[configuration] Fail on unsupported identifiers

[mx-psi]

Fixes #3981, fixes #3915

Changes

Fail to parse when invalid identifiers are found (this is basically anything that uses the ${...} syntax but contains an invalid blob in between). The rationale is to allow for changes in a backwards compatible way. Right now it would not be possible to extend the syntax in a non-breaking way to support more sh/POSIX/Bash syntax like the ones here or potentially to support other providers (like we support in the Collector e.g. ${file:<something>}).

As an example both #3974 and #3948 would have been nonbreaking had we done this change first (we would go from "fail to parse" to "do something", instead of going from one way of parsing to a different one).

For non-trivial changes, follow the change proposal process.

• Related issues Logging format for trace context #398
• CHANGELOG.md file updated for non-trivial changes

[mx-psi]

cc @open-telemetry/configuration-maintainers @TylerHelmuth

[jack-berg]

The rationale is to allow for changes in a backwards compatible way. Right now it would not be possible to extend the syntax in a non-breaking way to support more sh/POSIX/Bash syntax like the ones here or potentially to support other providers (like we support in the Collector e.g. ${file:}).

I think we we can accommodate this differently. We can adjust the regular expression definition to define a group for the optional env:, file: prefix, and state later state that the only supported group is env:. This would allow us to add additional substitution sources later in a non-breaking way.

EDIT: I think I misunderstood the goal. You're not saying to fail when an env var is referenced but unset. You're saying fail when an env var substitution expression doesn't quite match the regexp, and thus would be left as a string unless we fail.

[mx-psi]

I think I misunderstood the goal. You're not saying to fail when an env var is referenced but unset. You're saying fail when an env var substitution expression doesn't quite match the regexp, and thus would be left as a string unless we fail.

Indeed, that's what I meant :)

[codeboten]

Thanks @mx-psi

[jack-berg]

FYI, this also closes #3915, but in a different want than suggested in the convo on that issue.

[mx-psi]

FYI, this also closes #3915, but in a different want than suggested in the convo on that issue.

Updated PR description to also close this one

[jack-berg]

Planning on merging in two days if there are no additional comments. I know @pellared and @marcalff expressed a preference to not error in the #3915 discussion. Please review / comment.

[mx-psi]

docfx is failing because of open-telemetry/semantic-conventions/issues/987