It will likely be faster to split using a regex than calling split 2 times.

The performance will be the same in this case for 'well formed' headers, but worse for non-well formed headers.

I think this could actually be made faster though by scanning for the colon in c and then calling GoStringN() on the respective parts of the *char.

I rewrote the header splitting to ignore all whitespaces after ':', it will now also directly allocate key and value from the *char.

This is a bit faster than strings.SplitN(C.GoStringN(cstringForSplit, length), ": ", 2), but only a small optimization in the context of a whole request.

Test with go test -bench=. (only splitting this specific header without adding it: "Header-Key: value")

BenchmarkHeaderSplitOld-20              21330975                50.82 ns/op
BenchmarkHeaderSplitNew-20              39266910                29.69 ns/op

Good one! Thanks.