validation: add args validation for process

[Mashimiao]

Signed-off-by: Ma Shimiao mashimiao.fnst@cn.fujitsu.com

[wking]

This check is too strict. Perhaps the user intends to copy the executable into the right location before running start (see earlier discussion here). Or maybe the user will bind-mount the command in from somewhere else. Anyhow, I think it's to complicated to try and make this a validation-time check, and we should just leave it to the runtime/kernel to error out if the command isn't there at execve-time (or whatever syscall the runtime is triggering with ‘start’).

[wking]

See also my notes here on the current spec phrasing being inconsistent. The in-flight opencontainers/runtime-spec#427 should help make the specification more self-consistent.

[Mashimiao]

@wking Thanks for you information. I think you are right. I will remove code which checks whether command exists or not
From runtime-sepc, it seems args is required not optional. So I think checking whether args is empty is necessary.

[Mashimiao]

sorry, It seems no need to check args is empty or not. I will close this PR.

[wking]

On Tue, Jun 21, 2016 at 06:47:39PM -0700, Ma Shimiao wrote:

Closed #116.

Wait, what? I was pushing back on the executable lookup 1, but I
think the current spec wording unambiguously requires at least one
entry in the process.args array 2, and after my suggested spec
changes that will still be the case 3.

[Mashimiao]

@wking there is a function checkMandatoryUnit in validation.go can check args wheter should be empty or not. So, I think there is no need to check it again.
Am I right?

[wking]

On Tue, Jun 21, 2016 at 09:58:49PM -0700, Ma Shimiao wrote:

there is a function checkMandatoryUnit in validation.go can check
args whether should be empty or not. So, I think there is no need to
check it again.

Ah, you're right:

$ ./ocitools generate
$ CONFIG=$(jq 'del(.process.args)' config.json)
$ echo "${CONFIG}" >config.json
$ ./ocitools validate
FATA[0000] Mandatory information missing: ['Process.Args' should not be empty.].

So I agree that this PR wasn't needed.