amazonka: fixing sigv4 path encoding so that it happens once for s3 and twice for other services

[mwu]

Should fix #529.

I did some manual smoke testing:

• EC2 describe instances calls
• lambda invokeFunction using all name forms (arn, partial arn, partial arn + alias, function name, name + alias)
• S3 getObject on keys including and excluding uri reserved chars

[endgame]

This is excellent work, thank you - I particularly appreciate the extensive test suite. I have a few small nits to pick but I look forward to merging this soon.

[endgame]

All right, so once you get your secondary PR merged I think this is good to go. Would you mind firing off your manual smoke-test requests one last time, too, just to make sure the tweaks are all good?

[mwu]

All right, so once you get your secondary PR merged I think this is good to go. Would you mind firing off your manual smoke-test requests one last time, too, just to make sure the tweaks are all good?

Ok. These all returned success:

• Lambda InvokeFunction with full arn, partial arn, full and partial arn with alias suffix, full and partial arn with alias in qualifier, function name, name with alias suffix, name with alias qualifier.
• S3 GetObject with keys test/myron/test, test//myron//test', abc:def$1.foo`
• Some other requests with non-empty paths: APIGateway GetStages, Route53 GetHostedZoneLimit
• A few of the examples in the examples project

[endgame]

Excellent, thank you for being so thorough. One more nit, and one last question: I think I have misunderstood the purpose of the various metaCanonicalFoo in the metadata record - it seems like they don't get used anywhere because they get built and immediately passed to signRequest. I think you pointed this out earlier? When I asked you to remove it, I didn't realise there was an established practice of keeping this data around. My question is - do you think we should reinstate metaCanonicalPath? I think your tests are fine as they are.

[mwu]

Excellent, thank you for being so thorough. One more nit, and one last question: I think I have misunderstood the purpose of the various metaCanonicalFoo in the metadata record - it seems like they don't get used anywhere because they get built and immediately passed to signRequest. I think you pointed this out earlier? When I asked you to remove it, I didn't realise there was an established practice of keeping this data around. My question is - do you think we should reinstate metaCanonicalPath? I think your tests are fine as they are.

I think it's fine as it is, too. The canonical path gets logged already as part of the canonical request and I can't see any other use for returning it in the metadata. Carrying it around is only useful for testing but as you said, we should really approach that by expanding the test suite to test the whole signature.

[endgame]

All right then, sounds good. Thanks very much for this.