Removing dave from dave_sim.repos

[bsb808]

At this point this is just to link to the comment in the merged PR so that it doesn't fall through the cracks. @M1chaelM is probably best positioned to describe what needs to be done post 4.3.0 release.

See: #199

This looks good, and we're going to try it out. It sounds like removing the dave repos from dave_sim.repos would ultimately make sense also, though we will do that later just for logistical reasons relating to our upcoming release.

[M1chaelM]

This is an optional optimization, so I was going to ask in our meeting whether we want to do it.

The situation is that when vcs was set up, the person who created the dave_sim.repos file inlcuded the dave repo in the file. This leads to a little bit of redundancy because you have to retrieve the dave repo to get the dave_sim.repos file in the first place. Then when you use vcs to get the rest of the repos for the first time, you have to ignore the fact that you downloaded the dave repo already. After the first time, it's very easy to use vcs to update all the repos at once, including the dave repo.

We've been living with this tiny bootstrapping requirement and using the --skip-existing flag to avoid getting an error the first time we run vcs to clone all the repositories. Since we are usually not wiping out our workspace and starting from scratch, this is not a big problem.

Now that we are moving to a more sophisticated CI script that is going to rely on our dave_sim.repos file, having the dave repository in it makes the CI take a bit longer. The reason is that the CI script process is to automatically build and test the main repo (dave) and then build and test everything in the dave_sim.repos file. Since that file currently includes the dave repo itself, we end up testing it twice.

All of this is just a restatement of what our kind contributor pointed out when submitting the PR to add industrial_ci. There are two ways to avoid this redundancy and make the CI faster:

1. Remove the dave repo from dave_sim.repos. Our contributor says this is "preferred." If that's the case then it means we have probably not been using vcs in the best way, and there is some better workflow. This seems plausible but I'm not yet sure what that workflow is, so we need to learn a little more about vcs. So this change would affect our current workflow and installation instructions.
2. Keep dave_sim.repos the same, but create a new file, say dave_ci.repos that we use just for CI. This gets us the efficiency in the CI without need to update our own brains or processes. However, I agree with the contributor that this is not preferred. Having two versions of anything always incurs a synchronization cost and is very error prone, so I think there's a strong likelihood that somewhere down the line we'll update dave_sim.repos and forget to update dave_ci.repos and then have a CI workflow that doesn't test what we want it to test. So I'm hoping we can figure out the right way to use vcs and go with option 1.

[M1chaelM]

Fixed the title since this issue is not about moving the dave_sim.repos file.

[M1chaelM]

After some further testing, it looks like there is no good reason to keep the dave repo listed in the dave_sim.repos file. We still have the same install process but we no longer need the --skip-existing option in the initial import. After the initial import, vcs does not rely on the .repos file for things like pulls and updates. Instead, it automatically finds all the repositories under the present working directory by default, so it will include dave anyway.

In light of this, I'm going to make a simple PR to update the repo file.

[bsb808]

Great description @M1chaelM - thank you.

[woensug-choi]

Closing this since the PR is merged