Premarshal struct does not maintain JSON tags (omitempty) about genqlient

[benjaminjkraft]

Thanks for the clear repro! It took me a while to understand the subtle case even looking at all the code so the fact that it was all there was very helpful.

I think you're right that we should just propagate the json tags. It's happening in the first place because the premarshal types are generated in a separate code path (regular types, premarshal types), which arguably we could change. But anyway that's probably not necessary right now, probably we can just add a check for omitempty in the premarshal code.

I'm hoping to have some time to work on genqlient this weekend and if I do will try to get to this, but if you are excited to, I think this would actually be a good first PR; the code change is probably one line (in the premarshal types linked above; I think the data we need is already available) and the main thing is just adding some unit and integration tests to show that it works.

[kevinmichaelchen]

Hey @benjaminjkraft — glad my repro was helpful!

Me, taking a guess

If I had to take a crack at it, I'd say we want to add

{{if .Omitempty -}},omitEmpty{{end}}

in the necessary places.

For example, we'd change this:

    {{.GoName}} {{repeat .GoType.SliceDepth "[]"}}{{ref "encoding/json.RawMessage"}} `json:"{{.JSONName}}"`
    {{else}}
    {{.GoName}} {{.GoType.Reference}} `json:"{{.JSONName}}"`

to something like this:

    {{.GoName}} {{repeat .GoType.SliceDepth "[]"}}{{ref "encoding/json.RawMessage"}} `json:"{{.JSONName}}{{if .Omitempty -}},omitEmpty{{end}}"`
    {{else}}
    {{.GoName}} {{.GoType.Reference}} `json:"{{.JSONName}}{{if .Omitempty -}},omitEmpty{{end}}"`

where Omitempty comes from goStructField.

Contributing

If that sounds like the correct approach, I can try to put up a PR and look into adding some test cases where needed.

[benjaminjkraft]

Yep I think that's right! Good catch that we need the change in both places.

[kevinmichaelchen]

Hey @benjaminjkraft — threw up a PR here: #267

Still need to sign the CLA and add a changelog entry.

Should I add/modify documentation anywhere?

[kevinmichaelchen]

In the meantime, I'll try to run my fork and see if it generates Hasura sub-sub-fields in the expected way.

Update: I ran it against nested BoolExp structs, and it looks like I'm still missing something!

[benjaminjkraft]

#267