Address AOT warnings in Exporter.OpenTelemetryProtocol

[eerhardt]

This addresses the AOT warnings in the last library in opentelemetry-dotnet.

1. Google.Protobuf v3.19.4 isn't trimming and AOT compatible. We need to upgrade to a newer version. I chose the latest patch in v3.22.x. v3.22.0 was the first version that is AOT compatible.

2. There were 2 places in Exporter.OpenTelemetryProtocol that was using System.Reflection.Emit to set a private field on Protobuf's RepeatedField class in order to return it to a pool. Setting this private field is no longer necessary since v3.22.0 because the Clear method was updated to support this scenario. See protocolbuffers/protobuf@a4fd216

Fix #3429

@vitek-karas @Yun-Ting @reyang @utpilla

[eerhardt]

@CodeBlanch - I see you commented on the Google.Protobuf issue and PR that fixed the Clear() method to allow for pooling. Was there a reason OTel hasn't already updated to use this method instead of using Ref.Emit?

[codecov]

Codecov Report

Merging #4859 (e194f5e) into main (7cb92d3) will decrease coverage by 0.10%.
The diff coverage is 50.00%.

@@            Coverage Diff             @@
##             main    #4859      +/-   ##
==========================================
- Coverage   83.25%   83.16%   -0.10%     
==========================================
  Files         293      293              
  Lines       12012    11984      -28     
==========================================
- Hits        10001     9966      -35     
- Misses       2011     2018       +7     

Flag	Coverage Δ
unittests	83.16% <50.00%> (-0.10%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files Changed	Coverage
...tryProtocol/Implementation/MetricItemExtensions.cs	0.00%
...metryProtocol/Implementation/ActivityExtensions.cs	100.00%

[reyang]

LGTM with a changelog entry.

[Yun-Ting]

@CodeBlanch - I see you commented on the Google.Protobuf issue and PR that fixed the Clear() method to allow for pooling. Was there a reason OTel hasn't already updated to use this method instead of using Ref.Emit?

This commit: CodeBlanch@18c2d40
was reverted because of this proposal: #4396 (comment)

I'll bring this up for discussion in tomorrow's SIG meeting.

[utpilla]

@CodeBlanch - I see you commented on the Google.Protobuf issue and PR that fixed the Clear() method to allow for pooling. Was there a reason OTel hasn't already updated to use this method instead of using Ref.Emit?

@eerhardt @Yun-Ting We had updated this package version earlier but had to revert the changes in #4407. The reason behind it was mainly to not affect any auto-instrumentation users who could be using a lower version of the package. Auto-instrumentation dynamically loads this assembly and its dependencies. Check this comment.

Back then we didn't have any immediate need to be AOT-safe. However, AOT-safety is the major feature to be offered in 1.7.0 so updating this package version is much needed.

[Kielek]

@eerhardt, please check this PR #4201

You should be able to drop reference to System.Reflection.Emit.Lightweight in src/OpenTelemetry.Exporter.OpenTelemetryProtocol/OpenTelemetry.Exporter.OpenTelemetryProtocol.csproj.

You should be able to remove also https://github.com/open-telemetry/opentelemetry-dotnet/blob/7cb92d3b6dbad169048e5aa07939ef613f2587e2/Directory.Packages.props#L44C1-L45 to fully cleanup this reference.

The reason to keep older version is AutoInstrumentation (it will narrow the set of application which can be handled without code changes).

[Yun-Ting]

@CodeBlanch - I see you commented on the Google.Protobuf issue and PR that fixed the Clear() method to allow for pooling. Was there a reason OTel hasn't already updated to use this method instead of using Ref.Emit?

@eerhardt @Yun-Ting We had updated this package version earlier but had to revert the changes in #4407. The reason behind it was mainly to not affect any auto-instrumentation users who could be using a lower version of the package. Auto-instrumentation dynamically loads this assembly and its dependencies. Check this comment.

Back then we didn't have any immediate need to be AOT-safe. However, AOT-safety is the major feature to be offered in 1.7.0 so updating this package version is much needed.

Thank you @utpilla for sharing this!

[Yun-Ting]

LGTM assuming this comment got addressed: #4859 (comment). Thanks!