Use strings concat where possible in the request signing

[dlsniper]

Hi,

This is a small improvement, but one nonetheless.
On my machine, this: https://play.golang.org/p/a_ow6e7II7 using go 1.7 beta2 produces the following results:

go test -bench=.
BenchmarkStringsJoin-8          10000000               140 ns/op
BenchmarkStringsConcat-8        20000000                92.6 ns/op
PASS
ok      some/package        3.505s


go test -bench=. -benchmem
BenchmarkStringsJoin-8          10000000               136 ns/op             128 B/op          2 allocs/op
BenchmarkStringsConcat-8        20000000                96.6 ns/op            64 B/op          1 allocs/op
PASS
ok      some/package        3.565s

Also I wonder if it isn't a mistake that we have two new lines before v4.canonicalHeaders and not after?

Thank you.

[jasdel]

Thanks for taking the time to submit this PR @dlsniper. I don't think this will be a great change to take. The performance difference in negligible, and arguable hurts readability.

Right now these sections are a little gnarly to read. I'm not sure if a fmt.Sprintf would make them any easier to read.

[dlsniper]

Hi, sorry for the delayed reply. For many of the usecases yes, this might make no difference. I do think that for high perf environments, even a few ns shaved of while generating the key would be nice but up to you. I can rebase if you want or close this. Please let me know which direction you'd prefer. Thank you.

[jasdel]

@dlsniper, thanks for the update. I think the SDK would be better served using pprof/profiling to identify big fish within the SDK that can make significant improvements. Once we've knocked out those items I think that then is a good time to consider micro optimizations vs the cost of readability and maintainability.

One item I'm working on right now is to collect metrics for performance analysis of the SDK so we can identify areas to target for further improvements. Please continue to investigate and bring up areas you find where the SDK's performance could be improved.