vcabbage / amqp

73%

master: 73%

push

travis-ci

Additional encoding/decoding optimization

* Replace reader/writer interfaces with concrete `buffer` type.
* Avoid intermediate buffer during map encoding.
* Avoid `milliseconds` marshal allocation.
* Remove `errNull`/`isNull` from marshal and avoid defer in hot path.
* Cleanup of decoding code.
* Minimize bounds checks by adding hints.
* Check interface types last in type switches.
* Add fastpath in unmarshal for `**uint32`.

```
name                                       old time/op    new time/op    delta
FrameMarshal/transfer-8                       420ns ± 4%     197ns ± 5%   -53.04%  (p=0.000 n=10+10)
FrameUnmarshal/transfer-8                    1.84µs ± 0%    0.61µs ± 5%   -66.64%  (p=0.000 n=8+10)
Marshal/*amqp.performOpen-8                   787ns ± 3%     293ns ± 5%   -62.83%  (p=0.000 n=10+10)
Marshal/*amqp.performBegin-8                  621ns ± 1%     216ns ± 3%   -65.27%  (p=0.000 n=10+9)
Marshal/*amqp.performAttach-8                2.45µs ± 1%    1.12µs ± 3%   -54.13%  (p=0.000 n=10+10)
Marshal/amqp.role-8                          30.5ns ± 2%    20.5ns ± 2%   -32.75%  (p=0.000 n=10+10)
Marshal/*amqp.unsettled-8                     258ns ± 8%     100ns ± 3%   -61.10%  (p=0.000 n=10+9)
Marshal/*amqp.source-8                        900ns ± 1%     401ns ± 5%   -55.43%  (p=0.000 n=10+10)
Marshal/*amqp.target-8                        516ns ± 1%     244ns ± 3%   -52.79%  (p=0.000 n=10+10)
Marshal/*amqp.performFlow-8                   576ns ± 1%     193ns ± 4%   -66.42%  (p=0.000 n=10+10)
Marshal/*amqp.performTransfer-8               404ns ± 6%     184ns ± 2%   -54.35%  (p=0.000 n=10+10)
Marshal/*amqp.performDisposition-8            215ns ± 5%     102ns ± 4%   -52.54%  (p=0.000 n=10+10)
Marshal/*amqp.performDetach-8                 612ns ± 2%     279ns ± 0%   -54.37%  (p=0.000 n=10+6)
Marshal/*amqp.performDetach#01-8             1.23µs ± 8%    0.42µs ± 4%   -65.71%  (p=0.000 n=9+10)
Marshal/amqp.ErrorCondition-8                69.9ns ± 1%    27... (continued)

1131 of 1131 new or added lines in 8 files covered. (100.0%)

3596 of 4909 relevant lines covered (73.25%)

2.4 hits per line