Use Ryu for Float/Double to String conversions

[re-thc]

Hi, see https://github.com/ulfjack/ryu and the paper / benchmarks / information. Might help a lot in JSON output from Float / Double.

Cheers!

[cowtowncoder]

@hc-codersatlas Thank you for the link! Looks potentially interesting. I don't have much time to look into this right now, but maybe someone else has, so it's good to have an issue filed.

[re-thc]

Sounds good. Hopefully it gets picked up. Jsoniter adopted this and is 1 of the reasons it's faster.

[vy]

For the records, jsoniter fame @plokhotnyuk kindly shared the following links in a Reddit comment:

1. Fast & medium paths for parsing of doubles
2. Fast & medium paths for parsing of floats
3. Fast parsing of BigDecimals and using them to parse BigIntegers
4. Fast parsing of ISO-8601 representation of java.time.* instances
5. Fast parsing of UUIDs
6. Fast serialization of UUIDs
7. Fast serialization of ISO-8601 representation of java.time.* instances
8. Fast serialization of floats
9. Fast serialization of doubles

@cowtowncoder Let me know if you would prefer me creating individual tickets for each of these 9 improvements.

[cowtowncoder]

Qq: since #517 is related, wonder if these should be combined?

But as to @vy's question: I think that splitting definitely makes sense, but might make most sense to do "on-deman", when someone is considering working on specific aspect -- need not be a firm commitment, but just evaluation, so collaboration is easier wrt specific type or types.

[plokhotnyuk]

I switched jsoniter-scala from Ryu to Schubfach algorithm ~1 year ago. It brought guarantied shortest output with proper rounding. Also it gave a better throughput and minimized the tail latency.

Here is an original repo of the algorithm author that compiles only for Java 9+.

Here is its adoption in jsoniter-scala that works efficiently with Java 8 too.

[pjfanning]

I have https://github.com/pjfanning/double-reader-writer and #749 -- there is an open discussion on #746 about upgrading to Java 8 as a base in jackson 2.14

This issue seems closely related to #577

[cowtowncoder]

Will close, replaced by above-mentioned PRs for fast floating-point reads and writes, nominally #577 for reads and #749 for writes.

[pjfanning]

We support Schubfach - which is also going to appear in new JDKs in future (#577 (comment))

[cowtowncoder]

@pjfanning Feel free to improve titles and descriptions of relevant issues/PRs with references to algorithm(s) used. I am trying to figure out best ways to quickly indicate how improvements are achieved. I changed #577 title a bit.