I created #15311 that uses a slightly different approach just to test out whether it would work. Noticed that for webflux instrumentation the header getter logic is duplicated for client and server instrumentation, fixed that too. The approach taken in this PR is fine too and I might even prefer it. Though first all tests would need to pass.

I created #15311 that uses a slightly different approach just to test out whether it would work. Noticed that for webflux instrumentation the header getter logic is duplicated for client and server instrumentation, fixed that too. The approach taken in this PR is fine too and I might even prefer it. Though first all tests would need to pass.

great - I'm also trying to fix this PR so that we can choose.

• I've also opted not to ignore exceptions (which has already made it hard to fix the PR)
• added explicit check for spring 7 instead
• not removed duplication yet

@laurit should I rebase - or just close this PR?

@laurit should I rebase - or just close this PR?

your choice. I like that it always uses GET_HEADERS.invoke(headers, name), unlike the current code that has a branch. I don't like that it uses Class.forName("org.springframework.core.Nullness"); for version detection as this is not a reliable way to detect versions. It fails when there are multiple versions of spring on classpath, unfortunately this sometimes happens.

It fails when there are multiple versions of spring on classpath, unfortunately this sometimes happens.

removed that again

@laurit thanks for the feedback

I also added nullaway to make the new code more maintainable