I tested the latest version using release build. IMU was working and SPI communication seems to be stable (recorded for around 1 minute with no bad CRCs).

Should this be merged?

I think this branch contains some of the commits of #98, so the latter should probably be merged first to avoid trouble with the git history.

Thinking about it: Would it be possible to make the use of the power board and the associated protocol change optional via a build flag? My understanding is, that landing the power board support will create incompatibility between the sdk and all the master boards that are flashed so far. This will cause a lot of breakage that I would like to avoid.

What do you think @thomasfla @luator ?

This is an interesting point. However, having the support for power board optional will also be challenging to maintain later.

So far, the addition of new sensors / data fields directly impacted the length of the network datagrams exchanged between the robot and the control computer. It was the case for example with the IMU, which is optional on some setup, but occupies a data field in the protocol regardless of its presence.
It is also the case with the uDriver, which is fixed to 6 at compile time, even if your robot uses less than 12 dof.

Because of this, the users need indeed to update their SDK and FIRMWARE at the same time. To ensure this, the solution so far was to add a protocol version in the doc, firmware and sdk : 7e1786a

A more efficient way to deal with the situation would be to handshake on the different fields that the Master board needs to expose / parse. This would increase flexibility in the case of a user not asking for a sensor or command option that was not supported by the firmware he flashed in the past. More importantly, this would reduce the traffic on smaller systems which can be beneficial for wireless for example. Finally, this would open an easier support of new sensor types (For example, using lower cost IMU, being able to overwrite the RGB LED, set some configuration variables such as SPI frequency, retry, Timeout, etc.)
This looks complicated, but we introduced a change in the protocol with the INIT and ACK packets last year that makes this handshake easier to implement: https://github.com/open-dynamic-robot-initiative/master-board/blob/master/documentation/masterboard_communication.md#data-packet The challenges might be more on how to rethink all the fixed size data structures used by the firmware and SDK to parse and pack the data according to configuration unknown configuration at compile time.

This is clearly out of the scope of this PR obviously, but I took the opportunity to discuss here what I think would be a nicer long term solution.. We might want to open a feature request issue if we think this is the way to go..

To come back to this PR, I think we should merge it and advise the user to update their master board. Especially since this also include an important bug fix. I was reluctant to open the PR because of this compatibility issue, with only a partial support of the powerboard, but this is becoming quite old now.

I like the idea of a more flexible setup but I also agree with Thomas that it would be too much for now and we shouldn't block this PR for that.

To handle the breaking changes on the short term (and in general), I suggest to use releases following some form of semantic versioning. So we would make a release (using git tag and ideally also the GitHub release functionality) with a minor or patch version increase now, before we merge the changes. Any change that breaks compatibility between firmware and sdk would then be indicated by a major version increase.
This way, it is easier for users to see if they need to update the firmware after an SDK update and with the tag they can also relatively easily stick to the older version if they don't want to update directly (e.g. due to a deadline ahead).
Ideally the firmware should have a way to report its version for this (not sure if it already has), so the SDK could check the firmware version and show a proper error in case of incompatibility.

Of course this does not cover other packages which depend on updates of the SDK, so there could still be issues with those.

I like the idea about semantic versioning and we definitely should add some code (if possible) that tells the user clearly the SDK and firmware version are out of sync as a error message to the command line. Also a post to the ODRI form would be helpful.

Right now this PR has changes for the power board and the mailbox changes. I propose we cleanup this PR to have only the mailbox changes. My understanding is that these changes do not need an update to the SDK. We can land these changes first and then the power board changes. From looking at this PR this would boil down to doing a cherry-picking of 768f83f, 4706076, 4e3c911.

The reason I would like to do things this was is that we can land the mailbox improvements, which are important to stabilize experiments (we get hit with SPI timeouts quite often in New York) and give the other PR about the power board some more time until the power board is ready/used more often before landing on master.

What do you think?

I am happy to do the cherry-picking/new PR with only the mailbox changes. I don't have access to hardware anymore (I graduated). @luator , could you test a PR that has only the mailbox changes in Tübingen maybe?

Also, does anyone has an idea where the 17 minute maximum runtime error comes from at this point? We are planning to use the NYU Fingers for longer-running manipulation experiments.

(I graduated)

Oh, I missed that, congratulations!

Separating this to make it independent of the powerboard-related changes sounds good to me (assuming that it is easily possible). I have a setup here, were I could test it relatively easily.

Regarding the 17-min-issue: All I found out so far is that it fails around 17m 45s-50s after the boards are powered one (i.e. independent of when the software on the PC-side is started). My guess is that there is some time-related counter somewhere that is overflowing but this is really just a guess.

I'm also in favour of splitting the PR in two, especially if @jviereck proposes to do it and @luator to test it ! ^^

I'm also in for the versioning. If we could also provide a binary of the firmware for every release, that would be much easier for the end user to update. > #101

For informing the user on protocol version mismatch on the PC side, we will need to slightly change the protocol to do it.
Right now, the PC sends a INIT packet:

The masterboard checks the protocol and only send an ACK packet if the protocol version is valid.

We might change this packet for a larger one containing the protocol used by the firmware, but we might want to think about it to insure interoperability..

About the 17 minutes problem, let's open an issue to discuss it. > #108

Is there already a firmware version stored that can be queried from the PC? I think independent of the protocol version, this would be something nice to have, so we can check which version is currently running.

So maybe the new ACK packet could include the firmware version, the protocol version (albeit this could be derived from the firmware version, so might not be needed) and probably a flag indicating if the protocol matches or not.

I merged #98 so someone need to clean the git here.
@thomasfla @nim65s or @luator ?

@thomasfla you need to take a look at this PR and resolve the conflicts.
I can help if needed but it's mainly your code.

Already implemented