Reorganize repository for unified release - #39
Conversation
Remove NSI scripts and move towards dotnet tool deployment.
Brings the repository into compliance with the standardised build infrastructure.
glopesdev
added
documentation
|
It would be nice to also generate the register table for the API using Otherwise, the documentation portions of this PR look good to me! |
|
Hi @glopesdev! @banchan86 and I are working on having the first few docs sites ready by end of September and are looking forward to having them served on harp-tech.org. I wanted to check what is the timeline for this repo reorganization so we can coordinate. The Sound Card and Timestamp Generator repos are also part of that first doc upload. Thanks for letting us know. |
|
@ChucklesOnGitHub that sounds very much on target, with the release of the python interface we will be focusing on wrapping this next. |
The workflow-images job reads the need-workflow-image-render output now declared by build-interface. The condition previously named a job that does not exist, so workflow-images never ran.
The software folder now holds one subfolder per toolchain. The .NET projects, the solution, the shared build files and the license move into software/dotnet.
The package is harp-behavior and the module is harp.device.behavior. The interface is generated with the package option, so the module is an __init__.py beside a py.typed marker. The version comes from setuptools-scm, configured to read only plain semver tags, since the default pattern would resolve to a firmware tag otherwise. The tests assert that the device identity matches the metadata and that the register map merges the core and application registers.
Generation moves into a verify-generated job that regenerates every target and fails when the committed output differs, so the check runs once rather than on every leg of the interface matrix. The interface job becomes build-dotnet, and every build job now waits on the check, including firmware, so no artifact is published from a tree whose generated code has drifted. A build-python job runs the tests and builds the wheel on Python 3.11 to 3.13, and a publish-pypi job attaches the distributions to the release and publishes them with trusted publishing.
The valve mimic moves out of app_ios_and_regs.h into a hand-written mimic.h, which wraps every pin macro it needs as open_POKEn_VALVE, close_POKEn_VALVE and toggle_POKEn_VALVE. The IR half of the same feature was already called from interrupts.c, so both halves now sit outside the generated header. The three headers are now exactly what the generator emits, and verify-generated regenerates them and fails on any difference. The raw set, clr and tgl pin macros still exist and no longer mimic, so a call site must reach for the wrapper.
A .python-version file sets the default interpreter to the oldest supported version, so local work runs against the floor of requires-python. Because the file also overrides --python on the sync step, the CI job now selects the matrix version through UV_PYTHON, which takes precedence over the file.
aspaNeuro
left a comment
aspaNeuro
left a comment
There was a problem hiding this comment.
Tested with mircoharp converging with this setup and approach
4 participants
Adopt the standardised repository layout described in harp-tech/protocol#145, aligning this repository with the conventions shared by other Harp components and enabling a unified release of firmware, software, and documentation from a single CI pipeline.
Repository layout and build infrastructure
Firmwarefolder becomesfirmware,Hardwarebecomeshardware, andInterfacebecomessoftware, which now holds one subfolder per toolchain,software/dotnetandsoftware/python. A newdocs/folder is introduced as the root for repository documentation and assets.software/dotnet/build/, includingCommon.csproj.props,Common.csproj.targets, andPackage.props, for consistent project metadata and packaging across software components.Generated code is verified
A
verify-generatedjob regenerates every target fromdevice.yml, the ATxmega firmware headers, the C# interface and the Python package, and fails when the committed output differs. It runs first and every build job depends on it, so no artifact is published from a tree whose generated code has drifted.Bringing the firmware under that check required moving one device feature out of the generated header. The valve mimic had been hand-edited into
app_ios_and_regs.h, so a new hand-writtenmimic.hnow wraps the generated pin macros asopen_POKEn_VALVE,close_POKEn_VALVEandtoggle_POKEn_VALVE, and the call sites use those. The IR half of the same feature was already called frominterrupts.c, so both halves now sit outside the generated header. The three headers are now exactly what the generator emits, and the built firmware image is unchanged from a build of the previous revision, so nothing about the device changed.The
harp.toolkitpin moves to 0.4.0, which is what supplies the Python target and the per-language interface subcommands, and the C# interface is regenerated against it.Python device interface
Publish the generated Python interface as
harp-behavioron PyPI, importable asharp.device.behavior, alongside the existing NuGet packages. The module is generated as a package initializer with apy.typedmarker, so it resolves under a type checker and can be pinned like any other dependency.CI builds and tests it on Python 3.11, 3.12 and 3.13, and publishes the distributions to PyPI through trusted publishing on release. The version comes from setuptools-scm, restricted to plain semantic version tags so the firmware tags in this repository are not mistaken for releases.
Cross-platform App packaging
Replace the platform-specific NSIS installer with a cross-platform
dotnet toolpackage for the GUI App, installable on any platform with the .NET SDK and launchable asharp.behavior.Notes
After merging, contributors will need to run
git submodule update --init --recursiveanddotnet tool restorewhen checking out the repository.Firmware developers should now use the wrappers in
mimic.hrather than the rawset_POKEn_VALVEmacros, which still exist but no longer mimic. Regenerating with--implementationwould also rewriteinit_ios(), the one place where a raw call would be reintroduced silently.