add grpc protocol.

[morvencao]

resolve: #980

[morvencao]

/cc @yanmxa

[duglin]

Why do we delete the item from the map just to then re-add it here with nil ? Seems like this should be under an else or just remove the if+delete on lines 126-128. Same for below

[morvencao]

good catch, added else to handle non-nil case.

[duglin]

what if there's both 'contenttype' and 'ce-datacontenttype' ?

[morvencao]

'ce-datacontenttype' will be handled by:

sdk-go/protocol/grpc/message.go

Lines 122 to 128 in ab07228

	if strings.HasPrefix(name, prefix) {
	attr := m.version.Attribute(name)
	if attr != nil {
	err = encoder.SetAttribute(attr, attrVal)
	if err != nil {
	return err
	}

[duglin]

yes but which one wins? Should we require them to be the same (if both are there)?

[morvencao]

When both 'ce-datacontenttype' and 'contenttype' are set, 'ce-datacontenttype' takes precedence. In practical scenarios, only 'contenttype' is necessary. Check how the 'contenttype' attribute is configured here:

sdk-go/protocol/grpc/write_message.go

Lines 121 to 130 in 6523960

	case spec.DataContentType:
	if value == nil {
	delete(b.Attributes, contenttype)
	} else {
	attrVal, err := attributeFor(value)
	if err != nil {
	return err
	}
	b.Attributes[contenttype] = attrVal
	}

This follow the similar approach that the Kafka protocol and PubSub protocol take, refer to the implementations in the CloudEvents SDK for Kafka (https://github.com/cloudevents/sdk-go/blob/main/protocol/kafka_sarama/v2/write_producer_message.go#L103-L114) and PubSub (https://github.com/cloudevents/sdk-go/blob/main/protocol/pubsub/v2/write_pubsub_message.go#L70-L80).

[duglin]

Nit: you can remove all of these "return nil" lines since it'll happen due to the return on line 156

[morvencao]

thanks, done.

[duglin]

Let's define 'err' in the func instead of in the return arg - just for consistency

[morvencao]

done.

[duglin]

I see we have a sample, but can we get some testcases too?

[duglin]

unit tests are failing

[morvencao]

Added some test cases, PTAL @duglin @lionelvillard

[duglin]

@morvencao can you take a look at the failing checks?

[morvencao]

@duglin I've resolved the issues with testing and successfully executed it on my local environment, passing all tests.
I kindly request your approval to proceed with running the tests once more.

[yanmxa]

Why replace any with any1? Is this intentional？

[morvencao]

This code is generated by proto compiler, not manually altered.

[duglin]

I suspect it does this since any is now a data type in go

[yanmxa]

The Publish-subscribe model has the characteristics: The publisher sends events without specifically targeting any recipient. Subscribers express interest in certain types of messages and receive notifications or updates when messages of interest are published.

In my opinion, the gRPC does not explicitly refer to the publish-subscribe model. Instead, it resembles more of a request-response model, like the HTTP protocol. So I suggest aligning these interfaces with the http protocol.

[duglin]

please delete this file?

[morvencao]

I did delete the file. but I'm unsure if we should do that.
Based on the UT script at https://github.com/cloudevents/sdk-go/blob/main/hack/unit-test.sh#L20, it seems that coverage.tmp is produced as intermediate output by UT. Deleting it should pose no harm.

[morvencao]

I can restore it if there are additional uses that I overlooked.

[morvencao]

fixed the testing and rebased the code.

[duglin]

rebase is needed

[duglin]

ping @morvencao