Implement experimental marshaler for OpenAPI V3

[Jefftree]

goos: linux
goarch: amd64
pkg: k8s.io/kube-openapi/pkg/spec3
cpu: Intel(R) Xeon(R) CPU @ 2.20GHz
                                               │  old1.txt   │              new1.txt              │
                                               │   sec/op    │   sec/op     vs base               │
OpenAPIV3Serialize/appsv1spec.json-24            30.60m ± 2%   12.41m ± 3%  -59.43% (p=0.002 n=6)
OpenAPIV3Serialize/authorizationv1spec.json-24   2.541m ± 1%   1.143m ± 1%  -55.04% (p=0.002 n=6)
geomean                                          8.818m        3.766m       -57.29%

                                               │   old1.txt    │              new1.txt               │
                                               │     B/op      │     B/op      vs base               │
OpenAPIV3Serialize/appsv1spec.json-24            10.586Mi ± 1%   5.175Mi ± 3%  -51.11% (p=0.002 n=6)
OpenAPIV3Serialize/authorizationv1spec.json-24    931.7Ki ± 0%   538.6Ki ± 0%  -42.19% (p=0.002 n=6)
geomean                                           3.103Mi        1.650Mi       -46.83%

                                               │  old1.txt   │              new1.txt              │
                                               │  allocs/op  │  allocs/op   vs base               │
OpenAPIV3Serialize/appsv1spec.json-24            34.00k ± 0%   20.03k ± 0%  -41.08% (p=0.002 n=6)
OpenAPIV3Serialize/authorizationv1spec.json-24   3.021k ± 0%   2.025k ± 0%  -32.96% (p=0.002 n=6)
geomean                                          10.13k        6.369k       -37.15%

Slightly less improvement than OpenAPI V2 which is somewhat expected because of the smaller size of OpenAPI V3.

[Jefftree]

/assign @apelisse
/cc @alexzielenski

[apelisse]

Thanks,
I've compared this to the v2 versino of the same PR and looked at each individual comments I got on that other PR. I checked all the fields to make sure nothing was missing. Looks good to me, two minor questions/comments.

[apelisse]

I'm personally good with this, lgtm & approve for me.
Anybody else wants to have a look @sttts @liggitt ?

[alexzielenski]

mostly lgtm with few nits/questions

[alexzielenski]

These extra structs caused by Refable and VendorExtensible make me sad but do not see a way around them without breaking changes to the type.

[alexzielenski]

Could changing these parameterPropsOmitZero, operationsPropsOmitZero, etc. types to a pointer avoid a copy and boost performance at all?

[Jefftree]

I tried something like

    var x struct {
        Ref            string                 `json:"$ref,omitempty"`
+       ParameterProps *parameterPropsOmitZero `json:",inline"`
        spec.Extensions
    }
    ...
+   tmp := parameterPropsOmitZero(p.ParameterProps)
+   x.ParameterProps = &tmp
    ...

and performance numbers look identical. Is this what you mean by changing them to pointers?

[apelisse]

Yeah, Jordan called out the same thing in my PR for V2, and I ran the same experiment :-)

[alexzielenski]

I meant using like this:

x.ParameterProps = (*parameterPropsOmitZero)(&p.ParameterProps)

In my other comment I had a toy example that seemed to compile at least. Doing the cast with the pointer this way may avoid a copy

[apelisse]

#325 (comment)

[Jefftree]

| Name                                               | Before       | After        |
| (appsv1)OpenAPIV3Deserialize/jsonv2-24             | 13.68m ± 1%  | 13.65m ± 1%  |
| (authorizationv1)OpenAPIV3Deserialize/jsonv2#01-24 | 1.199m ± 1%  | 1.192m ± 2%  |
| (appsv1)OpenAPIV3Deserialize/jsonv2-24             | 5.260Mi ± 0% | 5.260Mi ± 0% |
| (authorizationv1)OpenAPIV3Deserialize/jsonv2#01-24 | 525.4Ki ± 0% | 525.4Ki ± 0% |
| (appsv1)OpenAPIV3Deserialize/jsonv2-24             | 26.89k ± 0%  | 26.89k ± 0%  |
| (authorizationv1)OpenAPIV3Deserialize/jsonv2#01-24 | 2.517k ± 0%  | 2.517k ± 0%  |

[alexzielenski]

Yeah doesn’t seem to do anything

[apelisse]

/lgtm
/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: apelisse, Jefftree

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [Jefftree,apelisse]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment