autogen_board_info.toml`? How does make know what to set those values to?
slender iron
that is created when you do a zephyr build
tulip sleet
is there a default .toml file? Where are the variations kept, in circuitpyt.toml?
slender iron
but there is an update script in cptools
tulip sleet
how does it know which ones to set to which values?
slender iron
zephyr2cp.py determines what modules to enable based on the zephyr device tree
the toml files are for the doc build so it doesn't need all of zephyr to determine it
tulip sleet
ok, I'll dive down on those, thanks
slender iron
what are you trying to fix?
tulip sleet
i added CIRCUITPY_SSL_NATIVE, which determines CIRCUITPY_SSL_MBEDTLS on some ports (e.g. espressif).
so on boards that used to say CIRCUITPY_SSL = 1, now they say CIRCUITPY_SSL_NATIVE = 1
(except for airlift-capable boards)
there's a CIRCUITPY_SSL_AIRLIFT
tulip sleet
so normally a native wifi board would set
CIRCUITPY_WIFI_NATIVE = 1
CIRCUITPY_SSL_NATIVE = 1
CIRCUITPY_SOCKETPOOL_NATIVE = 1
now (in my PR)
so need to do the same for zephyr
there is still a CIRCUITPY_SSL which gets set when any of CIRCUITPY_SSL_* gets set, etc.
slender iron
I think CIRCUITPY_SSL is set in build_circuitpython.py
but maybe it is in zephyr2cp
tulip sleet
there's a lot of logic in the .mk files to cascade these. circuitpython_flags.append("-DCIRCUITPY_SSL_MBEDTLS")
is the only mention of SSL in either file
slender iron
CIRCUITPY_<module name> is automatically added
tulip sleet
like now I have
ports/espressif/mpconfigport.mk
53:CIRCUITPY_SSL_MBEDTLS = $(CIRCUITPY_SSL_NATIVE)
no, it's still there, but now it's cascaded off the others, and not set explicitly for a particular board or port
tulip sleet
because there are still plenty of conditional compilation guards for SSL in general
ok, I see EXTRA FLAGS
i see, this is fairly incomplete, because it's just what you needed to get those particular boards to build
slender iron
That's where you can add the other extra flags
yup, but all module flags are implicit rather than explicit like they are the mk
tulip sleet
so I will try this:
EXTRA_FLAGS = {
"busio": ["BUSIO_SPI", "BUSIO_I2C"],
"ssl": ["SSL_NATIVE", "SSL_MBEDTLS"],
}
and remove the other mention
slender iron
👍
tulip sleet
probably need similar for socketpool and wifi
tulip sleet
@slender iron further on mpconfig stuff and zephyr:
In py/circuitpy_mpconfig.mk we have:
CIRCUITPY_IPADDRESS ?= $(CIRCUITPY_WIFI)
In the zephyr nrf7002dk/autogen_board_info.toml we have:
ipaddress = false
because it's not turned on explicitly anywhere.
So should IPADDRESS be in EXTRA_FLAGS or as yet another
circuitpython_flags.append(...)
? There is a lot of logic in circuitpy_mpconfig.mk that we have to put somewhere.
tulip sleet
if board_info["wifi"]:
enabled_modules.add("wifi")
module_reasons["wifi"] = "Zephyr board has wifi"
ok, I see
i will .add ipaddress
slender iron
module_reasons ends up as the comment in the toml
tulip sleet
do i have to do a clean build after modifying build_circuitpython.py?
slender iron
I don't think so
tulip sleet
i have one more issue where some routines in common-hal Socket.o are not being linked, despite being in the .o. They are referenced in a static routine in shared-bindings Socket.c. This is like not enough passes through the .a or .o files
years ago I saw something like this when files were in the wrong order in the linker line
slender iron
I thought we grouped them all together anyway
tulip sleet
that's what I thought, i have not seen something like this
there are other routines in the same .o that are linked just fine
and they are already declared in a .h that was included
/home/halbert/zephyr-sdk-0.17.4/arm-zephyr-eabi/bin/../lib/gcc/arm-zephyr-eabi/12.2.0/../../../../arm-zephyr-eabi/bin/ld.bfd: libcircuitpython.a(Socket.o): in function `socket_ioctl':
/home/halbert/repos/circuitpython/shared-bindings/socketpool/Socket.c:458: undefined reference to `common_hal_socketpool_socket_readable'
/home/halbert/zephyr-sdk-0.17.4/arm-zephyr-eabi/bin/../lib/gcc/arm-zephyr-eabi/12.2.0/../../../../arm-zephyr-eabi/bin/ld.bfd: /home/halbert/repos/circuitpython/shared-bindings/socketpool/Socket.c:461: undefined reference to `common_hal_socketpool_socket_writable'
since this USED to work, it's something that I have introduced in some way in the PR
slender iron
lemme look at your changes
tulip sleet
i pushed the latest
slender iron
you renamed the functions but didn't in the zephyr port
common_hal_socketpool_readable -> common_hal_socketpool_socket_readable
tulip sleet
i knew that -- i thought I had checked it 🙂
thanks for your eyes on that
now compiles 🙂
manic glacierBOT
orchid basinBOT
manic glacierBOT
orchid basinBOT
tidal kiln
is there something happening with PDMin? or something has changed? basic example is not working with latest but is working with older release.
```py
import time
import array
import math
import board
import audiobusio
Remove DC bias before computing RMS.
def mean(values):
return sum(values) / len(values)
def normalized_rms(values):
minbuf = int(mean(values))
samples_sum = sum(
float(sample - minbuf) * (sample - minbuf)
for sample in values
)
return math.sqrt(samples_sum / len(values))
Main program
PDMIn(clock, data)
PDM_CLK = board.SCK
PDM_DAT = board.MISO
mic = audiobusio.PDMIn(PDM_CLK, PDM_DAT, sample_rate=16000, bit_depth=16)
samples = array.array('H', [0] * 160)
while True:
samples_recorded = mic.record(samples, len(samples))
magnitude = normalized_rms(samples)
print(samples_recorded, (magnitude,))
time.sleep(0.1)
```py
Adafruit CircuitPython 10.0.3 on 2025-10-17; Adafruit QT Py RP2040 with rp2040
import pdm_test
160 (27868.4,)
160 (27800.1,)
160 (28557.6,)
160 (27592.9,)
160 (30034.0,)
160 (27794.3,)
160 (31051.4,)
160 (27718.9,)
160 (31760.0,)
160 (27820.6,)
160 (32098.1,)
160 (27696.6,)
160 (32036.7,)
160 (27840.9,)
160 (31387.6,)
```py
Adafruit CircuitPython 6.2.0 on 2021-04-05; Adafruit QTPy RP2040 with rp2040
import pdm_test
160 (68.3422,)
160 (85.6272,)
160 (56.5817,)
160 (70.3068,)
160 (817.971,)
160 (914.963,)
160 (2242.5,)
160 (1457.44,)
160 (2299.47,)
160 (1060.62,)
160 (401.675,)
160 (66.1168,)
160 (49.2124,)
160 (57.3926,)
"not working" = the values do not change when making noise vs. silence
tulip sleet
@tidal kiln I'm assuming you found a bug; I don't see an existing issue, so please open one. Maybe it's RP2xxx-specific? also report when it started failing -- that's very helpful of cours
tidal kiln
i searched issues too and did not find anything. wanted a second look before i opened something. thanks!
manic glacierBOT
tidal kiln
@danh ^^ (and anyone else)
tulip sleet
@danh ^^ (and anyone else)
I categorized the issue
slender iron
<@&356864093652516868> We'll have our weekly meeting in about an hour from now in this text channel and in the circuitpython voice channel. Please take the time to add your notes in advance to the document: https://docs.google.com/document/d/1wMpXc_QzzngY0Nk-AnDDyDzIqjy_pXXW76yKbKys838/edit?tab=t.0
slender iron
@lone axle could you record for me please? my obs isn't starting up
turbid radish
Send in your tips for newsletter to cpnews@adafruit.com
tidal kiln
👋
lone axle
Thanks for hosting Scott, have a great week everyone.
slender iron
@proven garnet I saw your note about using wokwi for memory testing and I think it'd be more worth while to try using the zephyr native_sim
lone axle
@slender iron Video is uploaded to yt now. https://youtu.be/3b7ZhhTR6NY
slender iron
Here is the notes document for next Monday’s CircuitPython Weekly Meeting. It is at the normal time of 11am Pacific / 2pm US Eastern here on Discord. Add your hug reports and status updates to the document before the meeting. If you are unable to attend but would still like to contribute, feel free to add your notes and we’ll read them off during the meeting. Hope to see you there! <@&356864093652516868> https://docs.google.com/document/d/1IsTbby5z8HJsNUdXBPoZI_zjVHO8dWR2t4TjUnRrmFc/edit?usp=sharing
short tendon
@tulip sleet I think I have every airlift board (and multiple wings). Are there any you would like me to specifically test, or just test in general?
tulip sleet
<@329766224093249548> I think I have every airlift board (and multiple wings). A...
Thanks! Just in general, maybe some more involved application than my test program. I didn’t test MQTT, for instance.
short tendon
Thanks! Just in general, maybe some more involved application than my test progr...
Will do. I have a few hours open tomorrow and will play with it
proven garnet
<@135415127879974912> I saw your note about using wokwi for memory testing and I...
Caught up on the meeting this evening (thank you @lone axle for uploading) - that sounds good and makes a lot of sense. Don't know much about it, but will look into it over the next week (Wokwi was cool but I had some problems trying to bringing it back up to speed maybe 6-12 months ago anyway, so nothing lost).
From my brief look into it, am I correct in understanding that the Zephyr port is an RTOS that basically runs the CircuitPython runtime as the application?
manic glacierBOT
slender iron
Caught up on the meeting this evening (thank you <@382939733107408897> for uploa...
yup! my native sim work is here: https://github.com/adafruit/circuitpython/pull/10795
manic glacierBOT
slender iron
yoto PR is ready for another look
and is basically a pre-req for the native sim one due to the codespell change
(which I should have done separately)
slender iron
@tulip sleet please look at the yoto PR before you are done for the day
tulip sleet
<@329766224093249548> please look at the yoto PR before you are done for the day
working on it -- was out for a walk
tulip sleet
<@329766224093249548> please look at the yoto PR before you are done for the day
all are working except monster m4sk is going into safe mode and is also getting a pin error
tulip sleet
i'm looking at the m4sk error in a little more detail
I get the RIGHT_TFT_SCK in use only sometimes. When I do it does NOT go into safe mode. When I go into safe mode there is no precursor error printed.
@slender iron offending line is
right_spi = busio.SPI(board.RIGHT_TFT_SCK, MOSI=board.RIGHT_TFT_MOSI)
previous interesting lines:
# Left screen spi bus
left_spi = busio.SPI(board.LEFT_TFT_SCK, MOSI=board.LEFT_TFT_MOSI)
left_tft_cs = board.LEFT_TFT_CS
left_tft_dc = board.LEFT_TFT_DC
left_display_bus = fourwire.FourWire(
left_spi,
command=left_tft_dc,
chip_select=left_tft_cs, # Reset on Seesaw
)
self.left_display = ST7789(left_display_bus, width=240, height=240, rowstart=80)
# right backlight on board
self.right_backlight = pwmio.PWMOut(board.RIGHT_TFT_LITE, frequency=5000, duty_cycle=0)
# full brightness
self.right_backlight.duty_cycle = 65535
# right display spi bus
right_spi = busio.SPI(board.RIGHT_TFT_SCK, MOSI=board.RIGHT_TFT_MOSI)
seems like maybe a static object is not being initialized and has some garbage somewhere
orchid basinBOT
slender iron
thanks @tulip sleet! I'll take a look now
(sorry I missed this earlier)
@tulip sleet what example are you testing with? you only shared the code.py and I'm missing a bmp running it
tulip sleet
<@329766224093249548> what example are you testing with? you only shared the cod...
\
I am not sure where this came from
some Guide example. It's not in the board guide that I saw
slender iron
I looked in the learn repo and didn't see it
tulip sleet
me neither
the eyes are from the Arduino code
it was just what was on the board from some previous forgotten debugging adventure
slender iron
it safe modes with main too
tulip sleet
if you press reset a few more times you may get the pin error. That's what happened with me
ok, i wrote Did work with 10.1.0-beta.1. I did not test it with main.
slender iron
I'll see if I can fix it anyway
suspect it is something with the dual screens
10.0.x works
tulip sleet
nvm 10.0.x crashes too if you ctrl-c, enter the repl and then go back to code.py
got it; I'll approve
slender iron
ty
slender iron
I'm thinking about how we could test multiple displays
manic glacierBOT
slender iron
@tulip sleet native_sim PR is ready and fixes the new m5stack board
tulip sleet
@slender iron not related except to zephyr, but might be of interest: https://github.com/micropython/micropython/pull/18678. I don't know if it applies to us.
slender iron
I don't think it does because we don't expose a transfer without a stop anymore
lone sandalBOT
steady mulch
Regarding NVM "Each assignment causes an erase and write cycle", does it have to be written in one go, or could you write to single bytes which currently hold the empty value? (Or perhaps that depends on the microcontroller?)
manic glacierBOT
short tendon
@tulip sleet I'm not sure where to post this (possible) bug. When trying to update py Airlift (via m4 or pyportal), on my Windows 11 machine, I can't use Adafruit WebSerial ESPTool, I get the error:
Invalid head of packet (0x4): Possible serial noise or corruption.
using the python ESPTool, I need to downgrade to4.7.0otherwise I get timeout errors
tulip sleet
<@329766224093249548> I'm not sure where to post this (possible) bug. When tryin...
make sure you are not connected in another window to the COM port, though I think you would get other errors
short tendon
I've made sure
tulip sleet
i also use esptool 5.x.x on Linux, but have not seen the timeout errors
use a slower speed like 115200 if you are not already doing so
short tendon
I'm doing that as well
I walked back the esptool every version until it worked (after jumping to a really old one to see if it would work)
tulip sleet
could you paste the command line you are using?
short tendon
esptool --port COM10 --before no_reset --baud 115200 write_flash 0 NINA_ADAFRUIT-esp32-3.3.0.bin
switching between 4.7.0 and 4.8.0 is works / doesn't work
tulip sleet
you could post this as a "Feedback? Corrections?" item on the AirLift update guide.
and you are using the passthrough UF2 from that guide, I assume
short tendon
I am, and can do
tulip sleet
this is also of interest to the wippersnapper people because they have an update guide
there is a circuitpython program that also does the update, but it requires storing the .bin in CIRCUITPY, which doens't always fit: https://learn.adafruit.com/adafruit-fruit-jam/upgrading-airlift-firmware
4.7.0 vs 4.8.0 is good -- we can see what changed
thanks for the detailed testing
manic glacierBOT
manic glacierBOT
wraith crow
@tulip sleet should I have seen a MAC address change when updating the airlift firmware?
tulip sleet
<@329766224093249548> should I have seen a MAC address change when updating the ...
No, I don't think so -- the MAC address is in hw. The sample program in the PR will print the version info.
wraith crow
After the firmware update your script says: MAC address: 94:e6:86:06:9a:88
but before (looking back at what I had posted on the PR) it said: MAC address: 88:9a:06:86:e6:94 Oh nevermind, the values are just reversed 😁
According to my router the new order is the correct one....
And the MAC address lookup website
manic glacierBOT
upper radish
ICYMI - Official support for web-assembly-based Python on Emscripten
https://docs.python.org/3/whatsnew/3.14.html#whatsnew314-capi-config
zinc walrus
Several months ago, when Win10 reached EOL, I switched to Arch Linux (CachyOS) after having prepared for a year by practicing my workflows in WSL. I'm now up and running again in KDE, with most of my needs met and configurations satisfactorily tweaked.
With the basics squared away, I'm beginning to rebuild my embedded development toolchain, but this time I want to avoid Microsoft as much as possible. I was very fond of VS Code and its ecosystem, but I just can't support that company any longer.
I've investigated VS Code's opensource builds (VSCodium and Code-OSS), and while these work well for some frameworks (eg. Zephyr), others are unsupported due to their dependency on MS plugins (eg. CircuitPython2 and PlatformIO).
One of my eventual goals is to play with the CircuitPython Zephyr port on the nRF5340 Audio DK, but that's still slightly beyond my reach. In the interim, I'd like to at least find a comfortable CircuitPython/Arduino coding environment with linting and completion.
KDE's Kate seems a promising candidate at the moment (with tio as serial monitor), but I'm soliciting advice and recommendations from those devs here with applicable experience. Suggestions and pointers to further resources appreciated.
serene token
Several months ago, when Win10 reached EOL, I switched to Arch Linux (CachyOS) a...
If you are going to stick with KDE, Kate is pretty awesome. I used KDE up until KDE 4, which was buggy and way too much like Windows Vista. Sadly I had to let Kate go, as it is dependent on way too much of KDE. I use Vim now, but if I was still using KDE, I'd probably still be using Kate.
As far as Vim goes, it depends on what you need. You can get plugins for a lot of stuff, but I mostly debug from the command line, I have multiple terminal windows open and use separate terminals for editing and compiling. If you like that kind of workflow, and are willing to put in the effort to learn Vim (it's not easy), it's well worth it. If you prefer GUI with integrated stuff though, you won't like Vim.
I also used Bluefish for a bit after leaving KDE. It's not bad, and maybe it's better now (that was over 10 years ago). The last time I used it, it was way too web centric. It had syntax highlighting and other features for a ton of languages, but the toolbar was full of stuff designed for writing HTML, JS, and CSS, even when working in C, C++, Python, etc... Again though, that was a decade ago. Maybe it's better for more than just front end web now? It might have more familiar IDE features than Kate (though I haven't used that in a decade either, so maybe they've added those).
(I generally disable most IDE features in any coding editor I use. I learned to code when features like autoindent and autocomplete didn't exist, and when my editor tries to do them for me, it just ends up mangling my typing. So there may be pros and cons to some of these that I'm missing, if you use these features.)
zinc walrus
If you are going to stick with KDE, Kate is pretty awesome. I used KDE up until...
Thanks very much for the advice. I used vi all through college (in the university computer lab, where I had to teach others how to even close it), so I'm perfectly comfortable in both Vim and the terminal.
But I don't miss the days of having to constantly sift through the docs for unfamiliar function references, especially now that IDEs can provide a fingertip onscreen reference, and prevent misspelled variables that have already been declared with helpful autocomplete.
I still have lots of options to explore, but I'll give Vim another look to see what it's capable of these days. Kate is based on the same text editing library as Notepad++, so it was instantly familiar to me from my decades spent on Windows.
upper radish
Several months ago, when Win10 reached EOL, I switched to Arch Linux (CachyOS) a...
Very cool! I've been working with WSL for a while now and would like to setup a Linux box at some point as well. Two of the options I'd consider are directly inspired by FoamyGuy and John Park. Tim (FoamyGuy) uses PyCharm - it features in his streams. There's also a setup overview video by Prof Gallaugher on YouTube I'll link. John Park uses Zed Editor, and you can see that during his CircuitPython Parsecs. Perhaps you might find one of those suitable?
serene token
Thanks very much for the advice. I used `vi` all through college (in the univers...
I use Notepad++ for somethings when I'm working in Windows. I didn't know it was based on the same library as Kate, but it makes sense!
I'm pretty sure there are autocomplete plugins for Vim, but how far you want to go with that is up to you. There are a lot of plugins for Vim!
Anyhow, good luck. My typical strategy when looking for a good application for something in Linux is generally just to look up options and then go through the list trying them until I find one that feels really comfortable. That said, somehow I ended up on Vim for my coding editor, so it's obvious that I don't always stick to that strategy! (Vim is certainly comfortable now, but when I first started using it? Not so much.)
wanton dirge
Hello friends! I'm back finishing up adding a couple more chips to the analog port; nearly ready to PR at this point. Before I do, I wanna check on an issue with USB connection. At the moment, when I connect via something like TeraTerm, it usually crashes on my first attempt to connect to the device. If I reset, then Teraterm is able to connect. I've seen similar issues with some other serial monitors like VS Code's serial monitor extension. My understanding is this could be due to a race between the serial monitor waiting for some specific packet from the target, and the target board already having delivered it by the time Teraterm tries to connect. Firstly, does that make sense? And secondly, is there a way I might work around that?
One of the quirks of our TinyUSB port is it doesn't support bidirectional endpoints (e.g. needs a unique ep for Tx/Rx), but I don't think that would cause what I'm seeing.
slender iron
@zinc walrus I use Zed or Sublime Text. I really like how fast Sublime Text opens. Honestly though, I'd recommend trying coding agents like Codex or Claude. I find myself often using them + Sublime Merge to see what they've changed. The editor matters much less to me now.
slender iron
Hello friends! I'm back finishing up adding a couple more chips to the analog po...
I'd push you to add the boards to the Zephyr port instead of the analog one
(I'd love to move the existing ones there too.)
tulip sleet
Hello friends! I'm back finishing up adding a couple more chips to the analog po...
I don't think Tera Term is looking for something in particular. Is it literally crashing? I have not seen that before.
The serial port might be going on and off and back on-line for some reason.
have you tried tio on Linux? That will reconnect automatically, and you will also see the disconnects.
might give more clues
wanton dirge
I'd push you to add the boards to the Zephyr port instead of the analog one
Last I tested (probably summer last year or so), it seemed like the Zephyr port still had some coupling to an existing port of CP. Is that not the case now? We have pretty good Zephyr support across the MAX32 parts, so if it uses the boards' .dts files + defconfigs & drivers directly, it should work fine.
slender iron
Last I tested (probably summer last year or so), it seemed like the Zephyr port ...
It's evolved a lot. I'm working on it primarily now. It uses the board dts to enable modules and setup board
I haven't tackled dynamic pin muxing yet but we'll get there. I switched to the Zephyr USB stack which makes it "just work" too
I'm working on BLE and networking now
wanton dirge
I liked how CP was doing pin aliases matching silkscreen...would I just add a .dtoverlay per board to give them aliases?
slender iron
right now it is based on the Zephyr gpio nexus
so if you have an "arduino pinout" it'll auto-add all of the names for that
that's how fixed busses like I2C work too
can I get dev boards for the new stuff?
(not sure what chip they are)
wanton dirge
It's the MAX32650 and MAX32666. In our HAL it's called the 32665, essentially the same underlying die
Boards are MAX32650FTHR / MAX32666FTHR
wanton dirge
I shall!
slender iron
maybe my zephyr_bleio branch too
I have a pretty big fix for the heap
I think you'll definitely want to switch because the analog port doesn't have ble
wanton dirge
Yeah, I don't think we have BLE support mainlined in Zephyr for the time being as well. Some issues with licensing restrictions there.
slender iron
ah
do you have a downstream with it?
I'm not against figuring out how to use vendor zephyrs
looks at nordic too
wanton dirge
Not publicly afaik
slender iron
👍 count us as a +1 then
zinc walrus
I appreciate everyone who provided Linux code editor suggestions, but watching @slender iron's recent deep dive yesterday was so demoralizing and discouraging that I'm reconsidering even bothering anymore.
I'm not at all interested in paying a computer to write my code, but that's clearly the only future left. Why even attempt to learn something this complicated? Expertise is obsolete; just outsource it to the machines.
manic glacierBOT
crimson ferry
Is this a sign of things to come... Copilot PR comments flooding this channel?
crimson ferry
@zinc walrus I wouldn't let it get you down. You can choose how you want to develop for your own projects. I enjoy the process of writing software from scratch or adapting from examples, and I'm not a professional trying to compete on productivity or get certain skills on my resume. I do not want to use an LLM. I don't even use editors with code completion.
lone axle
Is this a sign of things to come... Copilot PR comments flooding this channel?
We are testing it on PR reviews and all comments auto forward to here from the bot. Maybe we could adjust it to filter out copilot ones if there is a preference for that.
Whether it's a sign of things to come I think depends on how good it's reviews end up being. If they end up being useful i.e. good reviews with helpful suggestions then I imagine they will keep getting used.
and to @zinc walrus I agree with anecdata. Scott's or anyone elses choice of development environment doesn't really have to dictate how you work. I don't agree that is the only future left. There is room for all kinds of different people to develop CircuitPython itself and projects built with it in any number of different ways.
I think the answer for "Why even attempt to learn something this complicated?" will differ from person to person. For me the answer is because I think its fun, I want to help make it easier for other people to play with, and there is a great community of folks involved. Personally, the existence or use of these tools by me or anyone else doesn't really change those motivations.
serene token
<@834445433060589579> I wouldn't let it get you down. You can choose how you wan...
I do code professionally, and I don't like code completion or autotab. I also don't use LLMs (much). A large part of that is that my current work is research, and that means that LLMs have not been exposed to anything like what I'm writing, so they really can't do a very good job. Even for personal stuff, I'm mostly working on microcontrollers at the lowest level I can get away with, and LLMs aren't exactly great for that either. I've used basic search AI to generate example code, but even that I generally have to adapt to my specific use case.
A big part of it though, is that I want to know exactly what my code is doing (this saves an enormous amount of debugging time and makes maintenance much easier), and the time it would take me to go through code generated by an LLM to understand what it is doing is generally more than it would take me to write it myself.
That said, I have nothing against those who use LLMs. They can be really good for certain types of work. They just aren't as good for highly novel stuff and super low level stuff, which is both the space I work in and the space I prefer to develop in.
tiny peak
We are testing it on PR reviews and all comments auto forward to here from the b...
FWIW, if it becomes common to have storms of copilot review comments and human responses, it's not gonna be practical for me to monitor this channel any more. I enjoy being able to follow the pulse of what's going on with CircuitPython PRs and issues. But, if today's volume of notifications becomes the norm, I don't think I'd be able to continue.
Specifically, I mean that posts here by the GitHub app previously had a very good signal to noise ratio and were one of the best available ways to track what's going on with CircuitPython. I would be sad to see the average noise floor on that go way up. I don't want to use Discord moderation tools to ignore the GitHub app because then I'd lose access to its feed of human activity on GitHub.
tulip sleet
FWIW, if it becomes common to have storms of copilot review comments and human r...
We'll talk about how to manage this. Maybe we make a circuitpython github feed that's pruned in some way (no CoPilot, or only issues, and PR opens/closes, say), and leave it here, or put it in another channel. We could discuss this in the next In The Weeds in the weekly meeting.
tulip sleet
@stuck elbow note https://forums.adafruit.com/viewtopic.php?t=222491. I filed an issue in the stage library.
tiny peak
We'll talk about how to manage this.
zinc walrus
Manually studying the Zephyr port of CircuitPython at this point feels like learning to handcraft wagon wheels while watching Karl Benz drive by in his horseless carriage.
I've spent 3 solid years in the bottomless pit of complexity that is Zephyr, and I still only barely qualify as a beginner. West, cmake, kconfig, devicetree, threads, workqueues, semaphores, mutexes, etc. It's nearly intractable, and why torture myself further when I will never hope to compete with an LLM?
Like @crimson ferry, @serene token, @lone axle, and @tiny peak, I enjoy writing code. I like understanding complicated systems and coaxing them to my will, but I can plainly see where programming is headed. I loved the accessibility and immediacy of CircuitPython for rapid prototyping, even with its performance limitations, but I'm beginning to doubt that it's where I should focus my efforts.
I had at one point hoped to participate in porting CircuitPython to the nRF5340 Audio DK, which is an amazing dev kit just begging for a simpler language than NCS (my kingdom for synthio), but I no longer believe it's within my grasp. Maybe instead I should abandon my attempts to program it at all and simply spec an agent to do it for me. Or maybe I should go do something else more satisfying.
tiny peak
I can definitely relate to the feeling of Zephyr being beyond my capacity as a human mortal to comprehend. Like, I've been able understand enough bits and pieces to use it to do some interesting stuff, but it's really complicated. Some things are easy, but if you want to go in a direction that's a bit off the beaten path... oh man... it gets intense fast.
zinc walrus
I can definitely relate to the feeling of Zephyr being beyond my capacity as a h...
Excellent individual but competing subsystems with absolutely no guidance on how to cohesively integrate them. And did Nordic release yet another backwards incompatible SDK version? Time to rewrite your app (again)!
stuck elbow
<@351136009922019331> note <https://forums.adafruit.com/viewtopic.php?t=222491>....
Thanks for the heads up, I will look into it.
wanton dirge
Hey @slender iron , here's my Zephyr port update!
I was able to successfully enable the Zephyr port for all the boards I wanted, which is great. However, there seems to be a major issue causing USB enumeration for CDC / MSC to take ~2-3 minutes. At the moment this seems like a doozy to debug...but will continue looking into it. I might try and use SystemView to help me there. I also tried the zephyr_bleio branch from Scott's fork -- same behavior.
I tried to add a "digilent,pmod" connector for our SPI pmod on the APARD32690 board, but we have some small deviations in our devicetree (e.g. multiple CS pins available, plus an INT pin) which caused this to be challenging. I could refactor that node and the I2C one to be more compliant, but at the moment that's not done and I just deleted the node from our DT.
My board work is here:
https://github.com/Brandon-Hurst/circuitpython/tree/zephyr-add-max32
They depend on some driver fixes which I made a PR to mainline Zephyr for (marked as draft while I check on a couple small portability things I want to quickly sneak in):
https://github.com/zephyrproject-rtos/zephyr/pull/103720
I already am mostly finished with the baremetal port for MAX32666/MAX32650. I do want to keep working on the Zephyr port here as well, but in the meantime if I PR'd the work on our other port, would it be considered? We have some board development projects we're working on that could really use it 🙂
slender iron
Hey <@252717193496756235> , here's my Zephyr port update!
I was able to success...
Yup, happy to review it on the analog port. Thanks for trying the Zephyr one. It is still early.
We do have our own fork of Zephyr if you want to get it in CP on Zephyr quicker
slender iron
I appreciate everyone who provided Linux code editor suggestions, but watching <...
This makes me so sad. My intention of highlighting LLMs wasn't to deter people. I'm so excited to have a new powerful tool for making CircuitPython so much better. I'm excited to focus on what I care about: design, ease of use, and system architecture instead of typing and learning obscure code-isms (/me looks at Kconfig). It makes so much more possible sooner to me than before. For your example of the Audio DK. Not only do I see adding a board def happening but we could more easily get to BLE Audio in CP. To me, LLMs make that so much more possible.
I still feel like I need expertise even with LLMs. It's just at a higher systems and experience level instead of the weeds of C pointers and other language-isms. (Though there are lots of bugs I still have to help solve, like one with pulling in half float routines from libc.)
I get that the cost makes it hard to jump in. I'm lucky enough to be paid for my work which covers my LLM cost. My hope is that the open source models you can self run get good enough to use as a normal part of every future computer.
I also get that LLMs shift our future as software engineers. It is weird and uncomfortable to not know what software engineering will be tomorrow and in ten years. That weird, uncomfortable feeling is causing me to question the way I work and feel out a new normal. I hope that by showing my work I can help make what software engineering is becoming less unknown and more comfortable.
I really hope everyone sticks with Adafruit and CircuitPython during this shift. Please let me know how I can be better to not alienate folks.
lone axle
<@&356864093652516868> the weekly meeting will occur here on discord at the usual time 2pm Eastern / 11am pacific, about 1 hour 15 minutes from now. Add your hug reports and status updates to the notes doc if you'd like to participate: https://docs.google.com/document/d/1IsTbby5z8HJsNUdXBPoZI_zjVHO8dWR2t4TjUnRrmFc/edit?usp=sharing
tiny peak
@slender iron re your post above... One of the things that would be interesting to see Adafruit folks engage with is the impact that AI adoption is having on aspiring coders and junior devs. Various companies are firing huge numbers of people to free up funding for AI projects. It's very clear that companies who write software are trying to use increased AI adoption to avoid hiring. One of the things I read into Adafruit's recent LLM streaming stuff is that hiring of new CircuitPython devs is much less likely than increased LLM spend [edit: that might not be true, but to an outsider it looks very plausible in the current social context]. To somebody evaluating programming as a career path, it's easy to see very very strong signals that it's not looking like a great option. OTOH, AI can give superpowers to experienced devs with secure funding. It would be interesting to see some engagement with this dichotomy.
tulip sleet
AI spend
slender iron
Sure, if Zephyr is interesting to you. Testing the Zephyr port with any dev boards you have would be helpful and then see what isn't working yet. Coordinate with me if you work within Zephyr.
turbid radish
Hi all. I have been outputting VGA video from the Adafruit Feather RP2350. Using Claude Opus 4.5 in the second seat. After iteration, I have PIO generating the HSYNC and VSYNC and I can generate a PIO bitstream with the set operator. All efforts to output something like a file image require out and that seemed to not work in the same context.Has anyone run into similar issues or better at PIO?
slender iron
FYI I just made #coding-agents-and-llms for coding agent and LLM discussion.
This works too!
wanton dirge
On the topic of LLMs, this RFD from Oxide was a pretty pivotal piece of writing for me to read on the subject and my guiding light in terms of how to think about when to use/not use them. I have found them pretty valuable in many use-cases, but as with many software silver bullets they can become detrimental if not wielded carefully.
https://rfd.shared.oxide.computer/rfd/0576
For instance...I don't know how much command-line gdb I really want to be doing with my life. As it turns out Claude can be decent at that, and can talk to my debug probe. Kinda nice especially for open-source work.
slender iron
They have a good podcast too
wanton dirge
Very aware 😛 I've probably heard over a hundred episodes at this point haha. Call me biased. But I thought it was a good piece of writing
slender iron
thanks for hosting foamyguy! Talk with you all next week (Tuesday!)
turbid radish
Thanks everyone
tidal kiln
👋
lone axle
Thanks everyone, have a great week 👋
errant grail
Thanks!
lone axle
Here is the notes document for next ***Tuesday’s ***CircuitPython Weekly Meeting. It is on Tuesday next week due to the US holiday on Monday. It is at the normal time of 11am Pacific / 2pm US Eastern here on Discord. Add your hug reports and status updates to the document before the meeting. If you are unable to attend but would still like to contribute, feel free to add your notes and we’ll read them off during the meeting. Hope to see you there! <@&356864093652516868> https://docs.google.com/document/d/1VZJGhjSXqRALVQQ6UR053aAV73KtAvlTx3KgfF531bM/edit?usp=sharing
orchid basinBOT
manic glacierBOT
fiery ferry
Hello, everyone. 🙂 I was wondering if anyone could help me. A friend gifted me a Macropad RP2040 a while back, and I was wondering about programming in CircuitPython on it.
random junco
Hello, everyone. 🙂 I was wondering if anyone could help me. A friend gifted me ...
Hi, sorry to hear about the trouble. You're going to want to ask in #help-with-circuitpython - this channel is for development of CircuitPython itself. Cross-posting is usually frowned on, but in this case I'd suggest it. You'll also want to share the error you're getting.
fiery ferry
Okay, thank you. 😊
tidal kiln
is anything special needed for adapting older code for newer magtag HW? or will it "just work" as long as using 10.x+ ?
trying to figure out what is creating this behavior
tulip sleet
trying to figure out what is creating this behavior
it is supposed to "just work". I don't have a new one to test. This is https://learn.adafruit.com/magtag-weather/project-code-2, is that right?
tidal kiln
yep. and not my HW either. i also haven't gotten a new one yet.
tulip sleet
I looked over your code and I don't see anything tricky that might cause that
maybe appeal internally for someone with a newer display to test. Scott did https://github.com/adafruit/circuitpython/pull/10505
so he probably has one
tulip sleet
@slender iron if https://github.com/adafruit/circuitpython/pull/10813 is ok by you, I'll proceed with an rc.1 release (after rc.0 build failed). thanks
slender iron
<@252717193496756235> if <https://github.com/adafruit/circuitpython/pull/10813> ...
will look now
slender iron
yep. and not my HW either. i also haven't gotten a new one yet.
This can happen when the colstart or rowstart is incorrect
because it doesn't update the right region of ram
tulip sleet
This can happen when the colstart or rowstart is incorrect
mentioned by timeline in forum thread: https://forums.adafruit.com/viewtopic.php?p=1081144#p1081144
not sure if there was a further change to the displays after you added support
tidal kiln
^^ yep. that's the mystery at this point.
orchid basinBOT
tulip sleet
Highlights of this release:
- Fix long-standing Thonny disconnect issues.
- Add
hashlib.new("sha256"). - Add
bitmaptools.replace_color().
lone axle
How hard would it be to add something optional for devices with /saves/ that would save a history file of the last 5-10 REPL commands and then allow using up arrow to go through them again in a new REPL instance? Even if it can't handle more complex stuff like code blocks, this would be quite convenient for single line statements.
stuck elbow
I'm testing the current main branch on an esp32-s3 device with a built-in busdisplay, and it seems to be horribly slow, like half a second per line of code or so. Is that a known problem?
slender iron
How hard would it be to add something optional for devices with /saves/ that wou...
Not sure. It is handled by some "readline" code
tulip sleet
How hard would it be to add something optional for devices with /saves/ that wou...
it would have to write to flash a lot. Alternatively it could save the history in RAM and only lose it if the board was hard reset
a more sophisticated terminal program could provide this as well
lone axle
RAM would be good still I think, I'm generally not hard resetting. Hadn't considered flash wear.
fleet hollow
Thonny does this automatically with REPL
stuck elbow
RAM would be good still I think, I'm generally not hard resetting. Hadn't consid...
I thinnk you already have history in the repl?
lone axle
I'm not familiar with thonny, I'll have to try it out. In the REPL now accessed via tio there is history, but only until it resets with ctrl+d. I'm hoping to be able to go REPL -> commands -> ctrl+d & ctrl+c -> up arrow to get back to commands.
stuck elbow
I'm not familiar with thonny, I'll have to try it out. In the REPL now accessed ...
perhaps it would be possible to save the history to saves when you press ctrl+d? it wouldn't work when just switching the device off, though
tulip sleet
perhaps it would be possible to save the history to saves when you press ctrl+d?...
I think it wouldn't be too hard to save the strings saved by the history outside the VM, so they wouldn't get erased.
fleet hollow
The pico NES emulator recently added HDMI audio support on the Fruit Jam over HSTX. Wonder if that's something we could see in CircuitPython in the future... 🤔 https://github.com/fhoedemakers/pico-infonesPlus/releases/tag/v0.36
slender iron
Definitely interesting!
lone axle
When anyone has a moment, https://github.com/circuitpython/CircuitPython_Library_Screenshot_Maker/pull/34 ready for review. It's under circuitpython org so cannot request librarians group.
candid sun
i tried approving but i don't think i have the right access unfortunately
lone axle
Thank you! It does show as approved but uses different icon color in some places which is odd. I think thats fine though.
manic glacierBOT
thorny jay
Here is the notes document for next ***Tuesday’s ***CircuitPython Weekly Meeting...
So today is Washington's Birthday, but tomorrow is King Cake... I'm just hoping it's not yellowcake. ☢️🎂 Though I guess even that isn't enough to postpone your meeting!
Europe will celebrate Carnival, like you do in New Orleans or St. Louis.
Nothing to report yet except a group hug.
orchid basinBOT
iron finch
Friends, is it only me or has anyone else seen a performance reduction in rc.1 vs. the beta version on ESP32-S3?
tulip sleet
Friends, is it only me or has anyone else seen a performance reduction in rc.1 v...
there was a fix from beta to final that would improve display performance when doing explicit refresh() calls. An issue on performance reduction with a reproducible example would be welcome.
tulip sleet
are you using a display? which board is it?
iron finch
No display. ESP32-S3 - N4R2
It's strange as it seems to be across the board, not a single library operation.
tulip sleet
could you try some simple timing, like just a loop that adds to a count or similar?
iron finch
For sure.
tulip sleet
is this an Adafruit board?
10.1.0 rc.1 does include the BLE "no flash" PR you proposed and I modified
iron finch
I'm building it with those flags to speed up the startup process as I'm not using BLE workflow / serial. I had the fix in the beta version as well. It's probably something with the build, I will re-build it and also test with and without those flags set.
stuck elbow
I also noticed a big slowdown on esp32-s3 with a display
iron finch
iron finch
The size of the build is the same.. I thought I had set a JTAG/debug flag by mistake.
iron finch
Can it have anything to do with commit 44f3816191 ("Fixes for native sim") in PR #10795?
The fix added port_yield() calls to the shared background callback code (background_callback.c, lines 64 and 91), so it affects all ports including ESP32 — not only Zephyr.
On ESP32, port_yield() calls vTaskDelay(4) (port.c, line 487), which with CONFIG_FREERTOS_HZ=1000 could mean a minimum 4ms sleep?
Potential fuzzy math that could be wrong:
background_callback_run_all() is called via the MICROPY_VM_HOOK_LOOP macro, which fires on every branch/jump opcode through DISPATCH_WITH_PEND_EXC_CHECK() — not just once per loop iteration.
A typical counter loop with an if check could hit 2–3 branch opcodes per iteration, so the actual cost could be 2–3 × ~4ms ≈ 8–12ms per iteration → ~83–125 counts/sec. At 3 branches × ~4ms = ~12ms per iteration, that gives ~83 counts/sec — matching my RC.1 log.
tulip sleet
~~https://blog.adafruit.com/2026/02/16/circuitpython-10-1-0-released/~~
EDIT: Do not use 10.1.0 on Espressif or Pico W. It will cause terrible performance issues. 10.1.1 is on the way
tulip sleet
Can it have anything to do with commit 44f3816191 ("Fixes for native sim") in PR...
good catch! I"ll take a look
surreal perch
I'm not sure how to classify this. My original bug code now works without the error but executing the code in REPL via importing the code still exhibits. I don't know whether it is significant or not.
>>> import code_sck_bug_as_filed.py
Initing SPI
Done initing SPI
>>> import code_sck_bug_as_filed.py
Initing SPI
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "code_sck_bug_as_filed.py", line 13, in <module>
ValueError: SCK in use
I'm almost leaning toward calling it an edge case and closing the issue. But perhaps I should wait and double check against 10.1.1.
stuck elbow
I'm not sure how to classify this. My original bug code now works without the e...
that is strange, the second import should be a no-op
strange zinc
Hello,
Is there anything I need to consider before I start implementing what I proposed? Is my proposed solution considered acceptable?
manic glacierBOT
tulip sleet
I'm not sure how to classify this. My original bug code now works without the e...
It's significant. You can't successfully call deep sleep directly from the REPL, and apparently not from REPL-imported code either.
But in the case above this is going to be fake deep sleep, because you're USB-connected (is that right, or is it via web workflow?). So this sounds like a fake deep-sleep problem?
orchid basinBOT
tulip sleet
<@&356864093652516868> Weekly meeting is in 37 minutes at 2pm US ET / 11 am US PT. Add your notes here: https://docs.google.com/document/d/1VZJGhjSXqRALVQQ6UR053aAV73KtAvlTx3KgfF531bM/edit?usp=sharing