snap panics on empty config sections

[nanliu]

If you have an empty section in snapd.conf.yaml, snapd will panic:
such as:

---
scheduler:

Error:

snapd --config /etc/snap/snapd.conf.yaml
panic: runtime error: invalid memory address or nil pointer dereference
[signal 0xb code=0x1 addr=0x0 pc=0x40b08f]

goroutine 1 [running]:
panic(0xb28b20, 0xc82000e0d0)
    /usr/local/Cellar/go/1.6.1/libexec/src/runtime/panic.go:464 +0x3e6
main.applyCmdLineFlags(0xc8201eeac0, 0xc8200ddb80)
    /Users/nanliu/src/intelsdi-x/snap_packaging/artifacts/src/github.com/intelsdi-x/snap/snapd.go:627 +0x66f
main.action(0xc8200ddb80)
    /Users/nanliu/src/intelsdi-x/snap_packaging/artifacts/src/github.com/intelsdi-x/snap/snapd.go:222 +0xe3
github.com/codegangsta/cli.(*App).Run(0xc8200ba180, 0xc82000a120, 0x3, 0x3, 0x0, 0x0)
    /Users/nanliu/src/intelsdi-x/snap_packaging/artifacts/src/github.com/codegangsta/cli/app.go:130 +0xadb
main.main()
    /Users/nanliu/src/intelsdi-x/snap_packaging/artifacts/src/github.com/intelsdi-x/snap/snapd.go:208 +0xa8e

[geauxvirtual]

We should handle this better, but with YAML, if nothing will be in a config section, it can be commented out completely or removed from the file.

Also, there is no need to specify .yaml or .json behind the default configuration file. snapd will automatically pick up a configuration file located with the name /etc/snap/snapd.conf whether it is JSON or YAML without needing to specify the --config flag.

[IRCody]

I don't think any config file should be able to make us panic, valid or invalid.

[tjmcs]

Agreed, @IRCody. I am currently looking into this issue; I think it's the fact that those "empty" sections map to nil values; stay tuned...

[geauxvirtual]

The problem is in how JSON unmarhalling happens (yes, JSON is used for the YAML file). Even if a section is labeled as omitempty in the code, having scheduler: appear in the file doesn't omit it because it's empty. It's only omitted if it doesn't appear in the file. Specific unmarshalling code would need to be added to check for if the element exists but is empty.

[tjmcs]

Yeah, that was my assumption as well, @geauxvirtual; I'll dig into it a bit more and see what it would take to check for empty fields after the configuration file is parsed (skipping those in the merge process?). As @IRCody said I really don't think that a valid YAML (or JSON) file should cause a panic, so even though we can get around it by commenting that section out we should fix this issue.