Any plans to simplify customising float representation?

[jbasko]

If one needs to change the representation of floats then they have to extend the encoder and override iterencode by copying over most of the code and just:

1. replacing the _repr=FLOAT_REPR to something that they need.
2. removing the C encoder option

[etrepum]

No, this is not planned. Why would you want to do this?

[jbasko]

Because I need to reproduce behaviour someone has borrowed from pandas to_json which they called with set double precision on float serialization. The code is used for hash calculation so pretty critical to repeat that silly behaviour.

[etrepum]

That's a very unfortunate scenario, hashing a json encoded string is just asking for trouble. floats are printed at full double precision of course, but an optimal algorithm is used (the fewest digits that will round-trip through a correct parser exactly). The C extension can be disabled (globally), although this is more of a private implementation detail used by the test suite.