Populating query param semantics changed since v2.15.0

[roehrijn]

Populating query param semantics changed since v2.15.0

[johanbrandhorst]

Thanks for getting back to me. I think the safest thing for us to do is simply revert the PR that introduced the regression for now, then consider a new implementation that isn't subject to this problem. I'll take care of that.

[johanbrandhorst]

Hi Jan, thank you so much for this detailed issue. First of all let me express my apologies for this. Secondly, I'm extremely impressed with the lengths you've gone to, both to create a test to reproduce the issue and to drill down and discover the cause. Bravo! Would you be interested in submitting a fix for this? I'm ready to revert the PR that introduced the issue of course but if you intend to submit a fix we can avoid it. Thanks!

[roehrijn]

Hi Johan. You're welcome. I want this to be fixed and thus, I'm doing my best to support wherever I can. However there are two aspects which hinder me in providing a fix:

1. I have no glue about the very complex internal data model so far. I couldn't figure out how to create unit test in protoc-gen-grpc-gateway/internal/gengateway/template_test.go for this issue because there are some examples with nested items in repo, but they all work with HTTP POSTs and I wasn't able to turn them into working examples for GET.
2. I have not yet a real idea how to fix the issue. Somehow all parents of nested targets and all their JSON names need to be taken into consideration I guess. But again, currently I have no overview of the data model and what would be possible with the amount of metadata available at this point.

However, I'm eager to help and if somebody could assist by providing a test in the context of 1), it could give me a big knowledge-boost in that area. Maybe enough to fix that issue.