Closed grahamwhaley closed 6 years ago
OK, maybe now I understand the core issue, and a potential workaround.
Fundamentally, it is tied to golang packages, import paths and forks on github. We can see here: https://github.com/containers/virtcontainers/blob/master/cni.go#L25
that we directly import one of our own subdir packages:
cniPlugin "github.com/containers/virtcontainers/pkg/cni"
and, of course when you fork and do a normal go get or git clone from github if your fork, then the path where you are now building does not match that import path - and, hence the 'type clash' build failure.
One fix to get stuff to build is to rewrite all the imports to point at your new path (maybe using something like https://godoc.org/golang.org/x/tools/cmd/gomvpkg, or a script as found in the thread at https://stackoverflow.com/questions/14323872/using-forked-package-import-in-go), but that would then make a complete mess of your git diff
and trying to commit patches. A better solution (also on that stackoverflow page), is probably:
Anybody want to confirm this, and ack that adding a remote repo is the best workflow method? If so, I'll add it to the developer docs.
I will also likely add info on how to pin the vendoring in cc-runtime dep files so you can test virtcontainers changes with cc-runtime, unless somebody speaks up.
@grahamwhaley I think this is the way to go. I don't use this for virtcontainers since I push directly onto the main repo (I know it's wrong...), but I do that for Kata Containers and it works great. The Kata agent cannot build if you call into make from your forked repo, so I develop on the main one, and I push to a new branch on my forked one.
Adding this to the doc is the best thing to do since I am not sure there is something simpler we can do about it (as you've detailed it).
@grahamwhaley Adding a remote repo is the best workflow method. I usually clone the master/upstream repo and then continue with 2 remotes "upstream" (to point to the upstream repo) and "origin"(to point to my forked repo).
Right. My 'normal' workflow is to have those same two remotes, but to work in my forked repo :-) For vc, we need to flip that over. I'll PR a doc update.
I can build vc in its master checked out repo, but not in my user fork repo:
I believe this is a known issue - if somebody could add to this ticket any details of exactly what the problem with the CNI types are and why/how they are tied to the repo path, and any suggestions on how we effectively fix this - much appreciated.