Development

Source Code

The core of the flow3rbadge codebade is st3m, with part of it implemented in Python, part in C. To work on both, you will need to clone the flow3rbadge firmware repository.

$ git clone --recursive https://git.flow3r.garden/flow3r/flow3r-firmware

Don’t forget the --recursive, otherwise you’ll get weird errors from missing submodules, like:

CMake Error at esp-idf/tools/cmake/component.cmake:313 (message):
 Include directory
 '.../flow3r-firmware/components/micropython/vendor/lib/berkeley-db-1.xx/PORT/include'
 is not a directory.

If you’ve already cloned without --recursive you can update your submodules the following way:

$ git submodule update --init

Dependencies

If you’re using Nix(OS), just run nix-shell nix/shell.nix.

On other Linux-based distributions, you will have to manually install ESP-IDF alongside our custom patches (note that install.sh installs stuff to your $HOME so that you may want to use a container or nix):

    $ git clone --recursive https://git.flow3r.garden/flow3r/esp-idf
    $ cd esp-idf
$ git checkout 5.1-flow3r
    $ ./install.sh
    $ source export.sh

To compile, see Working on C st3m code.

For running the simulator, you’ll need Python 3 with pygame and wasmer:

$ python3 -m venv venv
$ venv/bin/pip install pygame requests pymad
$ venv/bin/pip install wasmer wasmer-compiler-cranelift

Warning

The wasmer python module from PyPI doesn’t work with Python versions 3.10 or 3.11. You will get ImportError: Wasmer is not available on this system when trying to run the simulator.

Instead, install our rebuilt wasmer wheels using

venv/bin/pip install https://flow3r.garden/tmp/wasmer-py311/wasmer_compiler_cranelift-1.2.0-cp311-cp311-manylinux_2_34_x86_64.whl
venv/bin/pip install https://flow3r.garden/tmp/wasmer-py311/wasmer-1.2.0-cp311-cp311-manylinux_2_34_x86_64.whl

On macOS: the above might work.

On Windows: good luck.

Working on Python st3m code

You can use mpremote and similar to copy edited files from python_payload/:

$ mpremote cp python_payload/main.py :/flash/sys/main.py

TODO: document mpremote mount, it’s currently broken

As with application development, you can first check your changes using the simulator:

$ python3 sim/run.py

Working on C st3m code

Make sure you have ninja installed - CMake will happily generate code for Make if Ninja is missing, but it won’t necessarily work.

To compile:

$ idf.py build

To flash the main firmware only (without overwriting the FAT32 partition or recovery image), put the badge in Flashing (low-level) mode and run:

$ idf.py app-flash

Note: do not run idf.py flash as that will prevent you from going into recovery mode. If you’re flashing a factory-new badge, you also need to flash the recovery partition/bootloader/firmware first. See flashing recovery.

To clean, do not trust idf.py clean. Instead, kill everything with fire:

$ rm -rf sdkconfig build

To edit the sdkconfig temporarily:

$ idf.py menuconfig

To commit your sdkconfig changes to git, run menuconfig, press d, accept the default path. Then, copy over build/defconfig onto sdkconfig.defaults.

Flashing Recovery

Tl;DR use the following script to flash everything:

$ tools/flash-full.sh

The long story is that the main firmware codebase has a slightly different partition layout (as seen by the flashing tooling) than the recovery tooling. The one used in the recovery project (recovery/partitions.csv) is the correct one. However, we can’t use it as the main partitions.csv file as ESP-IDF performs magical detection from that file on where the build artifact should be located, and it always defaults to flashing to the factory image. Thus, in the real/recovery partition table the recovery firmware is the factory image, while the main firmware is in the ota_0 partition. But to make idf.py app-flash work in the main firmware repository, there the main firmware is marked as factory. But if you flash the main firmware’s partition table to the device, the recovery partition will stop working.

In addition to Different-Partition-Table shenanigans, the second-stage bootloader is also a problem. As with the partition teable, the correct one is the recovery one. Using this bootloader allows you to pick the recovery image on startup by holding the right trigger.

So, in order to have a functioning badge you shoud:

  1. Flash the partition table from recovery

  2. Flash the bootloader from recovery

  3. Flash the factory image from recovery

  4. Flash the ota_0 image from main

Or, in code:

$ (cd recovery && idf.py erase-flash flash)
$ idf.py app-flash

Thich is what tools/flash-full does.

printf-Debugging

All printf() (and other stdio) calls will be piped to the default Micropython REPL console. For logging, please use ESP_LOGx calls.

If you’re debugging the USB stack, or want to see Guru Meditation crashes, connect to UART0 over the USB-C connector’s sideband pins (TODO: link to flow3rpot).

gdb Debugging

You can also disable the TinyUSB stack and make the badge stay in UART/JTAG mode:

idf.py menuconfig -> Component config -> debug config -> usb gdb mode

Console output (including REPL) is not currently implemented in this mode.

Do a clean build with rm -r build; idf.py app-flash

In one terminal:

$ OPENOCD_COMMANDS="-f board/esp32s3-builtin.cfg" idf.py openocd

In another terminal:

$ idf.py gdb

If experiencing issues with ctrl-c, try calling gdb directly (reusing the build/gdbinit/gdbinit created by the above command)

$ xtensa-esp32s3-elf-gdb -x build/gdbinit/gdbinit build/flow3r.elf

Porting Doom (or other alternate firmware)

You should be able to use the flow3r_bsp component from any ESP-IDF 5 project. Either vendor the files, use a submodule and a symlink…

You should stay compatible with our SPI Flash Partitions layout. The easiest way to do that is to copy partitions.csv and refer to it from your own project. Your firmware should fit the factory slot.

Then, you can run your firmware by distributing the resulting .bin file and letting people flash to it via Recovery Mode.

For an example, see our doom port at TODO.

Alternative Firmware Projects

If you fancy playing with Rust on the flow3r, check out the flow3-rs project.

A port of Open-Smartwatch is also in the works.

Hardware Generations

If you’ve received your badge at CCCamp2023, you have a Production Badge and thus you don’t need to worry about this section. Congratulations!

For those who have a prototype badge, there’s an idf.py -g pX flag which you can use to get the firmware running on your hardware:

Badge Generation

Markings

Flag

Prototype 4

B4xx

dead

Prototype 3

B3xx

-g p3

Prototype 4

B4xx

-g p4

Prototype 5

B5xx

port me

Prototype 6

B6xx

-g p6 (default, same as prod)

NOTE: Anything older than p6 is not (yet?) supported by the recovery firmware.

Writing Docs

Automatically updated on CI runs of the main branch and lives under https://docs.flow3r.garden.

You will need sphinx and sphinx_rtd_theme installed. If you’re not usinx Nix, install these via venv:

$ python3 -m venv venv
$ venv/bin/pip install sphinx sphinx_rtd_theme
$ . venv/bin/activate

To build the docs locally:

$ cd docs
$ make html
$ firefox _build/html/index.html

To continuously build on change:

$ watchexec make html

Releasing

  1. Check out a version of main that you’d like to cut a release from.

  2. Create a new branch named release/[major].[minor].[patch], eg. git checkout -b release/1.2.3.

  3. Tag a the first release candidate: git tag v1.2.3+rc1.

  4. Build and perform QA (TODO: document).

  5. If the release canidate needs more work, cherry-pick fixes from main, tag a subsequent RC (eg. git tag v1.2.3+rc2) and go back to step 4.

  6. If the release candidate is ready to be released, tag a full release (git tag v1.2.3) and push branch/tags to gitlab. (TODO: build CI pipeline for release tags)