Implement grpc-gateway's patch feature for GrpcJsonTranscoder

[sgtsquiggs]

grpc-gateway implements Google's "Standard API" Update method, which we use w/ grpc-gateway. We'd love to switch to Envoy to reduce the complexity of our services but this is holding us back. Is this functionality already implemented, or is this a feature request?

[mattklein123]

cc @lizan

[qiwzhang]

It is not implemented. This is a feature request.

[dio]

@qiwzhang just curious, should that be implemented in the filter or library side (https://github.com/grpc-ecosystem/grpc-httpjson-transcoding)?

[qiwzhang]

Most likely in the library side since it needs body field names to generate "update_mask", the filter doesn't have that data. The library will expose a boolean flag to enable such feature, the filter needs to turn it on for a PATCH method.

[stale]

This issue has been automatically marked as stale because it has not had activity in the last 30 days. It will be closed in the next 7 days unless it is tagged "help wanted" or other activity occurs. Thank you for your contributions.

[tzachshabtay]

We still want this, mr stale bot.

[RyanMarr]

Did this ever get solved?!

[TusconDanksman]

Most likely in the library side since it needs body field names to generate "update_mask", the filter doesn't have that data. The library will expose a boolean flag to enable such feature, the filter needs to turn it on for a PATCH method.

Hey @qiwzhang, Picking up this thread from nearly 3 years ago 📆

Is this feature request uncontroversial enough that it would be welcomed if implemented in the manor you suggested?

That is: add the functionality to the library (https://github.com/grpc-ecosystem/grpc-httpjson-transcoding) and expose it through a boolean flag, and then ultimately add support in Envoy to use that update?

--

Our team is using Envoy and its grpc-transcoder, but because of legacy reasons, we need a way to tell between an intentionally set to empty repeated field vs. an unset repeated field that defaults to as empty. Optional doesn't work with repeated fields, so updates with FieldMask seem to fit that bill.