rapidjson-ruby

rapidjson-ruby copied to clipboard

The capacity of handling arbitrary size integers is quite important to be able to replace the stdlib library.

Unfortunately RapidJSON isn't very flexible here.

For generating JSON it's quite easy to fallback to Integer#to_s.

However for parsing, the only escape hatch is to parse all integers ourselves, which significantly impact performance.

It could be a bit better by parsing in place, but all Ruby functions expect null terminated C-strings, so it wouldn't work well.

Ruby number parsing is also not as fast at RapidJSON's.

So for this to be performant, RapidJSON would need a flag that allows to only parse over-sized integers ourself, and not all numbers.

But that would require a new feature upstream.

(opening as a draft for comments, but I don't think it's mergeable in this state)

So https://github.com/Tencent/rapidjson/pull/2155 would allow to really keep the overhead down in the vast majority of cases where the number fits int64_t / uint64_t or is a Float.

No idea if it has any chance to be accepted upstream though.