Description
First of all, thank you a lot for this awesome library. I just ran a few benchmarks on my marshalling code, and I'm getting a speedup of 6x (!) and a reduction of 25x (!) on allocated memory compared to the Go standard library. That's a crazy difference!
Trying to optimize things even further (just trying to push things to the limit ;)), I've encountered the following problem: In the JSON format that I'm supposed to output, integers need to be encoded as strings (because many JSON implementations only support numbers up to 2^53 or so).
With the Go standard library this is pretty straightforward. I'd define a value like this
type frame struct {
Offset int64 `json:"offset,string"`
}
This would automatically output the int64
(in decimal representation) as a string.
As far as I've found out so far, the way to do this using gojay would be the following:
enc.StringKey("offset", strconv.FormatInt(int64(f.Offset)))
The problem is that strconv.FormatInt
has to allocate the string it is returning, just to have gojay copy the string contents to the JSON output being written, and garbage collected afterwards.
I can see two ways how to optimize this:
- I could allocate a byte slice (or string) as a buffer, encode my
int64
into that string, and then useStringKey
with this buffer, reusing it every time I need to write a number as a string. This solution would not require any changes to gojay. - gojay could offer an API to do exactly this, e.g. a function
Encoder.Int64AsString
. Considering that my use case is apparently pretty common and even supported by the standard library, maybe this is the cleaner solution?
Does this make sense? Or am I missing something here?