[floxis-admin]

Type of change

• Bugfix
• Bid adapter update

Context

Floxis splits requests by seat and routing host. A response to one outbound request could return a bid for another group's impression because bid parsing used the original request's full impression list. A supplied mtype also bypassed impression lookup. The later response validator checks the overall auction, so it does not enforce this outbound-request boundary.

Change

Validate each response bid's impid against httpCall.getRequest().getPayload().getImp() before resolving its media type. Invalid bids use the existing bad_server_response error and are skipped; valid bids in the same response remain accepted.

Only the Floxis adapter and its unit tests change. Request grouping and configuration are unchanged.

Test plan

At aa72a579a7f00914e72524682ce87f1b53ac3c97:

• Compiled the changed adapter and test sources targeting Java 25, using cached unchanged project classes/dependencies from the same upstream master ancestor.
• ./mvnw -B checkstyle:check surefire:test -Dtest=FloxisBidderTest -DargLine=-Xmx256m: 30 tests passed, zero skipped, zero Checkstyle violations. Maven heap was limited to 512MiB.
• New tests exercise an actual split request, keep the valid own-group bid, and reject another seat's impression for mtype 1 and 0. Unknown-impression cases cover both values; existing media-type cases still pass.
• The unchanged adapter fails three of these regressions. Independently restoring either faulty behavior makes its corresponding regression tests fail.
• JaCoCo: 100% Floxis line coverage (97/97), 36/37 branches covered. Both changed methods have 100% line coverage, and media-type resolution has all 17 branches covered.
• git diff --check passed.

The full project build/test suite was not rerun. Upstream CI must run on this PR.