Kijewski
Switch from fast_double_parser to fast_float
First off, thanks for maintaining this library! I'm trying to pitch in a bit to help and make sure things stay healthy. It's my first time working with Cython, so I'll be a little slow to make changes.
This PR swaps the fast_double_parser for fast_float, per the deprecation warning cited in #83. It all works cleanly except for a small number of tests regarding scientific notation.
I could use some input on how to push this over the finish line.
None of the fast_float behaviors matches what the JSON5 tests expect, and the one which is labeled json_or_infnan seems like the best fit for the spec in spirit.
Right now, I'm seeing pyjson5.loads(1e2.3) return 100, which is definitely wrong.
But I haven't yet figured out where the issue is. I would guess (from the above) that somehow it's truncating the data to 1e2.
Thank you very much for working on that! I'll try to have a look in the next few days.
(I wonder why the CI did not run for you PR, I'll have to investigate that.)
I was able to confirm that the string "1e2.3", when passed to fast_float, parses "successfully" as 100.0.
I couldn't find a way to get it working using clever fast_float usage, but once I satisfied myself that it wasn't possible, I added a manual little function to guard against weird invalid exponents (any non-digit char after the sign).
I think it's now in good shape -- all tests pass -- and I hopefully have some more contributions to follow!