## Summary: When genqlient generates output types, it generates whatever code is necessary to unmarshal them. Conversely, when it generates input types, it generates whatever code is necessary to marshal. This is all that's needed for genqlient itself: it never needs to marshal output types or unmarshal input types. But maybe you do! (For example, to put the responses in a cache, which is the use case that @csilvers hit at Khan, although there are others one can imagine.) While we can't support every serialization format you might want (at least not without adding plugins or some such), it's not unreasonable to expect that since genqlient can read JSON, it can write it too. Sadly, in the past this was not true for types requiring custom unmarshaling logic, for several reasons. In this commit I implement logic to always write both marshalers and unmarshalers whenever they're needed to be able to correctly round-trip the types, even though genqlient doesn't do so. I wasn't starting from scratch, since of course we already write both marshalers and unmarshalers in some cases. But this ended up requiring surprisingly large changes on the marshaling side, mostly to correctly support embedding (which we use for named fragments). Specifically, as the comments in `types.go` discuss, the most difficult issue is spreads with duplicate fields, which translate to Go embedded fields which end up hidden from the json-marshaler. Ultimately, I had to do things quite differently from unmarshaling, and essentially flatten the type when we write marshaler. But in the end it's not so ugly -- indeed arguably it's cleaner! Mainly it's just different. One thing to note is that we do marshal `__typename` based on what we know about the types; users need not fill it in (and if they do we'll ignore it). This seemed to me to be a better UX, and didn't add much complexity. In general, I begin to wonder whether using `encoding/json` at all is really right for genqlient: we're doing a lot of work to appease it, despite knowing what our types look like. I think it would still be a significant increase in lines of code to roll our own, but that code would perhaps be simpler, and would surely be faster (although if we just want the speed gains we could use another JSON-generator library, see also #47). Anyway, something to think about in the future. ## Test plan: make tesc Author: benjaminjkraft Reviewers: csilvers, StevenACoffman, benjaminjkraft, dnerdy, aberkan, jvoll, mahtabsabet, MiguelCastillo Required Reviewers: Approved By: StevenACoffman, dnerdy Checks: ✅ Test (1.17), ✅ Test (1.16), ✅ Test (1.15), ✅ Test (1.14), ✅ Lint, ✅ Test (1.17), ✅ Test (1.16), ✅ Test (1.15), ✅ Test (1.14), ✅ Lint Pull Request URL: https://github.com/Khan/genqlient/pull/120
155 lines
7.2 KiB
Cheetah
155 lines
7.2 KiB
Cheetah
{{/* We need to generate UnmarshalJSON methods for some types that we want to
|
|
handle specially. Specifically, we generate an UnmarshalJSON for each
|
|
struct with a field meeting any of these criteria:
|
|
- a field whose type is configured with a custom unmarshaler
|
|
- an embedded (anonymous) field; technically we only need an
|
|
UnmarshalJSON if the embedded type needs one, but it's easier to
|
|
generate it unconditionally
|
|
- a field of interface type
|
|
Additionally, since we add `json:"-"` to fields we handle specially, for
|
|
any field which requires a MarshalJSON, we also generate an UnmarshalJSON,
|
|
and vice versa.
|
|
|
|
Given that, we want to specially handle the above-described fields, but
|
|
unmarshal everything else normally. To handle fields with custom
|
|
unmarshalers, first we unmarshal them into a json.RawMessage, then call
|
|
the custom unmarshaler. Interface-typed fields are similar, except
|
|
instead of a custom unmarshaler we call the helper we've generated
|
|
(see unmarshal_helper.go.tmpl). Embedded fields don't need the
|
|
json.RawMessage; we just unmarshal our input again into the embedded
|
|
field. */}}
|
|
|
|
func (v *{{.GoName}}) UnmarshalJSON(b []byte) error {
|
|
{{/* Standard convention for unmarshalers is to no-op on null. */}}
|
|
if string(b) == "null" {
|
|
return nil
|
|
}
|
|
|
|
{{/* For our first pass, we ignore embedded fields, unmarshal all the
|
|
custom-unmarshaler or abstract fields into json.RawMessage, and
|
|
unmarshal everything else normally. To do this, we want to call
|
|
json.Unmarshal on the receiver (v). But if we do that naively on a
|
|
value of type <.GoName>, it will call this function again, and recurse
|
|
infinitely. So we make a wrapper type which embeds both this type and
|
|
NoUmnarshalJSON, which prevents either's UnmarshalJSON method from
|
|
being promoted. For more on why this is so difficult, see
|
|
https://github.com/benjaminjkraft/notes/blob/master/go-json-interfaces.md.
|
|
(Note there are a few different ways "hide" the method, but this one
|
|
seems to be the best option that works if this type has embedded types
|
|
with UnmarshalJSON methods.)
|
|
*/}}
|
|
|
|
{{/* TODO(benkraft): Omit/simplify the first pass if all fields are
|
|
embedded/abstract. */ -}}
|
|
var firstPass struct{
|
|
*{{.GoName}}
|
|
{{range .Fields -}}
|
|
{{if and .NeedsMarshaling (not .IsEmbedded) -}}
|
|
{{.GoName}} {{repeat .GoType.SliceDepth "[]"}}{{ref "encoding/json.RawMessage"}} `json:"{{.JSONName}}"`
|
|
{{end -}}
|
|
{{end -}}
|
|
{{/* TODO(benkraft): In principle you might have a field-name that
|
|
conflicts with this one; avoid that. */ -}}
|
|
{{ref "github.com/Khan/genqlient/graphql.NoUnmarshalJSON"}}
|
|
}
|
|
firstPass.{{.GoName}} = v
|
|
|
|
err := {{ref "encoding/json.Unmarshal"}}(b, &firstPass)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
{{/* Now, handle the fields needing special handling. */}}
|
|
{{range $field := .Fields -}}
|
|
{{if $field.NeedsMarshaling -}}
|
|
{{if $field.IsEmbedded -}}
|
|
{{/* Embedded fields are easier: we just unmarshal the same input into
|
|
them. (They're also easier because they can't be lists, since they
|
|
arise from GraphQL fragment spreads.)
|
|
|
|
Note that our behavior if you have two fields of the same name via
|
|
different embeds differs from ordinary json-unmarshaling: we unmarshal
|
|
into *all* of the fields. See goStructType.FlattenedFields in
|
|
types.go for more discussion of embedding and visibility. */ -}}
|
|
err = {{$field.Unmarshaler $.Generator}}(
|
|
b, &v.{{$field.GoType.Unwrap.Reference}})
|
|
if err != nil {
|
|
return err
|
|
}
|
|
{{else -}}
|
|
{{/* For other fields (abstract or custom unmarshaler), first, call the
|
|
unmarshaler (our unmarshal-helper, or the user-specified one,
|
|
respectively). This gets a little complicated because we may have
|
|
a slice field. So what we do is basically, for each field of type
|
|
`[][]...[]MyType`:
|
|
|
|
dst := &v.MyField // *[][]...[]MyType
|
|
src := firstPass.MyField // [][]...[]json.RawMessage
|
|
|
|
// repeat the following three lines n times; each time, inside
|
|
// the loop we have one less layer of slice on src and dst
|
|
*dst = make([][]...[]MyType, len(src))
|
|
for i, src := range src {
|
|
// We need the &(*dst)[i] because at each stage we want to
|
|
// keep dst as a pointer. (It only really has to be a
|
|
// pointer at the innermost level, but it's easiest to be
|
|
// consistent.)
|
|
dst := &(*dst)[i]
|
|
|
|
// (now we have `dst *MyType` and `src json.RawMessage`)
|
|
__unmarshalMyType(dst, src)
|
|
|
|
} // (also n times)
|
|
|
|
Note that if the field also uses a pointer (`[][]...[]*MyType`), we
|
|
now pass around `*[][]...[]*MyType`; again in principle
|
|
`[][]...[]*MyType` would work but require more special-casing. Thus
|
|
in the innermost loop, `dst` is of type `**MyType`, so we have to
|
|
pass `*dst` to the unmarshaler. Of course, since MyType is an
|
|
interface, I'm not sure why you'd any of that anyway.
|
|
|
|
One additional trick is we wrap everything above in a block ({ ... }),
|
|
so that the variables dst and src may take on different types for
|
|
each field we are handling, which would otherwise conflict. (We could
|
|
instead suffix the names, but that makes things much harder to read.)
|
|
*/}}
|
|
{
|
|
dst := &v.{{$field.GoName}}
|
|
src := firstPass.{{$field.GoName}}
|
|
{{range $i := intRange $field.GoType.SliceDepth -}}
|
|
*dst = make(
|
|
{{repeat (sub $field.GoType.SliceDepth $i) "[]"}}{{if $field.GoType.IsPointer}}*{{end}}{{$field.GoType.Unwrap.Reference}},
|
|
len(src))
|
|
for i, src := range src {
|
|
dst := &(*dst)[i]
|
|
{{end -}}
|
|
{{/* dst now has type *<GoType>; dst is json.RawMessage */ -}}
|
|
{{/* If the field is null in the input, skip calling unmarshaler.
|
|
(This matches json.Unmarshal.) If the field is missing entirely
|
|
from the input, we will have an uninitialized json.RawMessage;
|
|
handle that likewise. */ -}}
|
|
if len(src) != 0 && string(src) != "null" {
|
|
{{if $field.GoType.IsPointer -}}
|
|
{{/* In this case, the parent for loop did `make([]*MyType, ...)`
|
|
and we have a pointer into that list. But we actually still
|
|
need to initialize the *elements* of the list. */ -}}
|
|
*dst = new({{$field.GoType.Unwrap.Reference}})
|
|
{{end -}}
|
|
err = {{$field.Unmarshaler $.Generator}}(
|
|
src, {{if $field.GoType.IsPointer}}*{{end}}dst)
|
|
if err != nil {
|
|
return fmt.Errorf(
|
|
"Unable to unmarshal {{$.GoName}}.{{$field.GoName}}: %w", err)
|
|
}
|
|
}
|
|
{{range $i := intRange $field.GoType.SliceDepth -}}
|
|
}
|
|
{{end -}}
|
|
}
|
|
{{end}}{{/* end if/else .IsEmbedded */ -}}
|
|
{{end}}{{/* end if .NeedsMarshaling */ -}}
|
|
{{end}}{{/* end range .Fields */ -}}
|
|
|
|
return nil
|
|
}
|