sorry - ssh to rPI - and then serial output of DUT on rPI piped to my mac.
#circuitpython-dev
atomic summit
atomic summit
Device Under Test
My test jigs are rPIs and they flash and test each board and talk to the board and get data back from the bard during the test, and all of that is sent to the terminal that I am ssh-ing from. So I see the raw serial output
for example...
rPI: wifi connected to UM-C10 (192.168.4.59) in 4.917 secs
rPI: retrying 1
DUT: \x1b]0;\xf0\x9f\x90\x8dWi-Fi: off | code.py | 8.2.3\x1b\\\x1b]0;\xf0\x9f\x90\x8dWi-Fi: No IP | code.py | 8.2.3\x1b\\
rPI: Connecting to UM-C10 on channel 10```my wifi tests are designed to pass in under 4sec and if over, fail and retry
atomic summit
\xf0\x9f\x90\x8d is indeed snake when translated to UTF-8
Ok, so I now know it's a snake, thanks !
when I thought it was colour codes, I was going to see if I could also colour code my own output.
midnight ember
I'm unsure what the preprended \x1b]0; is though
could be color, possible
If it is a color will likely be a magenta
appears to be an escape character of some kind
would probably end up looking like \π\ probably used internally somehow
winter mortar
The \x1b (ESC) sequences are terminalio VT-100 speak
midnight ember
makes sense
jaunty juniper
I'm unsure what the preprended \x1b]0; is though
oh yeah that's the code for terminal title
I assume that whatever your code is that retrieves the data from serial an passes it down converts high ascii into escape characters
midnight ember
appears to be ansi
jaunty juniper
it's the code to put the status into the terminal title
midnight ember
makes sense
midnight ember
looks like Mu decided to just remove the whole sequence. at one time in 8.0 beta it was really really messing up almost every line. like every single repl output line had a snake it was a mess.
silent plaza
I don't know if this is the right place to ask this question but I'm wanting to create https://learn.adafruit.com/guggenhat-bluetooth-le-wearable-neopixel-marquee-scrolling-led-hat-display nut with circuit python using an nrf52840 or an esp32 s3 feather as the control board if not able to do with circuit python at the moment what board would be best to use with Arduino. Also if unable to be done in Circuit Python how best I ask for the features to be included in future versions of CP?
thorny jay
I would say that the right place is #help-with-circuitpython ... but I don't have the Discord Skills or Right to move it there.
That guide is from 2014 so CircuitPython did not exist at the time, but I don't see anything not possible in CP.
I found this guide that might help as it display text on a NeoPixel matrix: https://learn.adafruit.com/iot-led-sign
(but this is IOT so using the wifi of the ESP32S2, no Bluetooth)
The secret sauce might be this library to have a framebuffer made of pixel: https://github.com/adafruit/Adafruit_CircuitPython_Pixel_Framebuf
Then for BLE communication you could use the app "Bluefruit connect" https://learn.adafruit.com/bluefruit-le-connect
Now you just need to find some learn guide that use the text transmission of that application... and you need some other learn guide that scroll text on a framebuffer (you could try some MatrixPortal guide, and replace the matrix part with the Neopixel equivalent).
I don't say it is easy (but a lot of things are easy with CP), but my remixing two or three guide, you have piece of code for each part.
The difficulty nowdays is to find the right guide in the learn system, because maybe what you want to do already exist.
Don't forget the feedback you could put on the Guggenhat guide to ask if a CP would be possible, or share your code if you do it.
onyx hinge
@atomic summit The escape code "\x1b]0;" is "Change icon name and window title" and the title is terminated by "\x1b\". Sometimes just after a first connect, the sequence is incomplete or garbled; we're not sure why and would fix it if we could. And of course when whatever's receiving the sequence isn't a compatible terminal, it might appear garbled like that. And yes, b'\xf0\x9f\x90\x8d' is π in utf-8 encoding. [I see this was discussed to a natural conclusion hours ago but I wrote this so I'm sending it anyway]
brazen hatch
<@195765176035835904> The escape code "\x1b]0;" is "Change icon name and window ...
Terminals can be weird.
Ansi is very unreliable.
I think spamming it a lot more times around the code would be the only reliable way.
You have no idea how many bugs have I fought with to get terminal size detection going.
Though now I have my own .connected property for passive uart.
And now it's 100% reliable (with the power of spam) and even works over telnet.
Making another dotfiles pr in a bit, disabling dotfiles on boards that don't have native usb
buoyant lagoon
ive got some questions about the spi storage in circuitpython
i was running over some numbers in #1143997801274015774 and discovered a major bottleneck
when i was benchmarking erase/program times, i saw pico-sdk does 64kb erase blocks, so erase+program can run at about 80kb/sec
but circuitpython is using 4kb erase blocks, so it gets about 77kb/sec
however, if the blocks had been pre-erased, and your only having to program, it could get 625kb/sec!
i can also kinda see how you might bodge that ontop of fat
first, scan the entire FAT fs, find every used and unused block
second, pre-erase all unused blocks
third, keep a bitmap of pre-erased blocks, and store it somewhere, so you dont have to do 1/2 again
modify the fat block allocation algo, so it will prefer erased blocks
stuck elbow
what is your use case for needing fast writes?
buoyant lagoon
just better overall performance when copying things to flash
77 to 625 is a pretty big boost
stuck elbow
at a cost of slower startup
those things are usually tradeoffs, making them without a purpose in mind is not a good idea
buoyant lagoon
at a cost of slower startup
thats what the bitmap of erased blocks is for, so you dont have to scan on every startup
but a different filesystem, intended for flash, would also work
or a intermediate wear leveling layer
could also just be optional, so not ever board has it
slender iron
I think the ESP IDF has an intermediate wear leveling layer
I haven't looked because we don't really have a problem with wear
buoyant lagoon
i was thinking of using it more for pre-erase then proper wear leveling
but you kinda get both at the same time
if you just treat the entire flash as a ringbuffer, copy used blocks from tail->head, and erase everything in the middle, then your wear is more even, and things are pre-erased
slender iron
We're open to folks improving things. We don't have a shortage of ideas, only folks working on them.
buoyant lagoon
yeah
i also realized, i got some wires crossed
the original problem i was trying to solve, was programming a uf2 to an rp2040
so the circuitpython performance doesnt matter in that case
slender iron
yup, the uf2 code is in rom
slender iron
It can always be improved π Figuring out what to work on is the hard part
buoyant lagoon
yeah
the MBR code in the broadcom port is also rather fragile
if there is exactly 1 partition, it will create a second for the circuitpython drive
if there is exactly 2 partitions, it will expose the 2nd as MSD
anything else, and it breaks
so you cant multi-boot with pinn/noobs
you cant reserve the fake space on a 32gig card advertising itself as 256gig
buoyant lagoon
its a tricky thing to fix, because i can see it destroying data on an sd card when it goes wrong
so we would need to discuss options first
slender iron
issues are a better spot to discuss instead of discord since the history gets less and less accessible
buoyant lagoon
good point
slender iron
we have labels to filter by port too
buoyant lagoon
hmmm, bug report or feature request?
slender iron
I'd say enhancement since I didn't expect a CP SD card to be used for anything else
buoyant lagoon
we have labels to filter by port too
i think i lack access to add labels myself
slender iron
I'll add them. Thanks!
idle owl
@slender iron Are there going to be any changes to deal with the safe mode issues on the original Metro S3, or are we basically writing off those 40 boards? Mostly a curiosity. I don't need it for anything.
slender iron
With my PR it won't safe mode anymore. It just will say the pins are in use. I think Limor's plan is to refund folks if needed though.
idle owl
I knew her plan. I was curious about yours. Sounds good!
slender iron
manic glacierBOT
crimson ferry
Every once in a while, I have a device that consistently at least gets an exception, sometimes safemodes, from a NeoPixel exception. Espressif MCUs, but different board manufacturers.```
File "neopixel.py", line 180, in _transmit
RuntimeError: Input/output error
which port is that?
crimson ferry
espressif
buoyant lagoon
Every once in a while, I have a device that consistently at least gets an except...
from what ive seen, the neopixel is a write-only device
the driver has no way to know if the neopixel is working or not, and it should function the same if the neopixel is missing as when its present or malfunctioning
blissful pollen
the error is on neopixel_write which is in the core
crimson ferry
I used to get around it with a function that de-inited and re-inited the NeoPixel on exception, but it happens so frequently on those boards, it spends too much time doing that. Now I just move on to a different (identical) board and it's fine.
blissful pollen
But I'm not familiar with the RMT peripheral to really know more then that.
Just a quick look a couple previous setup calls don't check for error conditions.
crimson ferry
there must be something marginal in the hardware... somewhere ...for most identical boards to work and one not
stuck elbow
faulty microcontroller?
crimson ferry
marginal in some way maybe, i canβt think of another explanation for only one of otherwise identical units to have the issue
buoyant lagoon
@stuck elbow so if i wanted to add a new class, i would have to do it around here?
STATIC const mp_rom_map_elem_t videocore_module_globals_table[] = {
{ MP_ROM_QSTR(MP_QSTR___name__), MP_ROM_QSTR(MP_QSTR_videocore) },
{ MP_ROM_QSTR(MP_QSTR_Framebuffer), MP_ROM_PTR(&videocore_framebuffer_type) },
};
i'm thinking, there should be 3 objects, for the 3 displays, pre-created with the constructor blocked out
and then another object the user can create instances of...
how would that work exactly?
stuck elbow
<@351136009922019331> so if i wanted to add a new class, i would have to do it a...
I have no idea what that code is
buoyant lagoon
I have no idea what that code is
stuck elbow
I'm not familiar with the videocore module
buoyant lagoon
it looks to be a string=object mapping, for globals in this module, which can then be imported?
stuck elbow
yeah
buoyant lagoon
what i need more, is somebody that knows how C modules work in (micro|circuit)python
for example, what is MP_ROM_QSTR(MP_QSTR_videocore) doing?
why can i not find the definition of MP_QSTR_videocore elsewhere in the code?
build-raspberrypi_zero2w/genhdr/qstrdefs.generated.h:QDEF(MP_QSTR_videocore, 175, 9, "videocore")
ah, runtime generation
stuck elbow
it's a magical macro for interned strings
buoyant lagoon
git grep is always lying to me
stuck elbow
lie to it in return
buoyant lagoon
so, if i want to create my own class type, i would need to duplcate this line:
{ MP_ROM_QSTR(MP_QSTR_Framebuffer), MP_ROM_PTR(&videocore_framebuffer_type) },
and duplicate the struct its refering to in the other file....
root@d235fe3c2fdb:/circuitpython/ports/broadcom# grep test build-raspberrypi_zero/genhdr/qstrdefs.generated.h
QDEF(MP_QSTR_test, 51, 4, "test")
interesting, it seems to have just found that constant in my code
so i just act like it exists, and then it magically exists!
buoyant lagoon
@stuck elbow
.locals_dict = (mp_obj_dict_t *)&videocore_framebuffer_locals_dict,
what are these locals local to?
are they attributes of the resulting Framebuffer class?
>>> videocore.Framebuffer.
deinit height width
>>> videocore.Framebuffer.width
<property>
STATIC const mp_rom_map_elem_t videocore_framebuffer_locals_dict_table[] = {
{ MP_ROM_QSTR(MP_QSTR_deinit), MP_ROM_PTR(&videocore_framebuffer_deinit_obj) },
{ MP_ROM_QSTR(MP_QSTR_width), MP_ROM_PTR(&videocore_framebuffer_width_obj) },
{ MP_ROM_QSTR(MP_QSTR_height), MP_ROM_PTR(&videocore_framebuffer_height_obj) },
};
STATIC MP_DEFINE_CONST_DICT(videocore_framebuffer_locals_dict, videocore_framebuffer_locals_dict_table);
yeah, attributes of the class at least
libg.a(lib_a-sbrkr.o): in function `_sbrk_r':
hmmm, if i cant malloc() then how should i reserve some memory on the GC-able heap, for this class?
blissful pollen
hmmm, if i cant `malloc()` then how should i reserve some memory on the GC-able ...
gc_malloc I think. I just got in the door if you're still about I can try to find an example of it
buoyant lagoon
i am still about
blissful pollen
m_malloc will do it
buoyant lagoon
how will the object get freed when the python object around it is freed?
STATIC mp_obj_t hvs_sprite_make_new(const mp_obj_type_t *type, size_t n_args, size_t n_kw, const mp_obj_t *all_args) {
sprite_t *sprite = m_malloc(sizeof(sprite_t), true);
return MP_OBJ_FROM_PTR(sprite);
}
const mp_obj_type_t hvs_sprite_type = {
{ &mp_type_type },
.flags = MP_TYPE_FLAG_EXTENDED,
.name = MP_QSTR_Sprite,
.make_new = hvs_sprite_make_new,
};
the basic idea, is that the user code will spawn one instance of this, for each on-screen sprite
and if circuitpython reloads the code and GC's all objects, it should clean up all sprite objects
blissful pollen
Can't remember if you want the second m_malloc paramter true or not. If you looked it up and that's correct ignore me.
And it should be freed when the object gets freed. The GC checks for all pointers on the stack including C ones. Some objects have a deinit() python function that will explicity free the memory
buoyant lagoon
>>> import videocore
>>> videocore.Sprite
<class 'Sprite'>
>>> x = videocore.Sprite()
>>> x
and it seems to just hang now, when viewed from the CDC uart
ah, i did see a deinit on the framebuffer
blissful pollen
https://github.com/adafruit/circuitpython/blob/main/shared-module/displayio/Bitmap.c
Is a good examples that allocates a buffer and frees it in deinit
buoyant lagoon
i'll copy that if i need something more complex then free
i guess hw uart is next...
blissful pollen
Did you figure out the QSTR stuff you asked earlier?
blissful pollen
One thing that took me a while to figure out was that a QSTR is just an integer under the hood. Somewhere in the compile files you can see the mapping
buoyant lagoon
yep, found that file already
and ive delt with less magic-y interning before in the nix language
there is just a runtime function, to allocate/lookup a char* in the interned table
and then all maps use the int
buoyant lagoon
and it seems to just hang now, when viewed from the CDC uart
hangs the same way over hw uart
no backtrace or anything
buoyant lagoon
https://github.com/adafruit/circuitpython/blob/main/shared-module/displayio/Bitm...
next thing, also related to this link
i want to create a bitmap, and store it in an attribute on the sprite
basically:
import videocore
bitmap = something
s = videocore.Sprite()
s.backing_image = bitmap
and then be able to access the image from C
>>> b = displayio.Bitmap(16,16,255)
blissful pollen
https://github.com/adafruit/circuitpython/blob/main/shared-module/gifio/OnDiskGif.c
That one I know because the last major thing I contributed did just that
buoyant lagoon
heh
i'll need to find a uSD card then
You are in safe mode because:
CIRCUITPY drive could not be found or created.
guessed the right card on the first try!
import videocore
import gifio
sprite = videocore.Sprite()
gif = gifio.OnDiskGif("/discord-minecraft.gif")
sprite.image = gif
print("Hello World!")
dropped a random gif from my memes folder onto the drive, and it hung
@blissful pollen now what? its hanging on every boot π
blissful pollen
If you're stuck that way you can get into safe mode to edit code.py
buoyant lagoon
how do you get into safemode?
blissful pollen
if sprite.image is a bitmap too you need to assign it the bitmap from the gif. The docs in the shared-bindings folder should have a decent example
buoyant lagoon
typedef struct {
} sprite_t;
STATIC mp_obj_t hvs_sprite_make_new(const mp_obj_type_t *type, size_t n_args, size_t n_kw, const mp_obj_t *all_args) {
sprite_t *sprite = m_malloc(sizeof(sprite_t), false);
return MP_OBJ_FROM_PTR(sprite);
}
const mp_obj_type_t hvs_sprite_type = {
{ &mp_type_type },
.flags = MP_TYPE_FLAG_EXTENDED,
.name = MP_QSTR_Sprite,
.make_new = hvs_sprite_make_new,
};
sprite.image just doesnt exist
blissful pollen
how do you get into safemode?
Not sure off the top of my head, I think a couple resets the second after the neopixel lights up.
blissful pollen
Hmm yeah I'm not sure on the pi-zero
buoyant lagoon
my setup is also weird
the main circuitpython binary is loaded over the usb-device link
so i can change the C code freely, without having to swap uSD cards
even if its hanging like this
oh, i got a simple hack
just comment out the definition for videocore.Sprite
AttributeError: 'module' object has no attribute 'Sprite'
try and make a sprite now! π
import videocore
import gifio
sprite = videocore.Sprite()
gif = gifio.OnDiskGif("/discord-minecraft.gif")
# sprite.image = gif
print("Hello World!")
and that simple comment, solves the hang
so, i'm clearly missing something in my Sprite class
oh, i didnt populate the locals dict
bindings/videocore/Sprite.c:20:25: error: expected ')' before '(' token
20 | MP_PROPERTY_GETTER(test, (mp_obj_t)&get_test);
| ^~
@blissful pollen the macros make it hard to see what is wrong with this
blissful pollen
Do you have test and get_test defined properly?
//| width: int
//| """Width of the gif. (read only)"""
STATIC mp_obj_t gifio_ondiskgif_obj_get_width(mp_obj_t self_in) {
gifio_ondiskgif_t *self = MP_OBJ_TO_PTR(self_in);
check_for_deinit(self);
return MP_OBJ_NEW_SMALL_INT(common_hal_gifio_ondiskgif_get_width(self));
}
MP_DEFINE_CONST_FUN_OBJ_1(gifio_ondiskgif_get_width_obj, gifio_ondiskgif_obj_get_width);
MP_PROPERTY_GETTER(gifio_ondiskgif_width_obj,
(mp_obj_t)&gifio_ondiskgif_get_width_obj);```
As an exampleSTATIC mp_obj_t get_width_impl(mp_obj_t self_in) {
return MP_ROM_QSTR(MP_QSTR_foobar);
}
MP_DEFINE_CONST_FUN_OBJ_1(get_test, get_width_impl);
MP_PROPERTY_GETTER(test, (mp_obj_t)&get_test);
STATIC const mp_rom_map_elem_t hvs_sprite_locals_dict_table[] = {
{ MP_ROM_QSTR(MP_QSTR_test), MP_ROM_PTR(&test) },
};
as best as i can tell, its all correct
blissful pollen
Yeah I'm not noticing what's off
buoyant lagoon
bindings/videocore/Sprite.c:20:25: error: expected ')' before '(' token
20 | MP_PROPERTY_GETTER(test, (mp_obj_t)&get_test);
| ^~
| )
In file included from bindings/videocore/Sprite.c:3:
bindings/videocore/Sprite.c:23:44: error: 'test' undeclared here (not in a function)
23 | { MP_ROM_QSTR(MP_QSTR_test), MP_ROM_PTR(&test) },
| ^~~~
../../py/obj.h:312:24: note: in definition of macro 'MP_ROM_PTR'
312 | #define MP_ROM_PTR(p) (p)
| ^
i think the 2nd error, is because gcc ignored line 20, due to the 1st error
py/objproperty.h:#define MP_PROPERTY_GETTER(P, G) const mp_obj_property_getter_t P __attribute((section(".property_getter"))) = {.base.type = &mp_type_property, .proxy = {G}}
py/objproperty.h:#define MP_PROPERTY_GETTER(P, G) const mp_obj_property_t P = {.base.type = &mp_type_property, .proxy = {G, MP_ROM_NONE, MP_ROM_NONE}}
the 2 versions are a flash usage thing, so lets focus on the 2nd
const mp_obj_property_t test = {.base.type = &mp_type_property, .proxy = {(mp_obj_t)&get_test, MP_ROM_NONE, MP_ROM_NONE}}
that looks perfectly valid to me
blissful pollen
Yeah I still spot what's missing. Sometimes may have to do a clean compile too as some of the macros run before (random idea)
buoyant lagoon
wiped the build dir, no difference
@blissful pollen aha
#include "py/objproperty.h" was missing facepalm
blissful pollen
oops! It's late (at least here) so I can see it ha
buoyant lagoon
12:11 am here
still hangs though, so it wasnt the prop list
>>> videocore.Sprite.test
<property>
but the property is at least visible when i dont make a new instance
oh!
i'm calling m_malloc(0, false); ....
blissful pollen
Oh yeah
buoyant lagoon
nope
that wasnt it either
>>> s = videocore.Sprite()
sprite is at 0x3c5fa0
>>>
>>> s
a printf confirms it is allocating memory for the sprite when i construct it
how does circuitpython know how to print an unknown class?
blissful pollen
Not sure on printing that, never looked into it
buoyant lagoon
looks like you also support animated gifs, nice
is it changing the backing odg.bitmap?
blissful pollen
it keeps the same bitmap just re-writes it each frame
buoyant lagoon
yeah, so that will have tearing
the port i'm on has enough ram to just allocate a new bitmap for each frame, and can do vsync flips if you wait for vsync
but i can worry about that once i'm able to actually display anything
blissful pollen
I didnt notice any tearing on the small displays but there is barely enough RAM in a RP2040 for one frame let alone two
buoyant lagoon
but i'm on the rpi-zero, with at least 256mb of ram
so maybe make it an argument to the class
odg.next_frame(reallocate=True)
any kind of debug flag i could enable, to see why its just silently hanging?
buoyant lagoon
no difference
oh
i was wondering, how MP_OBJ_FROM_PTR(self->bitmap); turned into something that knows its of the bitmap class
there must be required fields!!
gifio_ondiskgif_t *self = m_new_obj(gifio_ondiskgif_t);
mp_obj_base_t base;
bingo
so circuitpython is casting my sprite to this, expecting to find the base
and its an undefined 32bit width....
π₯
self->base.type = &gifio_ondiskgif_type;
and that is critical
blissful pollen
oh yes it is
blissful pollen
π
buoyant lagoon
>>> s.test
'foobar'
and the getter works
now to change the code.py back...
AttributeError: can't set attribute 'image'
and then make another property
ok, so i have a mp_obj_t value, which came from the gif.bitmap
how do i assert that it is a bitmap....
displayio_bitmap_t *bitmap = m_new_obj(displayio_bitmap_t);
bitmap->base.type = &displayio_bitmap_type;```i could check for that exact type?
displayio_bitmap_t *bitmap = mp_arg_validate_type(args[ARG_bitmap].u_obj, &displayio_bitmap_type, MP_QSTR_bitmap);
this occurs in many places
TypeError: bitmap must be of type Bitmap, not OnDiskGif
it even gives a nice error, due to forgetting about .bitmap
>>> sprite.image
>>> sprite.image = gif.bitmap
>>> sprite.image
<Bitmap>
yep, progress!
>>> sprite.width
0
>>> sprite.width = 10
>>> sprite.width
10
#define simpleprop(name) \
static mp_obj_t c_getter_##name(mp_obj_t self_in) { sprite_t *self = MP_OBJ_TO_PTR(self_in); return MP_OBJ_NEW_SMALL_INT(self->name); } \
static mp_obj_t c_setter_##name(mp_obj_t self_in, mp_obj_t value) { sprite_t *self = MP_OBJ_TO_PTR(self_in); self->name = mp_obj_get_int(value); return mp_const_none; } \
MP_DEFINE_CONST_FUN_OBJ_1(fun_get_##name, c_getter_##name); \
MP_DEFINE_CONST_FUN_OBJ_2(fun_set_##name, c_setter_##name); \
MP_PROPERTY_GETSET(prop_##name, (mp_obj_t)&fun_get_##name, (mp_obj_t)&fun_set_##name)
#define prop_entry(name) { MP_ROM_QSTR(MP_QSTR_##name), MP_ROM_PTR(&prop_##name) }
simpleprop(width);
STATIC const mp_rom_map_elem_t hvs_sprite_locals_dict_table[] = {
{ MP_ROM_QSTR(MP_QSTR_test), MP_ROM_PTR(&test_prop) },
{ MP_ROM_QSTR(MP_QSTR_image), MP_ROM_PTR(&image_prop) },
prop_entry(width),
};
@blissful pollen behold!!, the power of macros!!
i'm surprised MP_QSTR_##name even worked, lol
>>> import videocore
>>> import gifio
>>> gif = gifio.OnDiskGif("/discord-minecraft.gif")
>>> sprite = videocore.Sprite()
sprite is at 0x3c61e0
>>> sprite.width
0
>>> sprite.height
0
>>> sprite.image = gif.bitmap
>>> sprite.width
313
>>> sprite.height
245
and now the w/h default to the input bitmap, when they started out at 0
blissful pollen
Nice, congrats!
buoyant lagoon
>>> sprite.maybe_regen()
regen time
>>> sprite.maybe_regen()
not dirty
>>> sprite.image = gif.bitmap
>>> sprite.maybe_regen()
regen time
>>> sprite.maybe_regen()
not dirty
and dirty tracking works
now, i need to actually generate the display list
https://github.com/librerpi/lk-overlay/blob/master/platform/bcm28xx/hvs/hvs.c#L142C24-L164
void hvs_regen_noscale_noviewport(hvs_layer *l) {
assert(l->dlist_length >= 7);
assert(l->premade_dlist);
uint32_t *d = l->premade_dlist;
// CTL0
d[0] = CONTROL_VALID
| CONTROL_PIXEL_ORDER(HVS_PIXEL_ORDER_ABGR)
| CONTROL_UNITY
| CONTROL_FORMAT(gfx_to_hvs_pixel_format(l->fb->format))
| CONTROL_WORDS(7);
// POS0
d[1] = POS0_X(l->x) | POS0_Y(l->y) | POS0_ALPHA(0xff);
// POS2, input size
d[2] = POS2_H(l->fb->height) | POS2_W(l->fb->width) | (l->alpha_mode << 30);
// POS3, context
d[3] = 0xDEADBEEF;
// PTR0
d[4] = (uint32_t)l->fb->ptr | 0xc0000000;
// context 0
d[5] = 0xDEADBEEF;
// pitch 0
d[6] = l->fb->stride * l->fb->pixelsize;
}
for basic sprites, it just needs 7 entries like this
and thats basically everything i need on the sprite side, for basic stuff
now i need the hvs side...
@blissful pollen now comes the fun part
i want to spawn 3 instances of the HvsChannel class, but not allow the user to create more
i think i can do the 2nd part, by just not having a .make_new
but then the 1st part, i need to create the 3 structs, and MP_ROM_PTR them??
wow, got that all right on the first try!
typedef struct {
mp_obj_base_t base;
uint32_t channel;
} hvs_channel_t;
extern hvs_channel_t hvs_channels[3];
hvs_channel_t hvs_channels[3] = {
[0] = {
.base.type = &hvs_channel_type,
.channel = 0,
},
[1] = {
.base.type = &hvs_channel_type,
.channel = 1,
},
[2] = {
.base.type = &hvs_channel_type,
.channel = 2,
},
};
const mp_obj_type_t hvs_channel_type = {
{ &mp_type_type },
.name = MP_QSTR_HvsChannel,
};
STATIC const mp_rom_map_elem_t videocore_module_globals_table[] = {
{ MP_ROM_QSTR(MP_QSTR___name__), MP_ROM_QSTR(MP_QSTR_videocore) },
{ MP_ROM_QSTR(MP_QSTR_Framebuffer), MP_ROM_PTR(&videocore_framebuffer_type) },
{ MP_ROM_QSTR(MP_QSTR_Sprite), MP_ROM_PTR(&hvs_sprite_type) },
{ MP_ROM_QSTR(MP_QSTR_HvsChannel0), MP_OBJ_FROM_PTR(&hvs_channels[0]) },
};
>>> videocore.HvsChannel1.width
1280
>>> videocore.HvsChannel1.height
1024
>>> videocore.HvsChannel1.enabled
0
mp_obj_new_bool was what i wanted
>>> import videocore
>>> videocore.HvsChannel0.enabled
False
>>> videocore.HvsChannel1.enabled
True
>>> videocore.HvsChannel2.enabled
False
>>> videocore.HvsChannel1.width
1280
>>> videocore.HvsChannel1.height
1024
@blissful pollen hows that look? π
now comes the next fun part...
i need to accept a list of Sprite objects, regen each, and then blast them into the display-list memory
buoyant lagoon
i wrote the rest of that code, ran it, and the screen just goes blank....
the stride, is half the width!!
w: 313, h: 245, stride: 157, bits per value: 16
aha
i forgot to call next_frame() the first time
what i should be seeing
what i'm actually seeing
i changed it from RGB332 to RGB555, and now i'm seeing 4 copies
the problem now, is that i need to understand what the data format is within displayio_bitmap_t
displayio_bitmap_write_pixel() should have answers
d[6] = s->bitmap->stride * 4; helps some
i can now just barely make out the pig, and the whole shape is there
and switching back to 555, i can now see just a single corrupted image, fully in the right shape!
let me try turning on animation, just for the hck of it
other then the colors, its looking amazing
the final code.py
oh, i might as well play with other sprite features, until i figure out color
now performance is horid π
ah, the printf's are to blame
i think its the framebuffer repl, thats always redrawing painfully slow
manic glacierBOT
buoyant lagoon
if anybody wants to test what ive done while i'm sleeping, its ^
wraith crow
On the road today/tomorrow otherwise I'd be on it! Can't wait to check it out when I get back π
buoyant lagoon
morning
with the newest code.py, its now able to cycle thru all of the gif frames, and it can animate the X/Y coords
blissful pollen
the problem now, is that i need to understand what the data format is within `di...
OnDiskGif sets the bitmap to RGB565_SWAPPED colorspace so two bytes a pixel I think (though looks like you figured that out)
buoyant lagoon
but now i need to understand how displayio bitmaps handle colors
ah, let me try 565
blissful pollen
The examples in the OnDiskGif docs give some hints how it works with displayio
buoyant lagoon
565 is still broken, but it looks better
the dark grey sides of the hole are showing up mostly green
but the brighter bottom sides are showing up dark purple
let me try the other color orders...
blissful pollen
There is 565 swapped too
buoyant lagoon
#define HVS_PIXEL_ORDER_RGBA 0
#define HVS_PIXEL_ORDER_BGRA 1
#define HVS_PIXEL_ORDER_ARGB 2
#define HVS_PIXEL_ORDER_ABGR 3
the hardware supports these 4 orders
ARGB and ABGR both look the same
but now i have an idea
from python, how would i generate a bitmap that is 32x32, and solid red?
or would it be faster to just whip up a test image in a gif, lol
actually....
if i just convert this to gif...
blissful pollen
Look up the displayio bitmap, it isn't too hard to generate a bitmap
I'm off to work for the day though so probably won't get a chance to check here, though this is more interesting
buoyant lagoon
ValueError: width must be <= 320
ah dang, your gif library has a max width
w: 32, h: 32, stride: 1, bits per value: 1
ah, but creating my own image, breaks things very much
it switched to a 1 bit per pixel format
because i gave Bitmap() diff args
looks like i will need more testing, for each bit-depth, and better docs on the internals
bitmap = displayio.Bitmap(32 * 3, 32 , (1<<16)-1)
bitmaptools.fill_region(bitmap, 0, 0, 32, 32, 31);
bitmaptools.fill_region(bitmap, 32, 0, 64, 32, (63 << 5));
bitmaptools.fill_region(bitmap, 64, 0, 32*3, 32, (31 << 11));
sprite.image = bitmap
HVS_PIXEL_ORDER_ABGR and the above, produces red/green/blue bars
HVS_PIXEL_ORDER_ARGB and the above, produces blue/green/red bars
i should just expose color order, as a param....
turbid radish
Some other test patterns
buoyant lagoon
after some more chat in #general-tech i got things working: https://github.com/adafruit/circuitpython/pull/8349
the root problem, was that gifio was byte-swapping all 16bit ints, to make SPI controllers happy
buoyant lagoon
manic glacierBOT
This requires a change in our fork of esp-idf:
My test platform is https://www.tindie.com/products/makerfabs/matouch-esp32-s3-parallel-tft-with-touch-7/ [800x480 model]
The display is unstable (horizontal tearing) at higher dot clocks, and on my test screen the low refresh rate of ~12Hz is a bit flickery. We think ...
mental nexus
Awesome, jepler! Itβs always a giant leap from black screen to something on the screen, as there are many different ways to make black screens and only limited number of ways to make pixels show up.
mortal kernel
Quick question about message translations. I'm putting new/modified strings into circuitypython.pot and assuming that there's some external process that will automagically get the translations into the various .po files. Is there something more I need to do?
onyx hinge
@mortal kernel pre-commit will tell you to run "make translate", which updates circuitpython.pot. We use weblate for users to contribute translations, and it also takes care of adding new messages to the individual language .po files, so there's nothing for you to do. https://hosted.weblate.org/projects/circuitpython/main/
mortal kernel
<@849815081251766292> pre-commit will tell you to run "make translate", which up...
Awesome. Thanks for saving me the work of updating the .pot!
onyx hinge
@slender iron design question for you. Some dot clock displays require configuration via a SPI bus; some real world designs place that SPI bus on an I2C I/O expander, because it's only used once at startup and the bus speed is unimportant (on the order of a few hundred bits are transacted once when setting up the display) ... and CP will want to do this at boot time outside of a VM for a board with an integrated display, or from circuitpython code if the display is NOT integrated or the user wants to re-initialize the display for any reason. SOOO... should bitbangio.SPI in the core be extended so that anything with a .value property can be used, in addition to a microcontroller.Pin? or would you prefer to see a different approach and if so what?
(this assumes the I/O expander would also have a core implementation I guess)
the espressif lcd devkit and limor's 2nd prototype dotclock tft board are both this way
buoyant lagoon
i had just thrown a random mp_raise_ValueError(translate("too many sprites, unable to pageflip reliably")); into some code i recently wrote
will that be handled automatically?
slender iron
<@252717193496756235> design question for you. Some dot clock displays require c...
Ya, that seems right. Very similar to the draft PR from pimoroni
finds it
onyx hinge
Ooh I'll look at that this afternoon!
slender iron
They need native io expander support there too
so I think it's worth doing
the CYW is kinda like that too
buoyant lagoon
<@252717193496756235> design question for you. Some dot clock displays require c...
the hyperpixel display also needs similar init, i forget which it is
originally, there was just a linux c binary, that bit-banged the whole thing
slender iron
@buoyant lagoon Ideally you wouldn't add a new sprite class. Instead change the TileGrid implementation to use them under the hood
buoyant lagoon
but more recently, linux was setup to drive it with an {i2c/spi}-gpio driver, so it showed up as a normal device and the driver bit-banged
buoyant lagoon
<@204112578425782272> Ideally you wouldn't add a new sprite class. Instead chang...
i'll have to look at what parameters the tilegrid stores, and how i can convert them over
slender iron
A tilegrid is a superset of a sprite
buoyant lagoon
does tilegrid support scaling?
does it support viewports?
slender iron
it only supports integer scaling now
buoyant lagoon
is it a int scale factor, or just seperate src and dst w/h?
slender iron
a tilegrid is a grid of indices into a spritesheet
a 1x1 tilegrid is a sprite
int scale factor
the advantage of making tilegrid use it is that existing code will just work then
slender iron
could you make one sprite per tile in the grid internally?
buoyant lagoon
what ive been calling viewports, is just fiddling with the starting addr, to render a sub-rect of an image
slender iron
right, so that's what the index of a tile does
buoyant lagoon
so you can take a 64x64 bitmap
treat it as a grid of 8x8 tiles, each 8x8 pixels
and then just render a single tile from that bitmap
in the little-kernel driver, its just viewport_x and viewport_y, an offset from the top-left of the original image
and then viewport_w and viewport_h, to set how big of a tile it crops out
buoyant lagoon
then it would accept a sprite sheet where the sprites vary in size
but the index of TileGrid is a far simpler api for users
slender iron
right now displayio assumes fixed tile sizes
buoyant lagoon
i can see how you might use TileGrid for fixed size sprites
and then maybe a new SpriteSheet for non-fixed sizes in the future?
slender iron
so you could add videocore.Sprite but also re-implement TileGrid with it
buoyant lagoon
i'll look into the docs, make some demos with TileGrid, and try adapting that...
slender iron
i can see how you might use TileGrid for fixed size sprites
and then maybe a new...
maybe! tough to know what direction to take displayio
foamyguy has done good work with it
buoyant lagoon
the other question, is about pixel formats, color order, and palettes
slender iron
the outer API is usually RGB888 and then converted internally
stuck elbow
maybe! tough to know what direction to take displayio
I would like to try and revisit the dma background writes
buoyant lagoon
from what i can see, displayio.Bitmap() acts like a palette with 1 to 65535 colors
slender iron
I would like to try and revisit the dma background writes
To overlap computing pixels with transmitting them?
buoyant lagoon
but a lot of things (gifio) treat it as rgb565
slender iron
ya, I think of Bitmap of "values" that get mapped to colors
buoyant lagoon
but there is no palette attached to the bitmap
so i have no way of just accepting a bitmap and displaying it
stuck elbow
To overlap computing pixels with transmitting them?
yes, basically add a paremeter that moves the code that waits for the transfer to end to the next call, and use double buffering
buoyant lagoon
sprite.image = gif.bitmap
sprite.color_order = 2
i need to add extra parameters like this
in this case, 2 refers to HVS_PIXEL_ORDER_XRGB
slender iron
but there is no palette attached to the bitmap
TileGrid is usually what combines a Bitmap with a "pixel_shader". A "pixel_shader" is a Palette or ColorConverter. The converter converts the source colors to the destination colorspace.
buoyant lagoon
ah
slender iron
Palette uses color converter code internally to map the RGB888 values to the destination color space
buoyant lagoon
but the rpi hw also supports palettes
1bit/2bit/4bit/8bit palettes are allowed
you give it an array of RGBA8888's, for whatever size you chose
slender iron
right, so by re-implementing tilegrid with hardware, you can have the hw do that conversion for you
I guess you'd reimplement pallete then too
buoyant lagoon
ah, nice
the other thing, is the REPL text console
i suspect its configured in the mmu as uncached ram, and just scrolling it is a major performance bottleneck
how best would i re-direct that to a bitmap/tilegrid, and then access it for inclusion in my sprite list?
slender iron
it is a tilegrid internally already
buoyant lagoon
so i just need to change it to use cached ram, and get a pointer to it
slender iron
right now the videocore implements a displayio framebuffer
so it uses all of the displayio code to render the tilegrids into the framebuffer
buoyant lagoon
the instant i run videocore.HvsChannel1.set_sprite_list(), that framebuffer vanishes, and never comes back
slender iron
https://github.com/adafruit/circuitpython/blob/main/supervisor/shared/display.c#L62 constructs that screen
buoyant lagoon
so i need to access its tilegrid object, and then i can have a videocore.HvsChannel1.show_text_console = True or similar
slender iron
it's all displayio still
buoyant lagoon
oh, and what about dynamically changing the size of that terminal?
lets say python code decides to put the terminal on the left half, and do gfx on the right half?
slender iron
the one weird trick the terminal uses is tilegrid's ability to change what index is starts the first row
it can do that already
buoyant lagoon
thats easy enough if you dont want wrapping
just add stride * row_offset to the starting addr
slender iron
it does wrap to make scrolling easier
buoyant lagoon
ah, i implemented exactly that on the rpi
slender iron
you could cheat it internally by making the buffer twice as long and saving values to two spots
buoyant lagoon
this was the original little-kernel gfxconsole
slender iron
on rpi you have plenty of ram
slender iron
displayio isn't fast for fullscreen updates now anyway
buoyant lagoon
i assume the current displayio terminal is using a ringbuffer for the display, but still has to memcpy() to paint it into the framebuffer?
slender iron
the tilegrid to framebuffer process is pretty heavy
buoyant lagoon
this is my re-implementation of littlekernel gfxconsole, using 2 sprites to draw the 2 halves of the ringbuffer
so it can scroll with zero copies
slender iron
it's more than just memcpy because you are doing tile indexing plus color conversion
nice!
it'd be very cool to hw accelerate displayio
buoyant lagoon
this will also allow dual monitor on all pi models
assuming you configure the displays first in config.txt
stuck elbow
@buoyant lagoon can it also do that when displaying on an spi display connected to the pi?
buoyant lagoon
and not just mirror, fully independent displays
slender iron
buoyant lagoon
<@204112578425782272> can it also do that when displaying on an spi display conn...
i cant see any reason why it would interfere with spi displays
slender iron
that'd be very cool. we don't support dual displays very well yet
buoyant lagoon
for the pi4, dual hdmi will just come up automatically
but i havent fully confirmed how pi4 hvs works
i was being stuborn, and trying to bring the pi4 up from reset with zero blobs π
stuck elbow
i cant see any reason why it would interfere with spi displays
I mean, can you use all that hardware acceleration with spi displays?
buoyant lagoon
ah, not yet
i would need to figure out how the transposer works
the transposer is basically a mode in the 2d core, where it just writes the fully composited frame back to ram
then you can just dma it out SPI
the 2d hardware natively supports dpi, smi, dsi, hdmi, and vec
dpi is just parallel digital color, like the hyperpixel
smi is undocumented, and is more of an 8080 style bus, where you assert a strobe for each write
vec is the ntsc/pal encoder
for anything else, you would just write it back to ram, and then dma it somewhere custom
buoyant lagoon
<https://github.com/adafruit/circuitpython/blob/main/supervisor/shared/display.c...
lets see...
splash_children looks like my best target
that seems to be an array of tilegrids that make up the screen
oh, one reason to use my custom Sprite class over TileGrid, is caching the displaylist
buoyant lagoon
void hvs_regen_noscale_noviewport(sprite_t *s) {
uint32_t *d = s->dlist;
// CTL0
d[0] = CONTROL_VALID
| CONTROL_PIXEL_ORDER(s->color_order)
| CONTROL_UNITY
| CONTROL_FORMAT(s->pixel_format)
| CONTROL_WORDS(7);
// POS0
d[1] = POS0_X(s->x) | POS0_Y(s->y) | POS0_ALPHA(0xff);
// POS2, input size
d[2] = POS2_H(s->bitmap->height) | POS2_W(s->bitmap->width) | (s->alpha_mode << 30);
// POS3, context
d[3] = 0xDEADBEEF;
// PTR0
d[4] = ((uint32_t)s->bitmap->data) // assumes identity map, should be physical addr
| 0xc0000000; // and tell HVS to do uncached reads
// context 0
d[5] = 0xDEADBEEF;
// pitch 0
d[6] = s->bitmap->stride * 4;
//printf("w: %d, h: %d, stride: %d, bits per value: %d\n", s->bitmap->width, s->bitmap->height, s->bitmap->stride, s->bitmap->bits_per_value);
}
but you have have tree of groups
buoyant lagoon
this is a 7 x 32bit structure, that describes a single sprite
if a sprite hasnt changed, then i dont have to re-generate this structure
and i can just copy it into the displaylist for the next frame
which saves some cpu
slender iron
so that's what tilegrid should maintain
and group should combine them to produce the displaylist
buoyant lagoon
ah, so the group is basically the same as my videocore.HvsChannel1.set_sprite_list([format1, format2, format3, format4])
is a tilegrid object bound to a single display driver?
stuck elbow
I think there is no explicit bond, but the display logic assumes only one display is displaying any object
because there is a linked list internally
buoyant lagoon
and on the rpi, all 3 hvs channels can consume the same format, and infact, share the displaylist ram
but each needs its own copy of the displaylist
basically, there is an uint32_t dlist[4096] in MMIO space
for each 1:1 scaled sprite, you write 7 slots
plus 1 slot to end the list
then you put the starting index into another control reg
a scaled sprite takes up 14 slots
a palette takes an extra 1 slot in the sprite itself, plus N slots to hold the N RGBA8888 colors
slender iron
I think tilegrid's ensure they are only in one group at a time
buoyant lagoon
so sharing a palette between multiple sprites will reduce the usage
slender iron
since groups have coordinates and scaling that propagate into the tilegrid
palettes can be shared
buoyant lagoon
and the TileGrid holds a pointer to the palette?
slender iron
ya, to the pixel_sharder
which is a palette or color converter
tilegrids are ordered by their order in groups
ie z-depth
buoyant lagoon
perfect
the entries in the hw dlist also have to be in z-depth order
it draws them first to last
internally, there is a FIFO that can hold a few scanlines
when that FIFO has room for 1 scanline, the hardware will start parsing the display list, and copying/blending pixels into the FIFO memory
it can draw 4 pixels/clock for unscaled sprites
2 pixels/clock for scaled sprites
and 1 pixel/clock for palette based formats
if it can draw all of the sprites in the scanline within the deadline (normal racing the electron beam stuff), then it can keep up and everything is fine
the FIFO gives it a couple scanlines of wiggle room, so it isnt too fragile
and the 2d core will take turns generating scanlines for up to 3 displays, each in its own FIFO
if no sprite covers a given pixel, then the color is undefined (whatever was already in the fifo), and you get weird effects
but there is an optional background fill, that will pre-write a configurable constant to the whole scanline, at the cost of more clock cycles
there is also hardware dithering support
for context, this was setup to do an r/g/b face from 0% to 100%
but the top 2 bits of green are mis-configured (hw uart conflicts)
in this example, dithering is off, so you can see some hard steps in the darker regions, when the camera isnt over-exposed
this image, then has hardware dithering enabled
but the top 2 bits are still mis-configured, which makes bit0 wiggling far more visible
slender iron
I think palette has a dither flag too
that or colorconverter
mainly for color eink improvements iirc
slender iron
for the rpi hardware, dithering is a screen-wide option
huh, interesting
buoyant lagoon
you define the output bit depth
24bit 888
18bit 666
16bit 565
or 15bit 555
and you define a dithering mode
0 = No dithering
1 = Accumulate rounding error horizontally.
2 = As 1, but also apply pseudo-random noise. The noise generator is reset at the start of each frame.
3 = As 2, but the noise generator is free-running. This setting will produce a shimmering effect on static images that may be distracting.
internally, it operates on rgb888, and then it will dither down to the configured depth, as it fills the FIFO with scanlines
https://forums.raspberrypi.com//viewtopic.php?f=41&t=45290&p=363475#p363475
this post explains how to configure dithering in config.txt
but the firmware/config format, only allow configuring it on 1 screen
while the hw lets you configure all 3 screens, with different dithering options
slender iron
ah, I see
buoyant lagoon
ive RE'd that part of the hardware, so i could expose it in circuitpython easily
but hdmi is already 8bit
most dsi i assume are 8bit?
composite i assume is 8bit, but its analog, hard to tell without a scope and test patterns
so only DPI really needs it
and then youve lost all your gpio
slender iron
we're just starting to support DPI
slender iron
most of what we support is 565 over spi or 8080
buoyant lagoon
there are 2 ways to do DPI
1: just use the legacy firmware options to setup DPI in config.txt, then it will show up as one of the hvs channels in my code
2: i know how to bringup DPI fully from scratch, so circuitpython could do all of the timing params
buoyant lagoon
most of what we support is 565 over spi or 8080
the SMI interface on the rpi lets you accelerate 8080 based interfaces
there is also an SMI video mode, but ive not figured it out yet
you configure how many clock cycles long the setup/strobe/hold/pace periods are
then you just load DMA up with a buffer, and point it at SMI, and let it rip!
slender iron
HDMI is the most interesting part of the pi for me anyway
but nice to know we could accelerate other displays too
buoyant lagoon
ive been struggling with hdmi init on the open firmware
but if your leaning on the closed firmware, its easy to hijack things after init is done
later tonight/tomorrow, i need to go over the tilegrid and group apis, write some example code, and then try throwing a group at my driver
also, how might i allow full rgba8888 or yuv images? the hw supports them
displayio.CIRCUITPYTHON_TERMINAL, ah nice, so thats just exposed to python
https://docs.circuitpython.org/en/latest/shared-bindings/displayio/index.html#displayio.Colorspace.RGB565
oh, and i have my demo backwards!
i'll need to update it then
idle owl
@buoyant lagoon Do you want to be added to the circuitpythonistas role? I dont' know if we've already offered it, or if you're even interested. But you've been super active lately with CP. The role means you get pinged twice a week about the Monday Community meetings. And your name shows up purple. Let me know if you'd like to be added! No worries if not!
buoyant lagoon
sure, add me
idle owl
Done!
buoyant lagoon
ive been listening in on those meetings, but somehow missed yesterdays
idle owl
Next week is Tuesday, FYI.
buoyant lagoon
ah, maybe it was last monday that i missed
idle owl
@buoyant lagoon The role also lets you speak during the meetings, if you're ever up for participating.
slender iron
also, how might i allow full rgba8888 or yuv images? the hw supports them
full alpha isn't something we have now. yuv would be another input and output colorspace I imagine
orchid basinBOT
manic glacierBOT
orchid basinBOT
manic glacierBOT
orchid basinBOT
buoyant lagoon
full alpha isn't something we have now. yuv would be another input and output co...
but yuv420 is complicated, by the luma plane being at half res, and the chroma plane being in a seperate array
orchid basinBOT
orchid basinBOT
buoyant lagoon
oh, on the subject of hdmi audio
you need to encode the audio data into a IEC958 format, which supports up to 8 PCM channels
then configure a DMA block to copy it into a hw fifo
other then the encoding step, that sounds very similar to how i2s and pwm would do audio on most other boards
does circuitpython have any pre-existing framework for managing the dma and ringbuffer, and keeping the buffer topped up with audio data?
karmic skiff
Is there an equivalent of the Decimal class?
Trying to port a normal python module over to CP. https://github.com/hrbonz/python-aqi/blob/master/aqi/algos/epa.py#L3
This worked for me https://github.com/jepler/Jepler_CircuitPython_udecimal
manic glacierBOT
mental nexus
Here is a barely-working GT911 CircuitPython driver for touchscreens, it reads and reports the first touchpoint. Unsure if the init code is required, but seems to work on my unit without it. Might be useful for some of these new dotclock displays. https://gist.github.com/kmatch98/097a34177b68480e511179f9eeb8d909
manic glacierBOT
orchid basinBOT
manic glacierBOT
orchid basinBOT
manic glacierBOT
buoyant lagoon
frame 6
frame 29
frame 52
frame 12
frame 35
frame 58
frame 18
i modified the C code, to expose the hw frame counter from the 2d core on the rpi
but just the framebuffer (which isnt accelerated yet), adds enough latency to prints, to make it skip over 20 frames, lol
frame 0
frame 4
frame 7
frame 11
frame 14
frame 18
frame 22
frame 25
frame 29
frame 32
frame 36
frame 39
frame 43
frame 46
frame 50
frame 53
frame 57
frame 61
disabling some code in board_init() removed that, and now its better
buoyant lagoon
delta: 3 scanline: 945/1024 overhead: 0.0359802 next: 0.05
delta: 3 scanline: 964/1024 overhead: 0.0360107 next: 0.05
delta: 3 scanline: 936/1024 overhead: 0.0370178 next: 0.05
delta: 3 scanline: 920/1024 overhead: 0.0359802 next: 0.05
delta: 3 scanline: 889/1024 overhead: 0.0370178 next: 0.05
delta: 3 scanline: 855/1024 overhead: 0.0369873 next: 0.05
and with a few more lines of code, it reports where the scanout is within a frame
manic glacierBOT
I have the same problem!
Firmware: adafruit-circuitpython-yd_esp32_s3_n16r8-en_GB-8.2.4.bin
Dev board: YD-ESP32-S3 N16R8, 2022-v1.3
After flashing the firmware and connected to it using putty, REPL did not work:
manic glacierBOT
lone sandal
manic glacierBOT
manic glacierBOT
manic glacierBOT
I haven't build CircuitPython in a while. Is there a way to collect the .bin artifacts for the Makerfabs 7" build or the HACKtablet?
I actually had to go digging to find mine.
manic glacierBOT
manic glacierBOT
brazen hatch
spare jacinth
Is that FTP from or to (or both) CircuitPython? π
brazen hatch
ftp server on CP
client on my thinkpad
I will get a ftp client on CP too, but much later
tbh, I don't see why this couldn't be a built-in module for the big boi boards.
it's not that complicated tbh..
crimson ferry
tbh, I don't see why this couldn't be a built-in module for the big boi boards.
...
PR it?
brazen hatch
I will first make it fully in python and then make it in C
Then I will π
It's honestly simplier than telnet
crimson ferry
storage.remount? SD?
brazen hatch
It just uses the os funcs in it's current form, so it will work with anything mounted
I didn't get as far as to write just yet..
I am doing the LIST command right now
I implemented ACTIVE, and will finish LIST with it.
then I will implement GET with PASV
In the core implementation we should prolly not have as many options as I have here though.
Maybe just PASV or it being should be settable by settings.toml
midnight ember
I was going to ask if that was on CP but then I was like nah no way. That's pretty amazing. π
buoyant lagoon
i saw mention before that the text console is a TileGrid
is it a bitmap being displayed in one, or is it one tile for every single character and then letting displayio draw the fonts?
brazen hatch
LIST is done, now onto modesetting and pasv
And I haven't even entered jank mode. It's all decent code thus far.
buoyant lagoon
π
buoyant lagoon
nice
random junco
<@&356864093652516868> Reminder - there is no meeting today due to the Labor Day Holiday in the US. Meeting is normal time tomorrow at 2pm EST / 11am PST. See you then!
upper ocean
Hey folks, does anyone here use VS Code when developing with circuitpython? I keep running into an issue where, due to the size of the circuitpython repo (ie all of its submodules), VS Code will often spawn dozens of rg processes whenever I do anything with the editor that could trigger indexing, searching, referencing, etc. It can sometimes get so bad that it completely maxes out my cpu and freezes VS Code.
I'm sure I can fix it by configuring VS Code to ignore certain paths, but thought I'd first check if anyone else here has run into this issue and already has a decent workaround? Thanks!
mortal kernel
Hey folks, does anyone here use VS Code when developing with circuitpython? I ke...
I've hit the same issue. My workaround is to checkout only the bits pertinent to the port I'm working on by using "make fetch-port-submodules" inside the port directory.
mortal kernel
Hey folks, does anyone here use VS Code when developing with circuitpython? I ke...
Also I've learned not the run fetch-port-submodules from the terminal inside VS Code as this gets things churning badly. Better to run it in a separate terminal and then open VS Cope when it's done.
upper ocean
Also I've learned not the run fetch-port-submodules from the terminal inside VS ...
@mortal kernel thanks for the input, I'll give that a try! I think that should still work when I'm testing changes to the core circuitpython code (eg in the py directory) since that isn't a submodule.
upper ocean
Also I've learned not the run fetch-port-submodules from the terminal inside VS ...
@mortal kernel Nice! Your suggestion is working quite well, VS Code is much more usable, thanks again!
manic glacierBOT
atomic summit
Hi folks. Based on there being new commits i the 8.2 branch, can I safely assume there's be an 8.2.5 or higher before 9.0 ?
tulip sleet
Hi folks. Based on there being new commits i the 8.2 branch, can I safely assume...
Yes, 9.0.0 has quite a ways to go. We'll be doing 8.2.x for a while. See for instance the previous post.
atomic summit
Yes, 9.0.0 has quite a ways to go. We'll be doing 8.2.x for a while. See for ins...
Ok, thanks Dan. I want to commit those FeatherS3 second I2C bus changes into 8.2.x as I did it in main thinking 9.0 was imminent - doh! I should have asked first!
tulip sleet
Ok, thanks Dan. I want to commit those FeatherS3 second I2C bus changes into 8.2...
np - if the PR is still open you can change the base
easier to merge from 8.2.x to main so we've been preferring 8.2.x instead of doing a lot of backports
atomic summit
Nah, it was merged already a while ago π¦
manic glacierBOT
manic glacierBOT
manic glacierBOT
manic glacierBOT
brazen hatch
And now we escale the situation to a safe mode issue..
I made a reproducible safemode with it..
brazen hatch
I just woke up and decided to break esp today.
manic glacierBOT
onyx hinge
Ahoy <@&356864093652516868> -- as yesterday was a US holiday, we're having the meeting TODAY in about 70 minutes. Take a moment to add your notes, and then if you can join us in the voice chat. https://docs.google.com/document/d/1P6AwndR-BBldT7jIzLpO72B9nSmnz7VJexO96ttI2O0/edit?usp=sharing
slender iron
i saw mention before that the text console is a TileGrid
is it a bitmap being di...
One tile per character from a bitmap font spritesheet
buoyant lagoon
ahhh, that will be an interesting stress test then
brazen hatch
I wonder, maybe it would be even better if I permit wifi conn (and only wifi), on reloads.
Cause now it only runs that stuff on boot.
@crimson ferry Opinion?
buoyant lagoon
assuming a basic 80x24 terminal, thats 1920 characters
and you need 7 words per sprite
which is 13k words, wont fit!
buoyant lagoon
One tile per character from a bitmap font spritesheet
so, to do text console on the broadcom port, i would need to render the text console into a bitmap, but treating the bitmap as a ringbuffer is allowed, and then i can sprite that onto the screen
slender iron
13k words won't fit where?
buoyant lagoon
the displaylist is only 4096 words long
and you need 7 words for each sprite/tile
so if you had one sprite for each character in the text console, you would run out after ~580 characters
buoyant lagoon
that would kinda negate the benefits of hw accel
so i was thinking, only something with a high tile count, like the text console, would pre-render into a bitmap
and things with a low tile count, can be fully hw accelerated
buoyant lagoon
ive also got another issue
how do i know when the tiles in a tilegrid have changed?
ive not found any hooks yet, and dont know if its timer based?
so ive just modified my test code.py to re-run board.DISPLAY.show() for each update
crimson ferry
<@360465312518504448> Opinion?
Without Web Workflow, raspberrypi persists wifi connection across reloads, espressif does not`. MicroPython acknowledges this as a bug https://github.com/micropython/micropython/issues/9505.
With Web Workflow enabled, I think the assumption is: always connected, though robust apps do need to manage the connection since there are ways it can fail that the core does not recover from.
brazen hatch
K, imma go make it reconnect on reloads and add that to the docs too.
crimson ferry
other core devs probably have more reasoned ideas about this
brazen hatch
Well, I am here and I can be pinged. I can also be pinged in the issue directly.
And it's not like I will (or can) merge it on my own.
The only reason it's not connecting upon reload, is because of the delay of starting up allothat
// TODO: Do our own scan so that we can find the channel we want before calling connect.
// Otherwise, connect will do a full slow scan to pick the best AP.
// We can all connect again because it will return early if we're already connected to the
// network. If we are connected to a different network, then it will disconnect before
// attempting to connect to the given network.
Yea, the slow scan, is a slow scan.
slender iron
how do i know when the tiles in a tilegrid have changed?
Tile indices are changed through the common_hal_displayio_tilegrid_set_tile() function
buoyant lagoon
so i would need to hook that function? but i want some tilegrids to render to a bitmap, while others update the hw accel
slender iron
I assume you'd reimplement it with your accelerated version of displayio
tulip sleet
@onyx hinge we can hear you
onyx hinge
give me a moment please
ember iris
If you're using a webbrowser sometimes refreshing the page helps when you can't hear someone
Tuesday meetings are weird π
midnight ember
off by one error but not an error
manic glacierBOT
slender iron
synthio doc bug? https://forums.adafruit.com/viewtopic.php?p=984421#p984421
gilded cradle
I can read
slender iron
I'm here too
random junco
I shared the wind chimes with Anne too, such a cool project
plucky tulip
This looks amazing, so much potential with such a big display
midnight ember
Very cool @random junco !
taking pictures of matrix panels is hard because of scan lines
random junco
quite!
midnight ember
congrats on the new office space @slender iron !
slender iron
midnight ember
Thank you for hosting @onyx hinge
ember iris
Thanks all! have a good week!
onyx hinge
@tulip sleet @slender iron do you want to take a moment to set a time to talk about 1.20? I am out FRIDAY, I'm busy much of THURSDAY afternoon, so today or tomorrow would be good if it's feasible for you two as well
slender iron
<@329766224093249548> <@252717193496756235> do you want to take a moment to set ...
I'm free later. heading home for lunch now and to deal with a grocery delivery. Free tomorrow too
tulip sleet
<@329766224093249548> <@252717193496756235> do you want to take a moment to set ...
today is fine; tomw i have someone coming to work on the boiler around 12-2 ET, so may have some interruptions
onyx hinge
OK, can we do it at around 16:00 ET / 15:00 CT / 13:00 PT ? I think that's about 80 minutes from now.
slender iron
i should be able to and will let you know if I'll be late
cedar veldt
Hi, I think I remember reading somewhere that the 9.x library bundle is waiting on something specific, is that right?
Also I imagine ESP-IDF v5 isn't going into CircuitPython 9 at this stage? https://github.com/adafruit/circuitpython/pull/6913
brazen hatch
Hi, I think I remember reading somewhere that the 9.x library bundle is waiting ...
For the merger of the micropython 1.20 changes iirc.
1.19 has gone through, causing mpy's version to change, but 1.20 will change it again
so libs won't be built for 9.x till that happens
midnight ember
and retooling mpy-cross to work with the new stuff they're doing.
brazen hatch
Also I imagine ESP-IDF v5 isn't going into CircuitPython 9 at this stage? https:...
that is still a wip last I checked
midnight ember
yup, without mpy there won't be any publicly released libraries, so at a minimum have to wait for that to happen.
you can still work on libraries in .py format if that's your goal
cedar veldt
Thanks, I see!
cedar veldt
you can still work on libraries in .py format if that's your goal
Or build them to mpy using mpy-cross matching the CircuitPython version I guess
midnight ember
with all the changes they're doing that would be a moving target as they update mpy-cross.
cedar veldt
that is still a wip last I checked
That's what I thought but it still has 9.0.0 milestone on it and 9.0.0 is in alpha already
midnight ember
alpha for the developers not for users, not even close to user ready and things will very much change before a 9.0 rc
cedar veldt
Ah so it could still happen?
brazen hatch
Or build them to mpy using mpy-cross matching the CircuitPython version I guess
If you must do mpy, like I do.. I got a module for that, https://github.com/bill88t/CircuitMPY
You can pass any version via ENV or leave it to autodetect it using the board and download it.
tulip sleet
That's what I thought but it still has 9.0.0 milestone on it and 9.0.0 is in alp...
we are not going to make any 9.0.0 releases before the .mpy file format stabilizes, after the MicroPython v1.20 merge. We have made tags, for labeling purposes
onyx hinge
Here is the notes document for next Mondayβs CircuitPython Weekly meeting. It is at the normal time of 11am Pacific / 2pm Eastern here on Discord. Everyone is encouraged to attend! Please add your hug reports and status updates even if youβll be attending the meeting - itβs super helpful! If you are unable to attend but would still like to include updates, feel free to include them in the notes and weβll read them off during the meeting. Hope to see you there! <@&356864093652516868> https://docs.google.com/document/d/1M51kkQVUANGjwe9rhnOsSs7U6e_x8YQp_7LXxGnx3Xg/edit?usp=sharing
midnight ember
i usually beta test releases looking for bugs, it's pointless for me to even try with 9.0 alpha builds currently because things will change too much between now and whenever they start merging and folding everything in for 9.0... it kind of hasn't even really started yet. the s3 bucket stuff is still using all 8.x libraries.
midnight ember
because it's all still on 8.x mpy and esp-idf. it's too soon.
cedar veldt
because it's all still on 8.x mpy and esp-idf. it's too soon.
OK, I'll leave it for a while! Just saw 9.0.0-alpha appearing and thought that meant it was time to start testing, but sounds like it's not.
manic glacierBOT
brazen hatch
Requesting rerun on spontaniously combusting action:
https://github.com/adafruit/circuitpython/actions/runs/6089030421/job/16521012570?pr=8367
slender iron
back to my desk now
brazen hatch
My PR's are in. I am finally done.
btw, now I can do pasv with web-workflow off
If I make a wrapper in ljinux that closes the telnet shell and immediately opens ftpd...
hehe
if then I detect the quit and swap back to telnet..
slender iron
@tulip sleet ready to chat?
tulip sleet
<@329766224093249548> ready to chat?
on the phone for a few minutes
onyx hinge
would be good to put this in, it fixes the script failing part way through https://gist.github.com/jepler/623db67bcdedba822b2e6a29d7136b18
slender iron
onyx hinge
@tulip sleet maybe you found this already but somebody did bluetooth classic for micropython https://github.com/shariltumin/esp32-bluetooth-classic-micropython/blob/main/debug/src/bt_spp_client.c
orchid basinBOT
mortal kernel
I'm hitting a problem with the 8.2.x branch. There are several uncommited changes on the branch which is a bit odd. Specifically, I'm encountering a problem building mpy-cross that looks related to "modified content" in tools/huffman. Wondering if anyone can lend me a clue?
I may have borked my forked repository by first forking just main and then several days later adding the 8.2.x branch from the upstream repository?
(.venv) rabeles@ub2004:~/Development/circuitpython-8xx$ make -C mpy-cross
make: Entering directory '/home/rabeles/Development/circuitpython-8xx/mpy-cross'
Use make V=1, make V=2 or set BUILD_VERBOSE similarly in your environment to increase build verbosity.
Traceback (most recent call last):
File "/home/rabeles/Development/circuitpython-8xx/mpy-cross/../py/maketranslationdata.py", line 30, in <module>
import huffman
ModuleNotFoundError: No module named 'huffman'
make: *** [../py/py.mk:274: build/genhdr/compression.generated.h] Error 1
make: Leaving directory '/home/rabeles/Development/circuitpython-8xx/mpy-cross'
(.venv) rabeles@ub2004:~/Development/circuitpython-8xx$ git status
On branch 8.2.x
Your branch is up to date with 'origin/8.2.x'.
Changes not staged for commit:
(use "git add <file>..." to update what will be committed)
(use "git restore <file>..." to discard changes in working directory)
(commit or discard the untracked or modified content in submodules)
modified: frozen/circuitpython-pcf85063a (modified content)
modified: frozen/circuitpython_ef_music (modified content)
modified: frozen/circuitpython_picoed (modified content)
modified: frozen/mixgo_cp_lib (modified content)
modified: frozen/pew-pewpew-lcd (modified content)
modified: frozen/pew-pewpew-standalone-10.x (modified content)
modified: lib/quirc (modified content)
modified: ports/espressif/esp-idf (modified content)
modified: ports/raspberrypi/lib/PicoDVI (modified content)
modified: ports/silabs/tools/slc_cli_linux (modified content)
modified: ports/stm/st_driver/stm32f4xx_hal_driver (modified content)
modified: tools/huffman (modified content)
no changes added to commit (use "git add" and/or "git commit -a")
(.venv) rabeles@ub2004:~/Development/circuitpython-8xx$
stuck elbow
I may have borked my forked repository by first forking just `main` and then sev...
you have to update the submodules every time to switch branches
mortal kernel
I did make fetch-all-submodules after checking out 8.2.x.
stuck elbow
if you enter the submodule directory and do git status, it will tell you exactly what's changed
you can do git reset --hard in that directory to force the files to the exact content of the checked out version
it will delete any changes that are not commited
mortal kernel
if you enter the submodule directory and do git status, it will tell you exactly...
No joy with hard medicine. Think I'll kill my forked repo on GitHub and fork it again.
stuck elbow
you don't need to fork it, just clone it again to a new directory
mortal kernel
you don't need to fork it, just clone it again to a new directory
Thanks, I got it!
manic glacierBOT
orchid basinBOT
idle owl
@tulip sleet If I have the Metro M7, with 8MB of QSPI flash, running CircuitPython, about how much flash memory is left over for me to do things with?
The RP2040 with 16MB flash had about half left. Is "half" correct, or is it a static amount?
tulip sleet
have to look, but you can just see how large CIRCUITPY is by asking the host OS to tell you
idle owl
Oh ok
I haven't gotten that far yet, so I figured I would ask.
I can verify it later. Or leave it out.
tulip sleet
i can plug one in and see
idle owl
That also works
tulip sleet
about 6.2MB
idle owl
Thank you
idle owl
I said "around 6"
manic glacierBOT