[pradeepsuryad]

Fixes #165.

This follows the proposal in #165, in three commits.

1. Install a CAN_RAW_FILTER for the registered receive ids. CANSocket gets add_recv_filter(can_id, mask) and remove_recv_filter(can_id, mask). They rebuild the filter array and call setsockopt(SOL_CAN_RAW, CAN_RAW_FILTER, ...). CANDeviceCollection::add_device() and remove_device() call them with get_recv_can_id() and get_recv_can_mask(). Until the first filter is added, and again after the last one is removed, the socket receives everything, so tools that register no devices are unaffected. CAN_RAW_ERR_FILTER is not touched.

Filters are counted per (id, mask), because the arm or gripper collection and the master collection hold the same devices. A filter stays until every collection that added it has removed it, which is why remove_recv_filter() takes the mask too. Re-adding a recv id to a collection adds the new device's filter before dropping the old one's, so frames for that id never stop in between.

2. diagnose, health: also listen on 0x00. With only the devices' own filters, every frame that passes a filter matches a device, so the RID 7 hint could never fire. A motor whose master id was never set replies on 0x00, which neither tool registers. Both now add a filter for 0x00 themselves, so those replies still arrive and are counted as unmatched. That is the "passes a registered filter but matches no registered id" case from the issue. I put this in the tools rather than the library because the hint is theirs, but I can move it if you'd rather.

3. Drop the frames queued before the first receive filter. The socket receives everything until the first device is registered. Whatever arrives in that window is still queued after the filter goes in, and the first recv_all() counted it as unmatched. On a busy bus that brings back the same false warning: 54 to 80 such frames in the flood test below. When the first filter goes in, the queue is now emptied first. Error frames in it still go into BusStatus.

Things to be aware of:

• A program that reads other ids through the same socket as its registered devices now needs to add a filter for them.
• add_recv_filter() isn't in the Python bindings yet. I can add it here or separately.
• Keep motor indices in the order passed to init_arm_motors() #176 also changes add_device() and remove_device(). I'll rebase whichever of the two lands second.

From reading the code, the other commands keep working:

• discover: I couldn't run it on vcan, because it reconfigures the bitrate. When it looks for a motor whose master id is unset, it already registers a device on 0x00, so those replies still get through.
• show_param, monitor and the status commands: they only read through registered devices, so the filters don't change what they show.
• Commands built on a bare CANSocket: they register nothing, so they still receive everything.

Test

No motors needed. On vcan0 (GitHub Actions, Ubuntu 22.04 and 24.04), small scripts stand in for the motors and for another process on the bus. diagnose --explain and openarm-can-health run against main and against this branch:

Scenario	main	This branch
Another process drives motors 0x01 and 0x02	diagnose lists 0x01 x117, 0x02 x117 and 0x7ff x234 as unmatched and prints the RID 7 warning, then "Nothing to report"; health prints unmatched: lines for the same ids	nothing unmatched; "Nothing to report"
Motor 0x02 replies on 0x00	0x00 reported as unmatched	the same
Another process floods 0x123	0x01 got 1 reply out of 101, and 0x123 x6463 is listed as unmatched	0x01 got all 102 replies; nothing unmatched

The counts are from 24.04, and 22.04 gave the same picture. With the first commit alone, the 0x00 motor goes unreported, and 54 to 80 flood frames from before the filter still show up as unmatched. Commits 2 and 3 fix those two cases.

A small program also checks the socket API directly:

• everything arrives before the first filter;
• nothing queued before the first filter is handed out after it;
• only the filtered id arrives after it;
• filters are counted across collections, including the same device in two collections and a device replaced in one;
• everything arrives again once the last filter is removed.

With the first commit alone, only the queued-frames case fails.

Run: https://github.com/pradeepsuryad/openarm_can/actions/runs/42085191330

The scripts are on a separate branch of my fork. I can add the socket check to the repository as a test if you'd like one.

[packit-as-a-service]

Congratulations! One of the builds has completed. 🍾

You can install the built RPMs by following these steps:

• sudo dnf install -y 'dnf*-command(copr)'
• dnf copr enable packit/enactic-openarm_can-178
• And now you can install the packages.

Please note that the RPMs should be used only in a testing environment.