Oh heh, a trip down the memory lane. I wrote the initial version of that, in an era where AWS docs did not match observed S3 behavior. The only way to make an S3-compatible API was to create a suite of over-the-network tests to run against both AWS S3 and radosgw.

We also had a little grammar-based fuzzer for S3 requests (really, any HTTP), but over the last 10+ years I've lost track of what happened to that code. That found some incompatibilities with allowed character sets etc too.