endgame

Seems to gen OK in #761

This has been fixed by #812 .

Checked latest `develop`; this still looks like it's a thing. But since nobody's screamed for six years, it can wait until after 2.0.

OK, did some digging. I assume this is for the `PreSignedUrl` fields that cross-region RDS requests need (#378)? `amazonka-rds` uses the default V4 signer: https://github.com/brendanhay/amazonka/blob/53bc5a196acebc6bb7c40da71223c8817fcb6f90/amazonka-rds/gen/Network/AWS/RDS/Types.hs#L1314 However, the default signer appears...

`DiffTime` sounds good, as it's a type about time, but will need to see what `http-client` is prepared to accept. But will think about that later.

Is this a regression, or have you only ever used amazonka-2.0 from `main`?

I believe you are right. Can you try using your own `_MatchServiceError` call to construct a prism as workaround: https://github.com/brendanhay/amazonka/blob/26466244b5a23b29ba8894ec89d922e6a6835a05/lib/amazonka-core/src/Amazonka/Error.hs#L40-L45 ```haskell _ResourceNotFoundException = _MatchServiceError defaultService "ResourceNotFoundException:http://iot.amazonaws.com/doc/2015-10-02/" . hasStatus 404` ```...

https://github.com/boto/botocore/blob/fec0e5bd5e4a9d7dcadb36198423e61437294fe6/botocore/parsers.py#L1006-L1015 boto3 splits colons away when parsing an error code out of a response header - we should do the same.

There are a number of services we can't generate for yet because they make code that doesn't compile (usually circular module imports). I'll look into fixing those by hand in...

While #761 gives us some progress on recursive data types, the service definition seems to encode possible exceptions as if they were struct members, and this causes `gen` to emit...