Setting `StreamReadFeature.USE_FAST_DOUBLE_PARSER` doesn't increase parsing throughput

[plokhotnyuk]

The feature for faster serialization of doubles that was introduced in 2.14.0-rc1 works fine when used in a factory builder for JsonGeneratorOps:

While a similar one for faster parsing of doubles doesn't increase parsing throughput:

[htmldoug]

Thanks for the heads up.

WeePickle can't use jackson-core to parse JSON doubles because of FasterXML/jackson-core#730, so the benchmark hits string.toDouble here (as of #102). Once jackson-core 2.14 lands, we should be able to change that over to use FastFloatParser directly, assuming it's part of the public API.

[pjfanning]

We are not treating FastFloatParser as part of public jackson-core API. It is a copy of a class from https://github.com/wrandelshofer/FastDoubleParser. That project is now publishing regular releases - you are better off using its classes directly. The classes have been renamed a little. Jackson v2.15 will use a shaded version of FastDoubleParser classes grabbed during the build.

You will probably want the JavaDoubleParser or the JsonDoubleParser classes.

Edit: jackson-core NumberInput util is part of public API. You can use its parseDouble(String, boolean) method. The boolean param controls whether Double.parseString (false) or the internal FastDoubleParser classes are used. This version of the method is new in jackson-core 2.14.x.