panic: canary - *should* be impossible to have a pkg-only selection here

[jmank88]

With latest dep (536f25f), running dep init on the latest github.com/iron-io/functions (32278fe2e02e38737fe87be32975255b81712f17) produces a panic that I don't see an open issue for.

Full Log
Relevant error:

panic: canary - *should* be impossible to have a pkg-only selection here

goroutine 1 [running]:
github.com/golang/dep/vendor/github.com/sdboyer/gps.(*solver).backtrack(0xc420886a20, 0xc420acf500)
	/home/jordan/go/src/github.com/golang/dep/vendor/github.com/sdboyer/gps/solver.go:953 +0xb34
github.com/golang/dep/vendor/github.com/sdboyer/gps.(*solver).solve(0xc420886a20, 0x0, 0x0, 0x1c)
	/home/jordan/go/src/github.com/golang/dep/vendor/github.com/sdboyer/gps/solver.go:409 +0x787
github.com/golang/dep/vendor/github.com/sdboyer/gps.(*solver).Solve(0xc420886a20, 0x30, 0xa63080, 0xab1138, 0xc420014098)
	/home/jordan/go/src/github.com/golang/dep/vendor/github.com/sdboyer/gps/solver.go:355 +0xab
main.(*initCommand).Run(0xab1138, 0xc420126d50, 0xc42000c460, 0x0, 0x0, 0x0, 0x0)
	/home/jordan/go/src/github.com/golang/dep/cmd/dep/init.go:156 +0xfbb
main.main()
	/home/jordan/go/src/github.com/golang/dep/cmd/dep/main.go:125 +0x6df

[jmank88]

It is reproducible, but it is not related to the preceeding Cached ... line (maybe that is obvious to someone looking at the code side)

[sdboyer]

@jmank88 canaries are working, this is probably a bug! 😄 thanks for the issue report.

Could you re-run with dep init -v? That'll give the solve trace output, which is essential for determining where things went wrong in the solving algorithm. Thanks!

[jmank88]

dep init -v https://gist.github.com/jmank88/1d1cb545f774ff3d34afb9bd15670200

[sdboyer]

Well, it's very odd that you'd end up on that panic (and that still indicates a bug). But it seems like you shouldn't have gotten here at all; it appears that there's another bug where it's not picking up the full version list for github.com/docker/distribution: https://gist.github.com/jmank88/1d1cb545f774ff3d34afb9bd15670200#file-dep_init_-v-txt-L490

That line is super wrong - there should be at least 40 or so more versions available to try.

[timoreimann]

Running the current latest version of dep (commit d010781), I am experiencing similar problems running dep init on containous/traefik, specifically commit a0d6594.

The verbose log output can be found here. FWIW, traefik also uses github.com/docker/distribution.

Let me know if I can do anything to help analyze the problem.

[jmank88]

I wonder if this could be related to the docker->moby name change.

[sdboyer]

While the docker->moby name change is...well, super fun, it's unlikely to be causing the problem. GitHub redirects requests to the old name. And we can see docker/docker showing up just fine in @timoreimann's log output.

To be clear, the problem here is, unequivocally, a solver bookkeeping problem. The canary panics in the solver (there are currently seven) are in place to identify situations where the solver's internal state invariants have been invalidated. This particular one could easily just be wrong, though - and the more I stare at it, the more I think it is.

[sdboyer]

Yes, looking at it more, that panic just shouldn't be there, as it's a perfectly valid case. Will get this fixed promptly after #453 😄

[sdboyer]

OK, this issue should be gone - it got through a dep init -v for containous/traefik correctly (nicely exposing some of the bookkeeping errors in our trace output - wow do those package selection counts get ridiculous!). I didn't try iron-io/functions, but it should be fine.

Man, does it feel good to be able to just fix something directly 😄

[timoreimann]

holy moly, that's what I call a quick fix (and I don't mean the bad kind)! 🚀

Thanks @sdboyer! 👏