Closed michaelpj closed 3 days ago
@marinelli has a PR up reverting to checked in generated code. I'll leave it a while in case anyone has any objections, otherwise I'll merge that.
There is an easy way to avoid this: just do the codegen manually and check in the result. It's ugly, and involves checking in generated code, but it's robust and not too painful if the input doesn't change that often (which I don't think it does).
Did a quick glance at https://github.com/open-telemetry/opentelemetry-proto/commits/main/ and it gets updated every few months or so, which doesn't seem often enough to warrant a custom setup.
We're back to manual generation for now.
I think we should seriously consider avoiding any build-time code generation. At the moment that's mostly
hs-opentelemetry-otlp
, but there is also https://github.com/iand675/hs-opentelemetry/pull/142 which I think is desirable.Why?
protoc
around in order to buildhs-opentelemetry-otlp
. Cabal does not know this, it is quite opaque to users.There is an easy way to avoid this: just do the codegen manually and check in the result. It's ugly, and involves checking in generated code, but it's robust and not too painful if the input doesn't change that often (which I don't think it does).
I also note that the last release of
hs-opentelemetry-otlp
did not use a custom setup, so this is going to be a new breaking requirement for our users. We already have people complaining about that.Overall, I would prefer to revert https://github.com/iand675/hs-opentelemetry/pull/122 before we release, and ideally commit to not doing this in the future.
cc @iand675 and @evanlauer1 , who are likely to have opinions.