start validation test panics if process was not set

[dongsupark]

#578 introduced a new test for StartWithProcUnsetGenError, which is checked by validation/start.go. However, when testing with runc, the test fails because Create() returns exit status 2. It turns out, that's because validateProcessSpec() in runc panics with nil-dereference when its input spec (type of *Process) is nil.

From my understanding, the original idea of the test was probably not to test whether spec.Process is nil or not. It should actually test if process in config.json is valid. So I think we should actually decode a config json file to see if it has a valid process entry.

/cc @liangchenye

[wking]

It turns out, that's because validateProcessSpec() in runc panics with nil-dereference when its input spec (type of *Process) is nil.

I have a patch in flight for this with opencontainers/runc#1726.

[wking]

From my understanding, the original idea of the test was probably not to test whether spec.Process is nil or not. It should actually test if process in config.json is valid.

From runtime-spec, you should be able to create but not start a container with an unset process (more background on that in opencontainers/runtime-spec#701). The purpose of the StartWithProcUnsetGenError test is to exercise that usecase and ensure that create works (as the spec expects) and the start fails (as the spec requires).

[dongsupark]

@wking Thanks.
I hope opencontainers/runc#1726 could be merged.

[wking]

I hope opencontainers/runc#1726 could be merged.

Maybe you can find the maintainer-convincing argument that has so far eluded me ;).