#circuitpython-dev
onyx hinge
oh no is there yet another uf2 performance bug on macs !? https://rasterweb.net/raster/2025/05/04/macos-finder-and-usb-flashing-format-uf2/
jaunty juniper
oh no is there yet another uf2 performance bug on macs !? https://rasterweb.net...
I see the Finder (or BBEdit) lock up on rare occasions, usually when something is going wrong with the board and I have to unplug it to recover. I believe I also have rare cases of renaming a file in the Finder leading to maybe some corruption (like the file disappears), leading me to erase_filesystem() some boards, as well as experimenting with remounting the drive rather than erase to solve it: https://gist.github.com/Neradoc/c1d0f74c0a3ee7d514d40d38686add7a
tiny peak
I see the Finder (or BBEdit) lock up on rare occasions, usually when something i...
If anybody's having trouble with this, a possible workaround might be doing something about like this from a Makefile:
xattr -cr ${BUILD_DIR}; \
rsync -rcvO ${BUILD_DIR} /Volumes/CIRCUITPY; \
sync
That's the pattern I typically use for programming on macOS Sequoia (edit local files in BBEdit then run make from Terminal) and it's given me zero problems.
However, I don't tend to leave boards plugged in when I'm not actively working on code. So, if the issue is something that depends on the board staying plugged in for an extended time, maybe my rsync + sync method is irrelevant.
tiny peak
Does anybody know where things ended up with refresh rates for Fruit Jam's DVI out? I just tried 10.0.0-alpha.4, and it booted with 640x480 @72Hz. That refresh rate is too high for my capture card to sync to, and my TV initially didn't like it. After a reset, the TV seemed okay. Is there any plan to use 640x480@60Hz?
also, if I try messing with microcontroller.cpu.frequency = ..., it breaks the timing for USB host mode, so I can't do that as a workaround.
lone sandalBOT
manic glacierBOT
Currently the builtin Terminal does not show any visual representation for the tab character. If you run print("\thello") in the REPL it will show the visible tab in the serial console in tio, but visually on the display there is nothing shown where the tab would be.
tio:
Display:
This poses a challenge for th...
slender iron
Does anybody know where things ended up with refresh rates for Fruit Jam's DVI o...
I didn't end up changing it. You could try changing the cpu frequency before initing usb host
tiny peak
I didn't end up changing it. You could try changing the cpu frequency before ini...
okay, thanks... I'll see if that works
tiny peak
I didn't end up changing it. You could try changing the cpu frequency before ini...
I don't seem to be clever enough to find a way of setting microcontroller.cpu.frequency on fruit jam prior to usb host getting initialized. Seems like the port gets constructed in board_init() and any subsequent calls to usb_host.Port() just return that singleton? Am I missing something?
slender iron
I don't seem to be clever enough to find a way of setting microcontroller.cpu.fr...
I think you'd need to edit board_init().
manic glacierBOT
manic glacierBOT
manic glacierBOT
mortal kernel
@lone axle It's almost as if you read my mind 👍
mortal kernel
@lone axle If you like, I can review 10322.
lone axle
<@382939733107408897> If you like, I can review 10322.
Yeah that would be great, thank you. I think my change falls in line with Scott's proposal in the issue, and it'll be nice to get it in if we can so that PRs can stop failing. I can always make further tweaks later on if/when Scott or Dan have a chance to look and think there are better ways to handle.
lone axle
Now you're reading gamblor's mind! 😄
mortal kernel
I'm getting good at hitting that big green button 😀
blissful pollen
Haha my work laptop is updating so I had a few moments to read it. Figured this was important to get in
tiny peak
I think you'd need to edit `board_init()`.
would you be open to a PR that added a settings.toml setting to put a conditional around the usb_host init call in board_init()?
slender iron
no, I don't think many people would need it. Just build it yourself
tiny peak
eh, not worth it then. thanks. I'll just make do with how things are now.
mortal kernel
@slender iron I've gone through the 10.0.0 milestone issues and selected these to work on: 9171, 9491, 10153, 9381, 9992, 10197, and 10224. That should keep me busy for a few beats. If you have any others that you'd like me to work on, please let me know.
slender iron
your capture card is the only one I have heard of that doesn't work
slender iron
<@252717193496756235> I've gone through the 10.0.0 milestone issues and selected...
do you have something to measure sleep current with? don't do sleep related things until you do
nordic ppk2 is my goto
mortal kernel
nordic ppk2 is my goto
I have a PPK II on hand, so it looks like I made the right choice!
manic glacierBOT
Resolves: #10319
We could maybe use -> if we wanted to try to mimic the little arrows that some IDEs use to show tabs.
I have only done very basic testing with the REPL reproducer from the issue so far.
With this change:
without this change:
It still differs visually from my tio showing the s...
steady mulch
Hi, I've got a situation on Pico W where c5c7d11 causes PulseIn to miss input in elaborate circumstances that include "only happens when on wifi", and is only noticeably affecting one type of IR sensor and not a slightly different one. Any ideas where I'd go from here?
Also I wasn't paying attention and did make fetch-all-submodules which takes up several gigabytes, so is there an easy way to undo that or would I just need to delete the clone and make it again?
Also, where can I find the detailed breakdown of RAM and flash usage for the build that I vaguely remember seeing somewhere?
short marsh
My PR was merged, woohoo: https://github.com/adafruit/circuitpython/pull/10309
Cool experience contributing and having my updates usable for all!
In what stable release version will these changes be included in, and made available on the .UF2 download page (specifically for the PyBadge) (https://circuitpython.org/board/pybadge/)
lone axle
My PR was merged, woohoo: https://github.com/adafruit/circuitpython/pull/10309
...
Congrats! 🎉 Thanks for your contribution! It will be in an alpha release before it makes it into a stable one. The current alpha release is 10.0.0-alpha.4 so it will be in the next one 10.0.0-alpha.5The first stable release that it will be in should be 10.0.0 whenever that is released.
blissful pollen
My PR was merged, woohoo: https://github.com/adafruit/circuitpython/pull/10309
...
Congrats, I like the idea of what you added and wondering if we should add it to more audio parts. Though MixerVoice usually ends up being the "final" step in the chain so good place to have it.
slender iron
Also, where can I find the detailed breakdown of RAM and flash usage for the bui...
firmware.elf.map in the build directory will show all sections.
slender iron
Also I wasn't paying attention and did `make fetch-all-submodules` which takes u...
Easiest to start again
slender iron
Hi, I've got a situation on Pico W where `c5c7d11` causes PulseIn to miss input ...
How did you get down to that commit? It looks like it should just use DMA earlier. Not sure why it'd fail any more than not using dma.
steady mulch
How did you get down to that commit? It looks like it should just use DMA earlie...
I observed that it works reliably in 8.2.8 and 9.0.0-alpha.5, and doesn't work reliably in 8.2.9 and 9.0.0-alpha.6, and then I built a 9.2.7 with that line reverted to const size_t dma_min_size_threshold = 32 and the problem was resolved
We didn't notice the issue until recently because it hadn't been tested with the affected sensor on wifi, only with the unaffected sensor on wifi and the affected sensor not on wifi
We're talking about non-modulated IR with pulses a few microseconds long
lone axle
I wonder why socketpool listing here is so much more specific than the rest
slender iron
@mortal kernel I'm excited to get your stablity fixes out
mortal kernel
I liked the register re-work you did in #10325. Just saw your review request, I'll get right on it. Any testing you'd like done?
slender iron
I liked the register re-work you did in #10325. Just saw your review request, I'...
Yes please. I'm just made sure it compiled. Not exactly sure how to test besides a smoke test.
Since I'm not sure exactly when pointers are only in registers
mortal kernel
What I'll do for RISC-V is back out my fix and test. That should show that your change is working for RISC-V. For other targets, the best I can think of is a close reading of the code. So far in my reading I don't see anything that's unclear.
slender iron
I'm willing to test in the wild in an alpha too. That's why I marked it ready to review
mortal kernel
@slender iron I'm close to done for the week. If you'd like, I can take up the interpreter crash on Monday. It's failing without backing out my change to parse_compile_execute.
slender iron
It fails even with your change?
mortal kernel
Yes, which is unexpected.
slender iron
I'm done too. Next week is fine. No release with this code anyway
mortal kernel
Have a great weekend! We've got the annual beer festival at the fair grounds I live next to. This year's theme is "The Great Lebrewski". I've got my bathrobe ready.
slender iron
Have a great weekend! We've got the annual beer festival at the fair grounds I l...
You too! Have a great time.
manic glacierBOT
manic glacierBOT
cobalt grail
Not sure what I did wrong here?
https://github.com/adafruit/circuitpython/actions/runs/14946735028/job/41991019037
blissful pollen
Not sure what I did wrong here?
https://github.com/adafruit/circuitpython/action...
Could only glance, but did you run make translations and also commit the circuitpython.pot file?
cobalt grail
I did not, as I can't find any instructions on what to run?
blissful pollen
just make translations I think (or to that effect if you type make on its own it gives you the commands). Do that in the root directory.
cobalt grail
make: *** No rule to make target `translations'. Stop.```make translate
Sorry had the wrong command
After that locale/circuitpython.pot should be updated and is required.
I'm not sure if you have the local pre-commit scripts running, they can flag this as well (for the future)
cobalt grail
Lol. Found a bug in the Makefile as well, I don't have a "python" command on my system.
Ok, all checks have passed. Thanks!
mortal kernel
@slender iron I couldn't resist, so I spent a few minutes on the RISC-V crash with #10325. It was quick and easy to fix, so I submitted a PR. I also verified that with #10325, #10301 is no longer needed, so I reverted it in the same PR #10337.
manic glacierBOT
tiny peak
@lone axle In case it's relevant to what you're doing with Fruit Jam input devices, I've started working on a USB configuration and HID report descriptor parser as part of a gamepad tester project. The configuration descriptor parser works pretty well, but I'm just getting started on the HID report side. My goal is to be able to distinguish between different types of gamepads, including those that use generic HID with different report structures (e.g. old style DInput vs. new Switch compatible). Link: https://github.com/samblenny/fruit-jam-gamepad-tester
lone axle
<@382939733107408897> In case it's relevant to what you're doing with Fruit Jam ...
That looks quite nifty! thank you for sharing.
jaunty juniper
@mortal kernel heh I was running tests and adding to my post when you posted on the issue. Running alpha 4 on the Feather ESP32 S2, with this code:
import board
import time
import neopixel
import adafruit_dotstar
pixel = neopixel.NeoPixel(board.NEOPIXEL, 90)
pidots = adafruit_dotstar.DotStar(board.SCK, board.MISO, 90)
while True:
for color in [0x200020, 0x002020]:
pixel.fill(color)
pidots.fill(color)
time.sleep(0.5)
The board crashes within 15 seconds (never mind the ctrl-C issue I mention)
(wifi.radio.ap_info.rssi is -41)
tiny peak
That looks quite nifty! thank you for sharing.
I plan to do more on parsing the HID report descriptors, but not sure how deep into it I'll go. The descriptors supposedly describe the different possible HID reports and how the button bitfields, hat-switch controls (dpad), joystick analog values, etc. are laid out. From what I read, it sounds like the cheaper DInput protocol gamepads are really quirky with non-standard device-specific reports. Probably a lot easier to deal with Switch compatible controllers that are still HID but more standardized. (XInput is non-HID but it's also pretty easy to deal with)
mortal kernel
@jaunty juniper It looks like we're seeing different behavior. The current tip of main has trouble due to #10325. If that's what you're testing with, that may be why your testing is failing while mine isn't.
lone axle
I plan to do more on parsing the HID report descriptors, but not sure how deep i...
maybe this functionality can move into https://github.com/adafruit/Adafruit_CircuitPython_USB_Host_Descriptors or else in a community bundle library. It would make it easier for people swap in different controllers to use with the projects that get published.
jaunty juniper
I restested with alpha 4 right now, I don't have a Metro to see if it's different
mortal kernel
I restested with alpha 4 right now, I don't have a Metro to see if it's differen...
Interesting. I'm thinking there's a ROM difference between the Feather and the Metro. I'll re-open the issue and defer working on it until I've got a Feather on hand.
jaunty juniper
esptool could give the ROM version ?
mortal kernel
Good suggestion, let me see...
jaunty juniper
I can get that
Detecting chip type... ESP32-S2
Chip is ESP32-S2FNR2 (revision v0.0)
(I did chip_id, which also results in Warning: ESP32-S2 has no Chip ID. Reading MAC instead.)
this does say to use chip_id to get the chip revision
https://docs.espressif.com/projects/esp-idf/en/stable/esp32s2/api-reference/system/chip_revision.html#backward-compatibility-with-bootloaders-built-by-older-esp-idf-versions
mortal kernel
I'm having usb/serial problems. I'll need to reboot, back in a minute or two.
I'm back. esptool.py does not want to connect to the Metro. Maybe I need to somehow exit tinyuf2?
mortal kernel
Thanks, that worked. Seeing the same info you were:
Warning: ESP32-S2 has no Chip ID. Reading MAC instead.
MAC: 60:55:f9:da:bb:f4
what about the revision ?
mortal kernel
v0.0
jaunty juniper
well 🤷
mortal kernel
The chip's tin says it's ESP32-S2-WROVER, but no ECO number visible.
jaunty juniper
I wonder if the wifi network could affect it 🤷
note that beyond the test code, I originally encountered the issue because I have a QTPY piloting 2 LED strips while running a server with a web page to change the animations and colors, and opening Discotool Manager would be enough to make it freeze. Discotool Manager scans MDNS for CP boards and does 3 or 4 requests to get the board info (to which I added some delays to try to mitigate the issue). So I would put the board in safe mode so I could edit the code without moving the board.
I now switched to a QTPY S3, since the issue is harder to reproduce there.
jaunty juniper
I also tried erasing the feather and putting the alpha 4 bin directly with the same result
(situations where 2 people see different results have been related to the version/presence of the bootloader in the past even if it seemed to make no sense)
mortal kernel
I've tried tinyuf2 0.18.2, 0.21.0, and 0.30.0, all with the same results.
orchid basinBOT
manic glacierBOT
steady mulch
Hi, I've got a situation on Pico W where
lone axle
<@&356864093652516868> The weekly meeting is set to occur here on discord at the usual time 11am US Pacific / 2pm US Eastern here on discord. That is about 1 hour and 15 minutes from now. Feel free to fill in your status updates and hug reports to the document ahead of time if you like https://docs.google.com/document/d/1SeffJXPObzSCRRPUpaYE8T5QIz8m8bc2UfMYpH9Rop0/edit?tab=t.0 we look forward to hearing from everyone.
turbid radish
Thanks Tim!
mortal kernel
Thank you, Tim.
uncut nexus
Thanks, foamyguy, for all that talking 🙂
lone axle
Thanks everyone, have a great week!
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/19MYKsKO3_4BN_Xw3mpOf1Tfc3urLAIPSewX-TWIro2E/edit?usp=sharing
manic glacierBOT
slender iron
I think I'll do an alpha.5 after #10332 is in
slender iron
#10332 passed CI for whenever someone is ready to review
manic glacierBOT
mortal kernel
@danh When you have time, I'd like you to help me sanity check a couple things re: RP2 IRQs. No hurry.
tulip sleet
@danh When you have time, I'd like you to help me sanity check a couple things r...
hi - I missed this. go ahead (@danh is in plain text and didn't ping me)
mortal kernel
hi - I missed this. go ahead (@danh is in plain text and didn't ping me)
NP. I just finished eating a hummus, olive, tomato, marinated tofu sandwich on toasted sprouted wheat. Breakfast of Champions!
I've been working out how to implement #9992. First off, after reading through the SDK implementation of external interrupts, I think we want to use DMA_IRQ_0 on core 0 and DMA_IRQ_1 on core 1. The reason we need to keep these separated is that the SDK only supports a shared vector table, so the only way I can see to ensure routing these IRQs to the proper core is by splitting them up like this.
(This is essentially what we're doing now)
tulip sleet
looking at some stuff...
tulip sleet
I've been working out how to implement #9992. First off, after reading through t...
So my impression from the above is that when you set up the handler, it's for the core that the setup routine was running on, and gets routed properly. I don't think you need to keep them separate. The underlying handlers take care of this, it appears?
though I see this: https://forums.raspberrypi.com/viewtopic.php?t=336613#p2014763
- Currently by default (we will add a #define option to address this) the same exception vector table is shared between the two cores. This is not usually a problem, as most people don't want to enable the same IRQ on both cores....
- ... the problem with enabling the IRQ on both cores, is that both cores will keep taking the IRQ until one of them has serviced it....
- ... this is one of the reasons for having multiple DMA IRQs, such that you can divvy the DMAs IRQs up between core1) Currently by default (we will add a #define option to address this) the same exception vector table is shared between the two cores. This is not usually a problem, as most people don't want to enable the same IRQ on both cores....
- ... the problem with enabling the IRQ on both cores, is that both cores will keep taking the IRQ until one of them has serviced it....
- ... this is one of the reasons for having multiple DMA IRQs, such that you can divvy the DMAs IRQs up between cores.s.
mortal kernel
Your final reference agrees with my reading of the SDK code.
tulip sleet
so you don't have to use separate irq's, since you can find out the current core, and only handle the IRQ if it's for you.
whether it's better I'm not sure. If a core gets an irq and says, that's not for me, it doesn't take a lot of time.
are we currently doing DMA on more than one core?
mortal kernel
OK, but then would there be any utility to having the IRQ fire on both cores?
tulip sleet
OK, but then would there be any utility to having the IRQ fire on both cores?
no, but that's just how it works?
mortal kernel
Yes, RP2040 DVI, for example.
mortal kernel
Not really, since each core has its own NVIC, it can mask off IRQ sources its not interested in.
tulip sleet
well, I think my idea when writing the issue is that we could use a shared irq, and use the shared-irq API to take care of that. RIght now they are separate irq's because the code was written without considering that that API ws available. Basically I didn't want statically-allocated irq's, because we run out really fast
so maybe that's in line with your conclusion,
mortal kernel
Agree we need to share. Just saying that we share IRQ_0 on core 0 and IRQ_1 on core 1. And that gets us to the next interesting topic, sharing an IRQ.
tulip sleet
there is something in one of those links about "the SDK doesn't do x yet", but I can't find it now again (I was looking at dozens of forum threads)
tulip sleet
Agree we need to share. Just saying that we share IRQ_0 on core 0 and IRQ_1 on c...
this makes sense to me, then
but now the sharing issue...
mortal kernel
Right. What it doesn't do is support a vector table per core.
First off, I see something in the current implementation that looks like a problem.
tulip sleet
I guess it could demux that in software, but it doesn't do that yet? Is that what they were saying?
mortal kernel
Less than that, it's completely, totally not there. If the control to turn it on is set, it will throw a pre-processor error.
mortal kernel
Back to the current implementations. In isr_dma0, when a DMA_IRQ_0 comes in, it scans playing_audio, background_pio_read, and background_pio_write for DMA channels in use and calls everyone it finds indicating the DMA transfer is complete.
The called code doesn't check whether the DMA has actually finished. When concurrent DMA is in use, this could cause problems, like weird audio artifacts.
tulip sleet
yeah, so I thought using the shared_handler mechanism would help us without having to put that code explicitly in the fixed handler (and we don't want to use the fixed handler)
mortal kernel
Well, the problem there is that at the IRQ level, it's completely non-specific as to which of several sources might be causing the IRQ to get fired.
tulip sleet
there isn't a register with the DMA control block address or whatever?
eh, so what good is a shared handler? I thought it could demux it
shared DMA IRQ use has to be serialized then, can't be parallel?
mortal kernel
There is, but the NVIC knows nothing about it. When a shared IRQ comes in, all of the sharing clients get invoked in priority order. It's just a raw xxx(void) call, no context.
mortal kernel
So, what I'm proposing is a layered approach.
We register a shared handler in audio_dma and another in rp2pio. Each of those handlers looks at the interrupt status bits in the DMA block along with whether an object is attached to its table (playing_audio, for example) and then calls when it needs to signal completion of a DMA transfer.
tulip sleet
this makes sense, and I was just reading this post, which I think proposes exactly what you said https://forums.raspberrypi.com/viewtopic.php?p=2123216
Yes, that's the definition of shared handler - with shared handlers, all of the handlers will get called for each IRQ (from both sources), and the handlers have to check if it's "for them" and return quickly if not.
The overhead of doing this is small (a few instructions (the beginning of each handler should be to read a status register, check if the relevant bit is set, and if not return without doing anything), so for many purposes it's acceptable.
the whole thread is about exactly this
mortal kernel
Yes, that's the essence of what I'm proposing. Clearly I'm not the first getting here.
tulip sleet
great minds think alike
mortal kernel
I have been to this particular rodeo a few times.
tulip sleet
so yes, doing shared irq's forces such a mechanism. The current code was done at a time when there was so little being done with DMA IRQ's that only two seemed necessary. (Some lack of foresight, i guess).
The SDK API could help with this but doesn't, I guess.
I appreciate your careful thinking on this, and your experience
mortal kernel
Going back to what we're currently doing, sorting out which DMA channel(s) is causing the IRQ could clear up some intermittent weirdness.
tulip sleet
definitely, at the very least it would be clearer when one thing was starving out another. the whole "which is worse: flickering RGB matrix or glitchy audio" is kind of a use case thing, and I'm not sure we want to provide a choice on that right now (ideally the glitches are actually due to the kind of bug you suggested rather than just lack of memory bandwidth), but we don't know now, I think
mortal kernel
I'm going to go ahead. I appreciate your help and expertise vetting this. Thank you!
tulip sleet
not sure if I have expertise, but I always think "someone else has already had this problem, there must be a discussion" of it
orchid basinBOT
orchid basinBOT
ionic elk
is there a safe way to use load when debugging espressif with gdb? Seems to put my boards in a reset loop if I use it. I assume that has something to do with the bootloader.
tulip sleet
is there a safe way to use `load` when debugging espressif with gdb? Seems to pu...
are you loading it at the same offset as you would with esptool? And you're loading the equivalent of the .bin?
ionic elk
My esptool offset is 0x0 so I'm not sure what extra info I'd be including. It's the equivalent .elf file for the bin yes
Something might be borked with my whole setup though. GDB is giving me blatantly wrong information on variable values and it won't respect stepping over functions with n. My feeling is it probably originates from my pre-compiled microros library, which has been the source of a lot of frustration since it only seems to compile on linux and has limited information on how to compile with debug symbols.
mortal kernel
My esptool offset is 0x0 so I'm not sure what extra info I'd be including. It's ...
I've never had any luck with gdb load with Espressif parts. Breakpoints and stepping are also problematic. This doc has some useful tips: https://docs.espressif.com/projects/esp-idf/en/stable/esp32/api-guides/jtag-debugging/tips-and-quirks.html. When your shell is in the ESP-IDF virtual environment (source esp-idf/export.sh) the toolchain that matches the version of ESP-IDF you're using, including gdb, is in your path. I've found that's the best one to use.
ionic elk
I am using the esp-idf gdb, particularly since it's xtensa-specific
mortal kernel
I am using the esp-idf gdb, particularly since it's xtensa-specific
What is Espressif chip you're targeting?
ionic elk
I'm on the esp32s3. I could maybe just use advice on my overall strategic approach. I've been trying to pin down why rclc_support_init in my microros port is returning a failure, which has been challenging since microros does not support human-readable logging. I pre-compile the library on ubuntu 24.04 and have gotten it to include debug symbols, but since GDB is not providing accurate variable prints and isn't stepping as expected, it hasn't been helpful in identifying the cause of the failure.
ionic elk
Really what I need to do is be able to break on various lines within the execution of the rclc init process and bisect to see where the error return value originates. But that's hard to do when I can't get it to correctly print what the return value actually is.
It's possible this originates from how I'm building the esp microros component, which is all in CMAKE and it's not entirely clear to me how they're enabling the debug flags in the first place.
mortal kernel
The compiler flags you want are: -Og and -ggdb. The first is similar to -O2, but turns off some optimizations that can be confusing when debugging with gdb. The second turns on symbols, so it sounds like you're already doing that.
It sounds like the .elf file you're loading into gdb might not match the binary you're flashing into the 'S3.
I'm not familiar with microros, so I'm not going to be able to help much there.
Generally speaking, when all else fails I resort to an LED on a GPIO pin.
ionic elk
If the library is compiled with different debug flags than the final build, would that mess it up? The library has its own debug information
I'd typically go with a hardware logic analyzer kind of approach too but the precompiled nature of the esp-idf components is throwing me off.
I think I need to understand better how cmake deals with debug flags.
mortal kernel
I don't think mixing flags for the various components of your final link would break debugging. I've been doing that with CircuitPython builds with no issues. Are you loading the .elf from the final link into gdb?
ionic elk
yes, it's xtensa-esp32s2-elf-gdb build-espressif_esp32s3_devkitc_1_n8r2_ros/firmware.elf
I'm seeing some sources say that register storage of variables can throw off what GDB returns?
mortal kernel
Since this isn't CircuitPython development specific, we should move to another channel. #help-with-projects maybe?
ionic elk
It's for integration of an espressif component into circuitpython's build system, and Scott suggested I post here. That said, we can put it on pause until I research cmake's handling of debug flags more thoroughly. It's weird that I can't even grep for ggdb in the component, despite it building with the symbols I need.
That said, if it sounds like I'm doing anything terribly misguided and ought to take a different high level approach, I'd be open to feedback
I've been banging my head on this final init failure for over a month now
mortal kernel
It can be a bear when the debugging tools require debugging. Factor in a multi-layered build process, and it can be a head-banger deluxe.
manic glacierBOT
slender iron
@ionic elk my suggestion would be to get a setup where you can build microros in circuitpython directly. That way you can add ESP_LOG statements into its code. I exclusively debug ESPs with ESP_LOG
slender iron
@tulip sleet we probably want this USB update PR in alpha 5: https://github.com/adafruit/circuitpython/pull/10331
tulip sleet
<@329766224093249548> we probably want this USB update PR in alpha 5: <https://g...
👍 If you've tested it, I'll just approve.
slender iron
ok, draft alpha.5 notes are saved
ionic elk
<@323278594258370561> my suggestion would be to get a setup where you can build ...
I will probably have to modify the component build process to achieve this, since it fetches the ROS source as part of the build process rather than including it in the module itself. But it may be worth trying.
Especially since I'm hearing that GDB has some unreliability with espressif
maybe I should have started out with the stm32 microros port huh 😮💨
tulip sleet
@slender iron I was thinking about how to stage the TinyUF2 no-OTA update, and also how to write it up in the guides. it seems to me that we could update the TinyUF2 version presented for S2/S3 in the web installer, etc., because I don't see any S2 or S3 board that has dualbank turned on. So 9.2.7 would work with the new no-OTA partition scheme. Originally I was wondering the installer shouldchoose the version of TinyUF2 based on CircuitPython, but I think that's not necessary. This also simplifies the guide updates. I wonder if I am missing something.
slender iron
<@252717193496756235> I was thinking about how to stage the TinyUF2 no-OTA upda...
what boards do have dualbank?
tulip sleet
8MB boards, and boards with CIRCUITPY_LEGACY_4MB_FLASH_LAYOUT set.
slender iron
I will probably have to modify the component build process to achieve this, sinc...
It likely still caches the source and you can modify it after
tulip sleet
but none of the latter are S2, S3
I did test 9.2.7 is OK on various S3 boards, and that dualbank is off
Thach just submitted a new PR for S2 boards: https://github.com/adafruit/tinyuf2/pull/445
this is a follow-on to https://github.com/adafruit/tinyuf2/issues/432#issuecomment-2814042915
slender iron
it looks like CIRCUITPY_LEGACY_4MB_FLASH_LAYOUT only ever applied to non-UF2 builds
tulip sleet
yeah, I thought it was for S2 or S3, but apparently not. That's good. So it's a change for those plain ESP32 and C3 and C6 boards, but not terrible. I already made that change in my comprehensive draft PR. I have a Feather S3 only branch I could make into a PR too.
slender iron
https://github.com/adafruit/circuitpython/pull/9222 turned off dualbank in 9.1 it looks like
tulip sleet
@slender iron do you want to do the alpha release or do you want me to do it tomw or whenever? Which unmerged PR's do you want to be in it?
slender iron
<@252717193496756235> do you want to do the alpha release or do you want me to d...
I think it'd be good for it to have: 10331
you can do it. I'm off the clock
(trying the linker modification stuff via llm agent)
slender iron
should only need to add a note about the usb changes to the release notes
and your nordic fix
tulip sleet
I am also doing a PR for the Feather S3 4/2 board to use the new layout. WIll assign Limor to review it. We discussed in the meeting. That doesn't have to be in this release, but might make it easier to test.
atomic summit
Hey folks - on my test jigs, I have a script that allows me to specify a CP version, and have it wget all of that version of CP for each of my boards, backup the old firmware, and set the downloaded firmware as the current one for flashing and testing...
Been working great for years - but now I'm blocked out from grabbing them...
HTTP request sent, awaiting response... 403 Forbidden
2025-05-15 15:18:04 ERROR 403: Forbidden.
Any idea what has changed? I've also tried curl but I also get forbidden.
Fill command for each board is:
wget \
--user-agent="Mozilla/5.0 (X11; Linux x86_64)" \
--header="Referer: https://circuitpython.org/board/unexpectedmaker_tinys3/" \
https://downloads.circuitpython.org/bin/unexpectedmaker_tinys3/en_US/adafruit-circuitpython-unexpectedmaker_tinys3-en_US-9.2.7.uf2
--2025-05-15 15:18:04-- https://downloads.circuitpython.org/bin/unexpectedmaker_tinys3/en_US/adafruit-circuitpython-unexpectedmaker_tinys3-en_US-9.2.7.uf2
Resolving downloads.circuitpython.org (downloads.circuitpython.org)... 18.65.11.76, 18.65.11.190, 18.65.11.59, ...
Connecting to downloads.circuitpython.org (downloads.circuitpython.org)|18.65.11.76|:443... connected.
HTTP request sent, awaiting response... 403 Forbidden
2025-05-15 15:18:04 ERROR 403: Forbidden.
I run this script on up to 5 different jigs and it allows me to keep all CP versions current, but also switch to older ones easily for testing/comparisons.
manic glacierBOT
manic glacierBOT
tulip sleet
Hey folks - on my test jigs, I have a script that allows me to specify a CP vers...
I am also getting this problem with wget. It fetches OK in the browser. I saw a similar issue with circfirm yesterday. I'll report it.
atomic summit
I am also getting this problem with wget. It fetches OK in the browser. I saw a ...
Ahh, ok, I'm glad it's not just a decision to block wget/curl access, thanks for letting me know Dan 🙂
tulip sleet
I just reported it, thanks for this. I thought I had some local problem. It's probably a CloudFront misconfiguration.
tulip sleet
Ahh, ok, I'm glad it's not just a decision to block wget/curl access, thanks for...
I got it to work for now by altering the user agent, e.g.
wget --user-agent "Mozilla/5.0 (X11; Linux i686; rv:10.0) Gecko/20100101 Firefox/10.0" https://downloads.circuitpython.org/bin/unexpectedmaker_tinys3/en_US/adafruit-circuitpython-unexpectedmaker_tinys3-en_US-9.2.7.uf2
--2025-05-15 07:50:42-- https://downloads.circuitpython.org/bin/unexpectedmaker_tinys3/en_US/adafruit-circuitpython-unexpectedmaker_tinys3-en_US-9.2.7.uf2
just some random Firefox user agent I plucked off an explanatory webpage
jaunty juniper
I don't see mention of audiopwmio for ESP32 in the issues, has there been any discussion on that ?
tulip sleet
I don't see mention of `audiopwmio` for ESP32 in the issues, has there been any ...
It could be a long-term issue for tag:espressif. I think it is just an oversight that there isn't such an issue. But with audiobusio and and audioio (latter not great), maybe not such high priority. But it might not be that hard to implement.
If there is zephyr support for pwm audio, maybe we'd just wait for that.
tulip sleet
I don't see mention of `audiopwmio` for ESP32 in the issues, has there been any ...
I see the query in #help-with-circuitpython . Could use audiobusio with more pins
jaunty juniper
yeah the Qualia S3 doesn't have easy audio out
tulip sleet
Ahh, ok, I'm glad it's not just a decision to block wget/curl access, thanks for...
This user agent works:
Mozilla/5.0 (X11; Linux x86_64; rv:10.0)
This does not:
Mozilla/5.0 (X11; Linux x86_64)
cobalt grail
I noticed that my translations are added using my "@bergdahl" Github account, which is my business account. Not sure how this works, but I want it to be my "@cobalt grail" account instead as I am nearing retirement and will eventually close down the business.
How do I change this?
tulip sleet
I noticed that my translations are added using my "@bergdahl" Github account, wh...
I think this has to do with how you are logged into weblate.
manic glacierBOT
cobalt grail
I think this has to do with how you are logged into weblate.
Yes, that is the strange thing. I can't find anything in my profile that is Github related.
tulip sleet
Yes, that is the strange thing. I can't find anything in my profile that is Gith...
When you set up your weblate account, did you use the option to authenticate using GitHub?
also look here: https://hosted.weblate.org/accounts/profile/#account
jaunty juniper
maybe it's the email associated with the commits ? I just noticed weblate used my weblate email which is why it didn't credit me with my github account when I used it. I can change it to my githtub email
(you might need to add your other github email to your weblate account with the "Verify new email address")
tulip sleet
Ahh, ok, I'm glad it's not just a decision to block wget/curl access, thanks for...
This is now fixed, and you should be able to use your original user agents. We were interested in blocking bot fetches, but it was not fine-grained enough.
cobalt grail
When you set up your weblate account, did you use the option to authenticate usi...
I did not. But the email address I used is the one I use for the "bergdahl" github.
Will try to change the email adress.
lone sandalBOT
atomic summit
This is now fixed, and you should be able to use your original user agents. We w...
Thanks heaps Dan for following this up. Time to update all of my TestJigs to CP 9.2.7 !
steady mulch
Pico W, trying to change from building 9 to 8, changed the compiler to 10-2020-q4-major, did make fetch-port-submodules, rebuilt mpy-cross, did make clean, and I'm getting this error when trying to build. Have I missed something?
In file included from /home/xubuntu/circuitpython/ports/raspberrypi/sdk/tools/pioasm/pio_disassembler.cpp:10:
/home/xubuntu/circuitpython/ports/raspberrypi/sdk/tools/pioasm/pio_disassembler.h:16:25: error: ‘uint16_t’ was not declared in this scope
you may need to back up the version of pioasm, for some reason, not sure to what
what is your goal for doing an 8.x.x build, if I may ask?
steady mulch
Still trying to track down my problem here https://discord.com/channels/327254708534116352/1370201051336019968
tulip sleet
also do git status to see if any submodules are skewed
steady mulch
There are various untracked libs
tulip sleet
I have this alias for getting the submodules at the right commits:
alias gitsubupdate='git submodule sync --quiet --recursive && git submodule update --init --filter=blob:none'
Ah I thought the other command was supposed to do it
tulip sleet
and if you see that various submodules are "modified', you need to inside them and run that as well
no, it just fetches, it doesn't correct. You could do make remove-all-submodules at the top level and then do the fetch-port-submodules. Another thing to do is just to clone your fork again, with a different name, so you have a fresh start. Then you don't lose context on main or whateger.
tulip sleet
https://blog.adafruit.com/2025/05/15/circuitpython-10-0-0-alpha-5-released/
Highlights:
- Add stability fixes for Espressif port builds.
- Add fixes for direct connecting USB devices to PIO USB host.
- Improve accuracy of
time.sleep()and similar functions. - Add
MixerVoice.end(). - Change partition layout for Adafruit Feather ESP32-S3 4MB Flash 2MB PSRAM board, allowing BLE and other features to be enabled.
steady mulch
no, it just fetches, it doesn't correct. You could do `make remove-all-submodule...
Looks like the alias fetches everything? So need to do one of these instead if I only want the one port?
tulip sleet
it updates everything. You don't need it if you did make fetch-port-submodules on a fresh clone
steady mulch
Looked like it was fetching modules I didn't have yet but maybe I was mistaken
tulip sleet
not sure; I think it should not fetch modules that were not inited at all
steady mulch
remove-all-submodules and fetch-port-submodules hasn't helped
tulip sleet
you mean you still have the uint16_t error?
steady mulch
Yes
tulip sleet
is this on a fresh clone?
steady mulch
No, might have to do that then
tulip sleet
i am just trying it on a fresh clone myself
tulip sleet
No, might have to do that then
ok, this might be due to the wrong compiler being used. What does arm-none-eabi-gcc --version print?
steady mulch
arm-none-eabi-gcc (GNU Arm Embedded Toolchain 10-2020-q4-major) 10.2.1 20201103 (release)
tulip sleet
an easy fix seems to be to go into ports/raspberrypi/sdk and there do git checkout 1.5.1.
at that point I was able to use gcc 13.2 and the picow build built
no wrong, still some issue
orchid basinBOT
tulip sleet
`arm-none-eabi-gcc (GNU Arm Embedded Toolchain 10-2020-q4-major) 10.2.1 20201103...
I don't have gcc10 available right now, need to download it
steady mulch
I think it's the one the instructions told me to download for circuitpython 8
tulip sleet
I think it's the one the instructions told me to download for circuitpython 8
that is right. I think there is something odd going on with cmake, though. In the output I see -- The C compiler identification is GNU 13.3.0
so it's using a newer /usr/bin/gcc to build pioasm, and that's what's complaining (it's not something in the circuitpython tree). so you might need to roll back the system-installed gcc as well to get it to compile. This is going to be a rathole.
CircuitPython 8.2.7 was released Oct 2023, so it was being build on Ubuntu 22.04 more or less, using whatever the standard /usr/bin/gcc was at the time.
pico-sdk 1.5.1 helps a lot, but there are further issues about warnings, which I wasn't able to work around
could maybe disable the warnings-as-errors flag; I didn't try that
steady mulch
Hmm I've got an older VM I can try
mortal kernel
@danh This may or may not be a bug, what do you think? Unless deep sleep is entered from code.py or friends in run_code_py deep sleep never happens. For example, executing alarm_io1.py from the REPL by typing import alarm_io1 will not enter deep sleep.
alarm_io1.py:
import board
import alarm
import microcontroller
import digitalio
LED_PIN = microcontroller.pin.GPIO15
led = digitalio.DigitalInOut(LED_PIN)
led.direction = digitalio.Direction.OUTPUT
led.value = True
pin_alarm = alarm.pin.PinAlarm(pin=board.IO1, value=False, pull=False)
alarm.exit_and_deep_sleep_until_alarms(pin_alarm)
that sounds like a bug, but just to be sure, are you running disconnected from USB, since that will be fake deep sleep?
[hit tab after typing @danh otherwise it does not complete into a username]
orchid basinBOT
mortal kernel
that sounds like a bug, but just to be sure, are you running disconnected from U...
Not sure what you mean by "running disconnected from USB"? In this case, it's an ESP32-C6, so it doesn't have a USB MSC endpoint.
From my reading of the code, the reason that code run from REPL doesn't enter deep sleep is that both pyexec_friendly_repl and pyexec_raw_repl don't break out of their process loops on a PYEXEC_DEEP_SLEEP.
tulip sleet
Not sure what you mean by "running disconnected from USB"? In this case, it's an...
I remember that it didn't work from REPL, but it didn't seem important at the time, since at that time there were no non-USB boards, so it was going to be fake deep sleep anyway. If you can fix that's fine, but it's not high priority.
steady mulch
CircuitPython 8.2.7 was released Oct 2023, so it was being build on Ubuntu 22.04...
Thanks, confirmed it builds on ubuntu 22
tulip sleet
Thanks, confirmed it builds on ubuntu 22
ok, whew 🙂
mortal kernel
I remember that it didn't work from REPL, but it didn't seem important at the ti...
Yes, probably not important. I've been chasing #10224 and since I was testing with the import blah.py method from REPL it looked like the root cause for a bit. I've learned how deep sleep works in the process, so it was worth it.
tulip sleet
i don't remember why I did the exception thing to initiate deep sleep, but it was the easiest at the time
mortal kernel
It makes sense. Kind of like NLR. Trying to pass a result up from down deep can be, well, problematic.
steady mulch
ok, whew 🙂
But I thought I was building dd9ecc8 and os.uname().version says it's 9072fe6
tulip sleet
what does git log -1 say?
steady mulch
It does indeed say the latter
tulip sleet
did you checkout 8.2.7 or whatever you wanted to start with? are you going to do a bisect?
steady mulch
I did check it out but at some point it seems to have decided to checkout main again for some reason
tulip sleet
is it because you started working in a new clone?
steady mulch
It was after that. But I'm also pretty sure this board has v8 mpy on it and it's working
Confirmed it does
tulip sleet
it should build what is checked out -- there's nothing that would alter the current commit
make sure you do a clean build
steady mulch
git checkout dd9ecc81
cd ../..
make remove-all-submodules
cd ports/raspberrypi/
make fetch-port-submodules
cd ../..
make -C mpy-cross
cd ports/raspberrypi/
make clean BOARD=raspberry_pi_pico_w
make BOARD=raspberry_pi_pico_w
And after this, main is checked out, and I have a build that says it's main but appears to be a v8
tulip sleet
i don't understand what you mean by "And after this, main is checked out"
steady mulch
Git status now says it's on main, rather than detached head at dd9ecc81
tulip sleet
I don't see that problem locally
steady mulch
My VM just crashed, probably something wrong with it
tulip sleet
i just did the above except with an atmel-samd build (to avoid the build issues you saw), and didn't see any change in the checkout.
off to bed, good night!
steady mulch
Good night
tulip sleet
good night and good luck
steady mulch
I thought disk might be full again but it's not
steady mulch
Worked that time
stuck elbow
oh wow, someone translated the SAM speech synth to plain micropython! https://github.com/jacklinquan/micropython-samtts
ionic elk
@slender iron have you had bad experiences with gdb on espressif in the past that influenced you to do only ESP_LOGI debugging? GDB is continuing to be really weird and if this is just inherent to the platform I guess I'll pivot faster than I thought
ionic elk
Has anyone gotten GDB with JTAG to actually work on an esp32s3, for that matter?
Seems like I must be missing a crucial memory configuration or build flag step
mortal kernel
Seems like I must be missing a crucial memory configuration or build flag step
I have. For a narrow definition of working, anyway. Did you check out the Espressif tips link I posted recently?
slender iron
<@252717193496756235> have you had bad experiences with gdb on espressif in the ...
print logging has the benefit of not interfering with time sensitive things like USB and wifi
since it doesn't stop the world
mortal kernel
@tulip sleet Would you like me to have a look at #10355. I can get SWD debug up pretty quick.
tulip sleet
<@329766224093249548> Would you like me to have a look at #10355. I can get SWD ...
sure - I was going to do a bisect but not tonight
mortal kernel
I'll give it shot for an hour or so.
tulip sleet
it's an alpha, expectations are not high 🙂
mortal kernel
True that.😉
mortal kernel
@tulip sleet No progress on #10355. This was the first time I've tried SWD with debugprobe on the new Mac. A cornucopia of tool failures ensued!
tulip sleet
i have never gotten debugging to work on RP2040, but I haven't tried very hard
I am working with a user here: https://forums.adafruit.com/viewtopic.php?t=218315, which may be of interest. CPX board. time accuracy is better with 10.0.0-alpha.5, apparently, but there is a MemoryError: might be due to the new collect/no_collect logic. Would you expect an improvement of whole-second sleeps on atmel-samd after your PR?
mortal kernel
Which port is that?
tulip sleet
CPX is atmel-samd
mortal kernel
I am working with a user here: <https://forums.adafruit.com/viewtopic.php?t=2183...
Ah, CPX is Circuit Playground Express. Sleep is now accurate (or as close as it can be given we're not an RTOS) to subticks for all ports that go down to that level of precision.
+/- ~30us
tulip sleet
do you think that before your subticks fix whole-second sleeps were noticeably inaccurate? The CPX does not have a crystal. When used with a battery it is going to use its internal RC oscillator for timing. (If USB is connected it can do clock recovery from USB timing)
2 minutes in an hour that the user saw seems like a lot
i'm not sure they did the testing consistently with USB not connected; have to ask that
mortal kernel
It depends on the interrupt rate, but if it's high enough the cumulative error could have been pretty large.
tulip sleet
they are using countio to count button pushes, but I think that's the only interrupt. and the neopixel usage is low
countio is going to interrupt just on pin transitions
tulip sleet
i think so, but there was also something we did to avoid that, and I can't remember now...
mortal kernel
On the RP2040, I suspect it's the factored assembly code for register unloads.
That's the only change from 4->5 that's plausible.
I'll take a look to see if anything jumps out.
On the CPX timing, the issue is that it's improved but you're not sure why?
@tulip sleet On M0, the register unload area is too small by 8 bytes. I'm going to patch and test now.
tulip sleet
since CPX et al are also M0, same issue, maybe?
mortal kernel
since CPX et al are also M0, same issue, maybe?
Yes, there's also a second bug I'm hunting down now. Something to do with the __ARM_ARCH_ISA_THUMB symbol.
mortal kernel
since CPX et al are also M0, same issue, maybe?
The unload area size is definitely a bug. With that patched, I'm still falling into safe mode. I've disassembled the code, and it's building with the M0 version of the unload, and it's identical to code it replaces, so there's something else going on. It's not the __ARM_ARCH_ISA_THUMB, I've verified that is correct and the usage is correct. When I've tested, I've been careful to fully nuke the flash between tests. I'll be AWK for the next few hours for my walking club.
mortal kernel
@tulip sleet The Mac is causing difficulties that are messing with debug. I've switched back to Linux, and the register unload size patch does correct the safe mode issue. I'm out the door, I'll PR it when I get back.
mortal kernel
@tulip sleet I'm giving up on MacOS as development platform for now. There appears to be a new issue that silently prevents uf2 flashing from completing, resulting in an incomplete or corrupted image.
mortal kernel
<@849815081251766292> is that on 15.5? This sounds like a recurrence of an old p...
I'm sad to say it does sound like they've regressed.
tulip sleet
I'll test your PR on some boards after it complete and make an alpha.6
silently is worse; the old bug threw an error
mortal kernel
I've spent the last few weeks struggling with my new M4 Mini. It's just not up to being my main development ax. Unfortunately for me, I'm past the return window. It's back to Linux for me.
tulip sleet
but you can run Linux on it under parallels?
mortal kernel
but you can run Linux on it under parallels?
I can. I've hit some flakiness with USB passthru that affects debugging, though.
tulip sleet
looks like dual-boot is possible, e.g. https://umatechnology.org/linux-can-now-be-run-on-the-mac-mini-with-apple-silicon/
but that may not be the non-work workflow you want
mortal kernel
I had high hopes for a one system solution.
tulip sleet
our CI did not do a good job deciding which builds to do. I'll build some manually and test.
mortal kernel
I don't think there's any potential downside to the patch, but the upside is huge. There won't be any code size change. There's no chance of build breakage. After a couple of smoke tests, I'd say ship it!
tulip sleet
I don't think there's any potential downside to the patch, but the upside is hug...
for the forum post I mentioned above, that's still getting a MemoryError on the relatively small code.py the user posted. But I added a print statement one place, and it went away :/
we're going out for a bit; I'll look more later. If I add print("foo") before the #Create PIN line, it works. If I add it before vcm =0 instead, it does not ??
Nevertheless 10355 works fine on rp2040, so I think I will move ahead
mortal kernel
for the forum post I mentioned above, that's still getting a MemoryError on the ...
I think that deserves a closer look. Was that issue new with Alpha 5?
tulip sleet
thank you for the saturday fix!
I haven't tried an earlier alpha yet. It works ok with 9.2.7
ok, back in an hour or two
mortal kernel
My pleasure. See you Monday.
ionic elk
I have. For a narrow definition of working, anyway. Did you check out the Espres...
I've read both the tips and quirks page, and also your setup gist, if that's what you mean. Unfortunately, neither has helped much with the incorrect memory readings from GDB. I'm trying scott's suggestion of switching my whole build setup to Linux and trying some print debugging.
If this doesn't work, I may bail on espressif and see if I can implement STM32 or something to salvage the project
tulip sleet
I think that deserves a closer look. Was that issue new with Alpha 5?
MemoryError with that program on alpha.3 and later, not on alpha.2
orchid basinBOT
tulip sleet
https://blog.adafruit.com/2025/05/17/circuitpython-10-0-0-alpha-6-released/
Highlights of this release:
- Fix regression causing errors on ARM processors.
onyx hinge
I wonder why socketpool listing here is so much more specific than the rest
apologies for necro-replying but that's to show the build has IPv6 enabled. Otherwise, there's no AF_INET6 property in socketpool.
thorny jay
I am just booting for the first time a Metro RP2350 (10.0.0.-alpha.6) and while contemplating my DVI output, I notice that the blinking cursor is missing.
Maybe it has always been missing, from all the REPL I have seen on an LCD, but there on a big screen where I can compare the serial REPL and the DVI output... it feel like it is kind of missing.
This could become an issue for an editor, I guess I need to watch some streaming when FoamyGuy was working on the FruitJam, and see if there is one or not, or if this issue has already been discussed.
tiny peak
In case folks might be interested in more test code to trigger USB bugs on Fruit Jam, I just posted a guide for the gamepad tester thing I've been working on. It will pretty frequently trigger a glitch where unplugged devices only raise usb.core.USBTimeoutError (which is indistinguishable from normal timeouts) rather than a usb.core.USBError (useful for detecting unplugs, but doesn't always happen). https://adafruit-playground.com/u/SamBlenny/pages/fruit-jam-gamepad-tester
I haven't found a way to reliably reproduce the glitch, but it seems to be more likely with a Full-Speed device than with a Low-Speed device.
Feels like some kind of a race condition thing. Perhaps involving the interaction of display updates and USB polling.
It's also possible to induce more glitchy behavior by doing display updates that take longer. For example, re-drawing 5-lines of text in a bitmap_label every time an analog joystick axis changes its value is a pretty good way to bog things down and trigger strange behaviors. Drawing a one line hexdump in a bitmap_label is generally fine though.
tiny peak
oh, also, if I try to flash new code while a USB device is connected and actively being polled for input, that will pretty frequently lock up the board (host computer is M1 mini with Sequoia). As long as I unplug the USB device before flashing new code, it's usually fine. The behavior I've mentioned happens on 10.0.0-alpha.6.
lone axle
I am just booting for the first time a Metro RP2350 (10.0.0.-alpha.6) and while...
As far as I am aware there never has been a visible cursor in the REPL that can be seen on the microcontroller display. If you are connected to the REPL from the PC it can show a cursor on the PC side depending on what software you use for connecting. But on the microcontroller display there is no visible cursor, when you type the characters just start appearing.
For the editor app that I've been working on there is a visible editor that is currently handled "manually" in the editor app code. It keeps track of cursor location, and uses the tilepalettemapper module to inverse the colors of the location the cursor is at. It's solid inverted instead of blinking.
Inside of terminalio.Terminal in the core technically there is some cursor position variables and logic IIRC, but it's internal only and not currently "hooked up" to do anything. It is on my radar to eventually try to implement proper visible cursor handling inside of terminalio.Terminal, which if it were to exist could allow the REPL to show a visible cursor on the MCU display.
lone axle
apologies for necro-replying but that's to show the build has IPv6 enabled. Othe...
Ahhhh, thank you!
lone axle
oh, also, if I try to flash new code while a USB device is connected and activel...
I have been seeing this too, we should make an issue for it.
thorny jay
As far as I am aware there never has been a visible cursor in the REPL that can ...
It is just a "nice to have" feature.
Maybe if you can add that (optional) to terminalio.Terminal, then you could avoid specific code in the editor.
There might be a lot of good reason to not always enable the cursor... like e-Ink screen, or if one use terminalio to display stuff, but there is no user input expected.
It is just my first feelting when testing my "Franken Fruit Jam", there was something missing, and it took me a few seconds to figure out why that terminal was uncanny.
mortal kernel
@tulip sleet I did some noodling around with the OTA partition change on an Adafruit Metro ESP32-S2 device. Using tinyuf2 0.31: On a fully erased device, I can install tinyuf2 and then load my new CP firmware.uf2 using the newly installed tinyuf2, but it will not boot CP on a subsequent reset. I can flash the same firmware.uf2 with esptool.py and that boots successfully on reset.
tulip sleet
<@329766224093249548> I did some noodling around with the OTA partition change o...
i did not try esp32-s2 yet. esp32-s3 was owrking for me. There is a 0.32 release but I think the bootloader is the same
with esptool.py, you loaded the firmware.bin?
mortal kernel
Yes, I mistyped. .bin.
thorny jay
Dear @lone axle , would you consider renaming this guide https://learn.adafruit.com/usb-game-controller-with-snes-like-layout to something similar to what you did for Keyboard and Mouse, such as:
"Using a USB Game Controller with SNES-like Layout with USB Host"?
mortal kernel
<@&356864093652516868> We'll have our weekly meeting in about 50 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/19MYKsKO3_4BN_Xw3mpOf1Tfc3urLAIPSewX-TWIro2E/edit?usp=sharing -- Looking forward to everyone's updates!
lone axle
<@382939733107408897> will you be making a backup recording?
Yes, I am set up to take one.
tulip sleet
thanks!!
turbid radish
tulip sleet
btw, diagram below was generated by Claude Code, given the .csv files as input, and tweaked with several further requests (please center this, please make this taller so it's legible, etc.)
lone axle
Thanks for hosting Bob. Have a great week everyone!
tulip sleet
Thank you, Bob!
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/11ZlI_p6ie_kbPV95FjjcOoiUu-Hy6aI8pnxyoOxxGXk/edit?usp=sharing
manic glacierBOT
spice gate
I wrote a RISC-V emulator in pure Python (https://github.com/ccattuto/riscv-python) and ported CircuitPython to it (https://github.com/ccattuto/riscv-python/tree/main/advanced/circuitpython). It features an emulated UART backed by a pseudo-terminal on the host, an emulated block device backed by a filesystem image, interrupt-driven timer and tick support, CTRL+C support for tight Python loops, autoreload, and a little more.
manic glacierBOT
solar whale
@tulip sleet FYI -- just followed the guide to update the bootloader on an esp32s3 feather with 4Mbyte Flash 2Mbyte PSRAM and loaded 10.0.0 -alpha6 -- My espcamera (OV5640) test code runs well! I no longer have to do a "custom build"
I had one "glitch" updating the boot loader -- After erasing the flash, My Chromium Browser would not reconnect to the board -- I rebooted the computer (Raspberry Pi 5) and everything went well after that -- loading the combined.bin and then the .uf2.
solar whale
Also -- I went through the process on a second board and it all worked fine. No glitch.
tulip sleet
I had one "glitch" updating the boot loader -- After erasing the flash, My Chrom...
Thanks for your testing! Did you try simply unpluggingand replugging the board, or try a different port?
solar whale
Yes - I unplugged it several times -- finally resorted to reboot.
tulip sleet
did you also try shutting down and restarting Chromium?
sometimes I see the boards switch ports from ACM0 to ACM1, etc.
solar whale
yes -- that too -- appeared to be something odd with the port detection.
No problems at all on the second board
This wa using the WEB Flasher
tulip sleet
I can add a warning about that.
Right now the main procedure is to load the "Factory Reset + UF2" .bin. We plan to update those .bin's, we think.
So the somewhat tortured instructions now of "use these instructions", but use this other ".bin" will not be soe tortured
solar whale
It was clear to me what to do.
tulip sleet
good 🙂
solar whale
This is only implmented for the 4Mbyte/2MBPSRAM board now, correct? not for the reverse TFT.
tulip sleet
We wanted to do just one board, to vet the procedure, before doing a big bang. We really appreciate your testing.
There is this draft PR: https://github.com/adafruit/circuitpython/pull/10265. I have to fix the boards that didn't build.
solar whale
No problem -- just curious -- It's nice to have this one available -- without needing the custom build any more.
manic glacierBOT
manic glacierBOT
I found my Teensy 4.0 with a USB jack soldered to it and tested this out. Yep, works!
It works with a USB hub too. But if you unplug the hub and then go back to the device plugged in directly, it's not recognized. Is this a known bug? I've not played with USB host stuff recently in CirPy.
Anyway, here's what my Teensy 4.0 USB host wiring from six years ago looks like:
Also, since `usb.core.Devi...
jaunty juniper
@tulip sleet I wonder if we should have the i2ctarget fix in a 9.2.8 ?
tulip sleet
<@329766224093249548> I wonder if we should have the `i2ctarget` fix in a 9.2.8 ...
We don't have much else for 9.2.8, and the 10.0.0 alphas are pretty stable (maybe that's a matter of opinion 🙂 ), so I think main is fine.
manic glacierBOT
manic glacierBOT
manic glacierBOT
worn egret
are there any circuit python based firmware or kernels for a digital synthesyzer
?
how much processing power do i need for that ?
fiery pecan
are there any circuit python based firmware or kernels for a digital synthesyzer
Yes. Certain microcontrollers have built-in support for synthio after flashing CircuitPython. Read through this guide for more information: https://learn.adafruit.com/audio-synthesis-with-circuitpython-synthio
mortal kernel
RP2350 Deep Sleep
manic glacierBOT
digital shoreBOT
adafruit
manic glacierBOT
dusty axle
Hi there, I was hoping for some help with my Pyportal, I can't seem to get the sdcard to mount, it's stating that it can't set the block size to 512, (OSError: can't set 512 block size). I've tried both the adafruit_sdcard and sdcardio imports, using the latest circuitpython-9 release at the moment (tried the latest 10 alpha release with the same result previously)
wrong channel sorry
tulip sleet
Here is the notes document for next Tuesday's CircuitPython Weekly Meeting. It is delayed one day due to a US holiday. 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/11ZlI_p6ie_kbPV95FjjcOoiUu-Hy6aI8pnxyoOxxGXk/edit?usp=sharing
cerulean acorn
Hi! I am trying to change the hardcoded resolution on the Pi Zero 2W. I've followed https://learn.adafruit.com/building-circuitpython/linux and gotten to the Run Your Build! step, but I'm getting errors without having made any changes.
../../shared-module/lvfontio/OnDiskFont.c: In function 'get_char_id':
../../shared-module/lvfontio/OnDiskFont.c:327:21: error: a label can only be part of a statement and a declaration is not a statement
327 | uint16_t idx = codepoint - self->cmap_ranges[i].range_start;
| ^~~~~~~~
../../shared-module/lvfontio/OnDiskFont.c:328:21: error: expected expression before 'uint16_t'
328 | uint16_t glyph_id = self->cmap_ranges[i].glyph_offset + idx;
| ^~~~~~~~
../../shared-module/lvfontio/OnDiskFont.c:329:28: error: 'glyph_id' undeclared (first use in this function)
329 | return glyph_id;
| ^~~~~~~~
../../shared-module/lvfontio/OnDiskFont.c:329:28: note: each undeclared identifier is reported only once for each function it appears in
../../shared-module/lvfontio/OnDiskFont.c:327:30: error: unused variable 'idx' [-Werror=unused-variable]
327 | uint16_t idx = codepoint - self->cmap_ranges[i].range_start;
| ^~~
../../shared-module/lvfontio/OnDiskFont.c:327:30: error: this statement may fall through [-Werror=implicit-fallthrough=]
../../shared-module/lvfontio/OnDiskFont.c:331:17: note: here
331 | case 3: { // Direct mapping - need to look up in the table
| ^~~~
cc1: all warnings being treated as errors
-e See https://learn.adafruit.com/building-circuitpython; Adafruit Discord #circuitpython-dev
make: *** [../../py/mkrules.mk:93: build-raspberrypi_zero2w/shared-module/lvfontio/OnDiskFont.o] Error 1
I assume the main branch should build, so -- any idea what I'm missing?
stuck elbow
Hi! I am trying to change the hardcoded resolution on the Pi Zero 2W. I've fol...
can you show us the diff of the changes you made?
cerulean acorn
can you show us the diff of the changes you made?
I haven't made any changes
I just cloned it and tried to get through the instructions
stuck elbow
I just cloned it and tried to get through the instructions
are you using the correct version of gcc?
cerulean acorn
are you using the correct version of gcc?
I'm using 10.3-2021.07 because it's for Pi Zero 2W, which is a broadcom port
stuck elbow
I'm using `10.3-2021.07` because it's for Pi Zero 2W, which is a `broadcom` port
then I have no other ideas, I can't try it myself right now, sorry
cerulean acorn
no worries, thank you
jaunty juniper
yeah that sounds like a difference in version of gcc 🤔
cerulean acorn
Is there a different branch I should be using for the broadcom make?
I just want to change the resolution so it's not 640x480 😭
Can I just make a GitHub Action to do the build? I assume that's how the final versions get made?
jaunty juniper
it (the CI) seems to use 13.3.rel1 like other ports ?
arm-gnu-toolchain-13.3.rel1-x86_64-aarch64-none-elf.tar.xz
jaunty juniper
Can I just make a GitHub Action to do the build? I assume that's how the final ...
yeah it should do it if you upload a change to your fork
cerulean acorn
I'll try that first and then go back and try the more recent toolchain if that doesn't work
(If it's supposed to be the latest, might need to update the directions on https://learn.adafruit.com/building-circuitpython/linux)
jaunty juniper
hmmmm
oh I see the version and the link are wrong, but later in the text mentions "13.2.Rel1"
dunno if Dan is still around, I'll leave feedback on the guide
looks like the CI uses a different version for M and A, 13.2 versus 13.3
cerulean acorn
well, on the one hand, it looks like the CI completed. on the other hand, did it Upload to S3??
0s so hopefully not
it makes a disk image, not a uf2
jaunty juniper
no it doesn't upload
cerulean acorn
phew
cerulean acorn
ah okay
jaunty juniper
(which is main, you'd have to checkout for stables)
cerulean acorn
oh, neat
so main is alpha
so i should change to 13.3
thank you! i will try that
hm, that also makes a disk img
(.py) kevin@tinkerbell:~/circuitpython/ports/broadcom$ make BOARD=raspberrypi_zero2w
- Verbosity options: any combination of "steps commands rules", as `make V=...` or env var BUILD_VERBOSE
Memory region Used Size Region Size %age Used
READONLY: 976849 B 1536 KB 62.11%
RAM: 32104 B 1022 MB 0.00%
aarch64-none-elf-objcopy -O binary build-raspberrypi_zero2w/kernel8.elf build-raspberrypi_zero2w/kernel8.img
cp build-raspberrypi_zero2w/kernel8.img build-raspberrypi_zero2w/firmware.kernel8.img
0+0 records in
0+0 records out
0 bytes copied, 5.1987e-05 s, 0.0 kB/s
mkfs.fat 4.2 (2021-01-31)
adding: build-raspberrypi_zero2w/circuitpython-disk.img (deflated 95%)
that seems correct, you put that image on the SD card
there's no bootloader for pi zero
also don't forget that's not a solid port, it's pretty experimental
cerulean acorn
oh right
got stuck on following the directions, but we use the raspberry pi imager
cerulean acorn
Looks like it worked! Thank you!!
Weirdly, it's got a bit of a border that I didn't expect to be there, because it goes full screen normally and that's 1920x1080 i think
tulip sleet
@jaunty juniper @cerulean acorn Sorry, i misread "Pi Zero 2W" as "Pi Pico 2W", and was specifying the version based on that.
dusty marten
Hi Guys
I need help connecting wm8960 audio codec module to raspberry pi pico using circuit python.
How can I do that? I just need to check that pico is connected successfully to wm8960 codec and able to detect it
My main objective is to output audio and input microphone voice when connected to computer using that codec. I can later update the code to c++
Thanks in advance!!
stuck elbow
Hi Guys
I need help connecting wm8960 audio codec module to raspberry pi pico us...
try #help-with-circuitpython or #help-with-projects, this channel is about development of circuitpython itself
dusty marten
try <#537365702651150357> or <#330406777009209346>, this channel is about develo...
thanks
manic glacierBOT
Yeah, the minimal example that I ran on my badger doesn't have release_displays():
manic glacierBOT
lone axle
<@&356864093652516868> The weekly meeting is today on Tuesday this week due to the US holiday yesterday. We'll be meeting here on discord at the usual time of 11am Pacific / 2pm Eastern which is about 2 hours from now. You can add hug reports and status updates to the notes doc any time: https://docs.google.com/document/d/11ZlI_p6ie_kbPV95FjjcOoiUu-Hy6aI8pnxyoOxxGXk/edit?usp=sharing We look forward to hearing from everyone at the meeting.
thorny jay
I am present, but will stay without mic.
onyx hinge
hi folks! Just listening in today!
mortal kernel
Thanks, Tim! Have a great week, everyone!
onyx hinge
👋
lone axle
Thanks everyone! have a great week.
thorny jay
Thanks.
I have a question about "network stack"...
I am analysing the minimqtt.loop() for either non-blocking or play nicely with asyncio.
Basically, I read all the related Open and Closed issues for hint.
I try to see what are the "usecase" to test if ever I find a fix.
What are the various IP network stack in CircuitPython:
- ESP32SPI (aka AirLift) build on some board like PyPortal or as a module
- Espressif (ESP32, ESP32S2, ESP32S3, ...) use native network stack
- WizNet (Ethernet) use the specific library
- PicoW (onboard wifi chip on PicoW, does not support Pico2W)
- Anything else?
I have a doubt about PicoW...
Is that a specific stack?
Does this have a name?
Does it only work on/with the RP2040 or Pico"1"W? (not Pico2W)
Wait Anne made a nice Networking learning guide... maybe it has some answer?
mortal kernel
I have a question about "network stack"...
I am analysing the minimqtt.loop() f...
The IP stack for the Pico's is LWIP, same as Espressif. It is supported by both the Pico W and Pico 2W.
thorny jay
Updated IP stack list:
- ESP32SPI (aka AirLift) build on some board like PyPortal or as a module
- Espressif+LWIP (ESP32, ESP32S2, ESP32S3, ... and Pico W + Pico 2W) use native network stack
- WizNet (Ethernet) use the specific library
- Anything else?
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/1LkWa9fIlK7wzdpeyxYi2RbaUSO2Dx84AvcDQm40nA7s/edit?usp=sharing
mortal kernel
@tulip sleet I'm considering putting my attempts to use RP2350 power manager aside for now. While it's theoretically possible to achieve significant power savings with this facility, there are too many problems to solve (including possible SDK bugs and/or bad datasheet info) that I think I can better use my time thoroughly testing and tuning the SLEEP and DORMANT states.I have a specific issue with an interrupt window during transition to DORMANT state that I'd like to go over with you.
@tulip sleet I looked at #10380 and think it's a worthy fix. Do we have a policy for taking an upstream PR from Micropython outside of the periodic mass merge?
manic glacierBOT
empty salmon
@mortal kernel & @tulip sleet - THANKS !!
tulip sleet
<@329766224093249548> I looked at #10380 and think it's a worthy fix. Do we have...
If it's a really good/important fix, definitely.
tulip sleet
<@329766224093249548> I'm considering putting my attempts to use RP2350 power m...
I can talk now, no problem. Getting alarms to work at all is more important.
manic glacierBOT
digital shoreBOT
adafruit
orchid basinBOT
tulip sleet
https://blog.adafruit.com/2025/05/28/circuitpython-9-2-8-released/
Highlights of this release
- Fix deque bug.
- Fix I2S audio file read causing memory corruption.
- Support “Spectra6” six-color e-ink displays.
- Fix audiodelays.Delay when freq_shift=True.
- Board fixes.
ionic elk
I need to deinitialize specific objects within my circuitpython module in order - I can't run the cleanup function safely on "Node" objects until all "Publisher" objects have already been cleaned up. Is the best way to do this with a manual registry of objects and cleaning them up in order in my reset function?
fleet hollow
Is there any support for the continuous capture methods of imagecapture.ParallelImageCapture? I can't find it implemented anywhere. 🤷♂️ https://github.com/adafruit/circuitpython/blob/002881826a04669abaaa4f2d5220379f786cb159/shared-module/imagecapture/ParallelImageCapture.c#L10-L24
tulip sleet
I need to deinitialize specific objects within my circuitpython module in order ...
do Publisher objects point to Node objects? Could Publisher objects do the deinit of Nodes themselves?
can gc do the cleanup, or does it have to be manual? If you make gc by making __del__() point to deinit(), then gc would clean up everything, but not in any particular order. The deinit functions could be careful and not try to deinit something that is already deinited if deinit is not idempotent
ionic elk
There might be multiple publishers per node so they'd need to keep track of other publishers
it's a tree that has to be deconstructed from the outside in
I'd have the gc do it if it wasn't random in order. My impression is that if I only ran checks on whether something is "valid" to deinit, it'd leave me with leftover objects that I'd have to manually clean up anyway
tulip sleet
There might be multiple publishers per node so they'd need to keep track of othe...
if you don't deinit them what happens?
do they hold resources?
if you broke all links to them so they could be gc'd, then all the unconnected-from-the-outside nodes would have __del__() called on them eventually, once each. If they marked themselves as deinited (often by zeroing out a pointer, etc.), then you could use that as a "deinited?" check
this is how our current deinit stuff works
tulip sleet
I'd have the gc do it if it wasn't random in order. My impression is that if I o...
during gc, all the reachable objects are marked. if an object isn't marked, then it isn't reachable and its finaliser can be called
so a publisher would just take care of itself, and not try to deinit its nodes. the unmarked (unreachable) nodes would clean up themselves when their finaliser was called.
ionic elk
do they hold resources?
microros has its own allocation system and manages objects internally, plus it actually has to send a message to a co-operating agent on a linux machine to remove the nodes/publishers from the ros map
so this could be considered more like a hardware deinit than a software one, hence why I was thinking maybe a registry
Trying to understand the GC here, so a publisher object of type:
typedef struct {
mp_obj_base_t base;
rclcpy_node_obj_t * node;
rcl_publisher_t rcl_publisher;
} rclcpy_publisher_obj_t;
after circuitpython execution ends, would not be linked anywhere, and could be collected. And then once it's removed, the node would not be reached and could also be collected? So this might happen automatically already?
in the correct order, I mean
ie all I maybe have to do is add the correct _fini calls (microros destroy functions) to the deinits and maybe it will work out the gate?
tulip sleet
You'll need to use mp_obj_alloc_with_finaliser() in the make_new() and have something like this in the dict_table:
{ MP_ROM_QSTR(MP_QSTR_deinit), MP_ROM_PTR(&busio_i2c_deinit_obj) },
{ MP_ROM_QSTR(MP_QSTR___del__), MP_ROM_PTR(&busio_i2c_deinit_obj) },
When you call the _fini() routines, then NULL out at least one of those pointers so _deinited() can tell. Also you need to check for deinited everywhere. A good example would be, say, shared-bindings/busio/I2C.c (it sets a pin value to a invalid value, but same idea)
manic glacierBOT
tulip sleet
CircuitPython builders: Note https://github.com/adafruit/circuitpython/pull/10386, which has updated the gcc version to use to 14.2Rel1. You'll be notified if you try to build with an earlier version. Download that version from https://developer.arm.com/downloads/-/arm-gnu-toolchain-downloads.
ionic elk
What's the best practice for handling a failure that happens in a deinit function? Given that they can run during GC
Ignore? Have a preprocessor option for non-circuitpython debug message?
manic glacierBOT
tulip sleet
What's the best practice for handling a failure that happens in a `deinit` funct...
Yes, ignore, give up, or go into safe mode or something.
ionic elk
What's the recommended way of writing the RST in the shared-bindings layer? I don't see anything about it in the contribution docs. It's kind of fiddly to do by hand
manic glacierBOT
