[CHANGED] avoid unnecessary data copy in processMsg

[ninjken]

This PR intends to address the github issue

may avoid unnecessary data copy in processMsg for large message payloads #787

I've also tweaked the benchmark test in file test/bench_test.go for local testing, but not included it in the PR because I am not sure of its necessity.

func BenchmarkPubSubSpeed(b *testing.B)
...
    //msg := []byte("Hello World")
    msg := []byte(strings.Repeat("Hello", 1000))

[kozlovic]

LGTM but small comment if we should reset this boolean after the other call to processMsg() (although I believe it is not needed, but same for nc.ps.msg.Buf which is clearly already nil, yet we reset).

[kozlovic]

For consistency, I wonder if we should also reset this boolean here: https://github.com/nats-io/nats.go/pull/788/files#diff-83eb8e32639d01cf443d6d8bde24c1c8be78766090d8c5f8586c36250cfedca6R194
I agree that at this location it means that the message fit in the current read buffer, so it is clearly not copied, so this boolean should not be true..

[ninjken]

At first I thought it was not needed, but I think I agree with you we should reset the flag as well for code consistency.
PR updated :-)

[coveralls]

Coverage decreased (-0.03%) to 86.933% when pulling 9b7c017 on Endeavourken:master into 91bdffe on nats-io:master.

[kozlovic]

LGTM. Thank you for the contribution!