[confmap] Remove top level condition when considering struct as Unmarshalers.

[atoulme]

This is a slice of #9750 focusing on removing the top level condition on unmarshaling structs.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.86%. Comparing base (ff7a485) to head (3958649).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #9862      +/-   ##
==========================================
- Coverage   91.88%   91.86%   -0.03%     
==========================================
  Files         360      360              
  Lines       16728    16732       +4     
==========================================
  Hits        15371    15371              
- Misses       1020     1022       +2     
- Partials      337      339       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[atoulme]

Blocked on open-telemetry/opentelemetry-collector-contrib#31802

[mx-psi]

I think I finally understood this! I filed #9879 separately that adds two confmap tests. I would want to merge that one and #9861 first if that makes sense to you.

[mx-psi]

It would also be interesting to try and reproduce the failing contrib test on a unit test here, so we can work on it. The offending Unmarshal function is here https://github.com/open-telemetry/opentelemetry-collector-contrib/blob/e4c5b51ec0e24ca163bcb64d944455577f90a09f/pkg/stanza/operator/helper/time.go#L49

[atoulme]

It would also be interesting to try and reproduce the failing contrib test on a unit test here, so we can work on it. The offending Unmarshal function is here https://github.com/open-telemetry/opentelemetry-collector-contrib/blob/e4c5b51ec0e24ca163bcb64d944455577f90a09f/pkg/stanza/operator/helper/time.go#L49

added a skipped test here: TestRecursiveUnmarshaling

[atoulme]

If you want a foolproof solution, we might have to do something a little more drastic. We might need to keep a hashset of all pointers of unmarshaler methods ever called, to make sure we don't run into cycles.

[mx-psi]

If you want a foolproof solution, we might have to do something a little more drastic. We might need to keep a hashset of all pointers of unmarshaler methods ever called, to make sure we don't run into cycles.

I guess we can forbid what the stanza package does instead. Do you think there is a benefit to being able to do what stanza does? I can't see any advantage right now

[atoulme]

No, I don't see any advantage, except maybe if you're trying to unmarshal and set all values on the struct atomically. Even then, this is a corner case. I think we're ok.

[mx-psi]

I am okay with the current approach, I guess we should document the limitation about nested unmarshalling before merging, but I think we can go with this.

cc @open-telemetry/collector-approvers, I would appreciate a second pair of eyes since this is somewhat subtle. Even if this is ready before, I will wait at least until Monday to merge to give others some time to review.

[atoulme]

One more PR in contrib needed: open-telemetry/opentelemetry-collector-contrib#32577

[atoulme]

The last contrib issue is solved by open-telemetry/opentelemetry-collector-contrib#32660. Unfortunately, I couldn't find a good way to fix this - in effect, a breaking change.

[evan-bradley]

Looks good to me, I think this is probably the cleanest solution given the current state of things.

[mx-psi]

LGTM after adressing Evan's comment. This will need an update on contrib since it's breaking the tests there

[codeboten]

@atoulme are these expected:

    config_warnings_test.go:92: 
        	Error Trace:	/tmp/opentelemetry-collector-contrib/exporter/datadogexporter/config_warnings_test.go:92
        	Error:      	elements differ
        	            	
        	            	extra elements in list B:
        	            	([]interface {}) (len=1) {
        	            	 (string) (len=125) "\"metrics::histograms::send_count_sum_metrics\" has been deprecated in favor of \"metrics::histograms::send_aggregation_metrics\""
        	            	}
        	            	
        	            	
        	            	listA:
        	            	([]string) (len=1) {
        	            	 (string) (len=125) "\"metrics::histograms::send_count_sum_metrics\" has been deprecated in favor of \"metrics::histograms::send_aggregation_metrics\""
        	            	}
        	            	
        	            	
        	            	listB:
        	            	([]string) (len=2) {
        	            	 (string) (len=125) "\"metrics::histograms::send_count_sum_metrics\" has been deprecated in favor of \"metrics::histograms::send_aggregation_metrics\"",
        	            	 (string) (len=125) "\"metrics::histograms::send_count_sum_metrics\" has been deprecated in favor of \"metrics::histograms::send_aggregation_metrics\""
        	            	}
        	Test:       	TestSendAggregations/metrics::histograms::send_count_sum_metrics_set_to_false

[mx-psi]

@atoulme So that we can merge this soon and address @codeboten's comment, would you be able to do the following?

1. File a draft PR on opentelemetry-collector-contrib that pulls in this version of opentelemetry-collector + adds the necessary fixes in contrib. This is to 'prove' that we can satisfactorily fix the issue in contrib
2. Post the PR from (1) here. Once this is done, we can merge this PR.
3. Update contrib and mark the PR from (1) as ready to review with only the fixes

[atoulme]

Here is the PR open-telemetry/opentelemetry-collector-contrib#32660

It is past midnight, I will follow up tomorrow on it.

[atoulme]

The contrib PR shows the changes combined with test changes to contrib will work. I think we have reached step 2. I have reached out to Alex to make sure he's ok with this.

[atoulme]

@codeboten are you ok with this approach and the steps I have followed, as well as the checks they provide? Can we merge this PR?

[atoulme]

Can we agree to merge this by end of next week unless someone objects?

[mx-psi]

I'll put a reminder to merge in a wekk, @codeboten feel free to merge before then if you are okay with it :)