inception: Reset undefined fields when reusing objects.

[cmeon]

When reusing objects with sync.Pool, for example, and unmarshaling
objects, this resets the old fields, which are not defined in the
current JSON to their Zero values.

For example:

type M struct {
     Y []int
}

type A struct {
     I int
 T *M
}

type B struct {
     X int
     Z *A    `json:"Z,omitempty"`
     Y int   `json:"Y,omitempty"`
     K []int `json:"K,omitempty"`
}

buf0 = []byte(`{"X":1,"Z":{"I":3, "T":[3,4]},"Y":4,"K":[1,2,3]}`)
buf1 = []byte(`{"X":1,"K":[1,2,3]}`)

x0 := B{}
x0.UnmarshalJSON(buf0)
x1 := x0
x1.UnmarshalJSON(buf1)

• This ensures that x0 != x1 by setting the fields in x1.Z = nil and x1.Y = 0
  after marshalling all JSON objects.

[erikdubbelboer]

All of this should be properly indented with tabs so the resulting code doesn't change with gofmt (like the other templates here).

[pquerna]

While nice, it doesn't matter too much, the generated code is all ran through fmt at the end, so the templates don't need to be whitespace perfect

[erikdubbelboer]

Looks very useful and would surely prevent bugs.

I think it should get some test cases?

[erikdubbelboer]

@pquerna did you have a look at this?

[erikdubbelboer]

@pquerna I just updated the pull request with code that hides this feature behind a command line flag. This is needed for ffjson to continue being a drop in replacement for the normal json.

I also improved the test case and squashed all commits into one. I also did some benchmarking and it doesn't affect the performance at all, which is expected as it's only a couple of if statements.

I think you can really merge this now so other people can also use it.

[erikdubbelboer]

Oh and I also fixed make clean and made sure make test rebuilds all the files as otherwise the tests might be using old generated files.

[erikdubbelboer]

@pquerna is this project dead?

[erikdubbelboer]

Thank you.