Return response body in FeignException from errorReading (#2618)

[seonwooj0810]

Fixes #2618

Problem

FeignException.errorReading(...) builds the exception by passing request.body() and request.headers() into the constructor parameters named responseBody / responseHeaders:

return new FeignException(
    response.status(),
    format(...),
    request,
    cause,
    request.body(),     // <- responseBody slot
    request.headers()); // <- responseHeaders slot

As reported in #2618, this means that when an IOException is thrown while reading/decoding the response (see InvocationContext.decode and ResponseHandler.handleResponse), FeignException.responseBody() / content() / contentUTF8() / responseHeaders() return the request body and headers instead of the response. The maintainer confirmed in the thread that this looks like a bug.

Fix

Read the response body into a byte[] and pass the response headers, mirroring the existing errorStatus(...) method. Reading is wrapped in a try/catch (IOException ignored) so a streamed or already-consumed body degrades gracefully to an empty body rather than throwing (consistent with errorStatus).

Test evidence

Strengthened the existing FeignExceptionTest.canCreateWithRequestAndResponse to assert the content: the exception now carries the response body ("response") rather than the request body ("data"), and exposes the response headers. Before the fix this assertion fails (contentUTF8() returned "data").

./mvnw -pl core test -Dtest=FeignExceptionTest
Tests run: 14, Failures: 0, Errors: 0, Skipped: 0
BUILD SUCCESS

Verification done: (1) no in-flight PR (gh pr list --search "responseBody"/"errorReading" empty); (2) the only self-claim is a 2024-11-03 question that never produced a PR (>21 days stale); (3) code-focused change in FeignException.java + test; (4) confirmed the request.body() pattern still present on current master; (5) maintainer acknowledged the bug in-thread. Commit is DCO signed-off; google-java-format validation passes.

[velo]

The response-body/header fix in core/src/main/java/feign/FeignException.java has focused coverage in FeignExceptionTest, and I do not see a compatibility or security concern in the patch itself. I still cannot approve this head while the required ci/circleci: pr-build check is failing. Please get that check green, then I can re-review.

[seonwooj0810]

Thanks @velo. The failing ci/circleci: pr-build was a pre-existing feign-core test-compilation error on master (JavaLoggerTest/LoggerRebufferTest calling the Request.create(..., byte[], Charset) overload that had been removed), not anything introduced by this PR — this change only touches FeignException. That error has since been fixed on master by #3406 (commit 18fe771), so I've rebased this branch onto the latest master to pick it up and retrigger CI (head now a8e3ac0d).

Verified locally on the rebased branch: feign-core test sources compile and FeignExceptionTest passes (14/14). Should be green now — ready for another look whenever you have a moment. Thanks!

[velo]

The current FeignException.errorReading change has a focused test for preserving the response body and headers, and I do not see a compatibility or security concern in the diff itself. I still cannot approve this head because the required ci/circleci: pr-build check is failing. Please get that check green, then I can take another pass.

[seonwooj0810]

Thanks @velo. I tracked down the red ci/circleci: pr-build: it was three pre-existing tests that asserted the old behavior this PR fixes. throwsFeignExceptionIncludingBody / throwsFeignExceptionWithoutBody in FeignTest, AsyncFeignTest and FeignUnderAsyncTest expected FeignException.contentUTF8() from the errorReading path to return the request body — which is exactly the #2618 bug. Now that errorReading carries the response body, I updated those assertions to expect the response body ("success!"), and changed the without-body case to enqueue an empty response so it still verifies the empty-body path.

Full feign-core suite is green locally now: ./mvnw -pl core test → Tests run: 627, Failures: 0, Errors: 0. Pushed as feb8c31 (DCO signed-off). Could you re-trigger the build when you get a chance?

[velo]

The added updates in core/src/test/java/feign/FeignTest.java, AsyncFeignTest.java, and FeignUnderAsyncTest.java are in the right area for the errorReading behavior, but the required ci/circleci: pr-build check is still failing on this head. I can't approve while that required check is red. Please get the branch back to green, then I'll re-review the current diff.