Revendor containerd/containerd to 5c72f92a5

[TBBle]

Walk in the snapshots.Snapshotter interface added a parameter, and the introspection.Service interface was refactored leading to a simpler Plugins call.

The prompt for revendoring is to remove the transitive CGO requirement on Windows, removed from containerd in containerd/containerd#3970.

This maintains the hand-managed 1.4.0 version per #1205

> go mod tidy
> # Fix pseudoversions, to make Go Modules happy
> go get github.com/containerd/containerd@5c72f92a5d924fdd699e761d022991266a77ed51
> go mod vendor
> # Revert pseudoversions, to make module-resolution rational

This process looks correct, but the diff shows a lot of // +build ignore files being deleted from other vendored modules, so I suspect this is not the same process others have used.

It also newly-vendored github.com/google/uuid, I'm not sure why, it doesn't appear to be a containerd/containerd change, so it's probably also due to using go mod vendor.

Full commit history for containerd is containerd/containerd@acdcf13...5c72f92, covering around 3.5 months of development, and 137 non-merge commits.

[AkihiroSuda]

duplicate: #1297

[TBBle]

duplicate: #1297

#1297 doesn't appear to bump containerd or hcsshim.

Edit: Whoops. It seems to have bumped it, but didn't change go.mod in the process. It's not bumped high enough though...

Perhaps I should rebase this on top of #1297?

Edit: I see now. #1297 is doing a replace for containerd and docker, but with a different commit to the entry in the require list, which seems like a mistake.

[TBBle]

WIP as it actually doesn't achieve what I want yet... The hcsshim version from containerd isn't visible to the go modules system as containerd doesn't have a go.mod. Instead it's picking up 0.8.5 as an indirect from tonistiigi/fsutil, which still has the CGO dependency.

[TBBle]

Hmm. Build fails due to

#10 35.69 verifying github.com/containerd/containerd@v1.4.0-0.20200125045834-5c72f92a5d92: checksum mismatch
#10 35.69 	downloaded: h1:M7RF1z0Sola5l99gmReufIg1oTwXVmupkhGq3KPSOto=
#10 35.69 	go.sum:     h1:Kxbw/mnIUsS8jFWr0RmvrCshoD4+ArcZeL1da0w5JgM=

https://sum.golang.org/lookup/github.com/containerd/containerd@v1.3.1-0.20200125045834-5c72f92a5d92 gives the same checksum as go.sum, but I guess the build script is running go 1.12, and gets the checksum from somewhere else.

Same thing is true of the version that's in master, https://sum.golang.org/lookup/github.com/containerd/containerd@v1.3.1-0.20191014053712-acdcf13d5eaf shows a different checksum to the committed go.sum.

I assume this is a 1.12/1.13 difference. -_-

[TBBle]

Putting this aside for now. I'll wait for #1297 to land before I play with this again, as it's already trying to work though the issues I'm expecting to see here.