Will use as few digits as is necessary to ensure that round-tripping the same number with pstrtod will result in the same vector. This prevents eg. 3.3 formatting as 3.29999995 while still ensuring that two floats that are not equal (other than nan) are guaranteed to have a different string representation.
Uses a hacky pftoa function that can be abandoned as soon as we adopt C++17, which has to_chars that does what we need
Fixes#1671
This is more idiomatic, simplifies the code, makes it easier to port to eg. HPy later on, and makes it easier to use the limited ABI (where it can be optimized to a tail call) later on
A polyfill is provided for compatibility with Python versions 3.8 and 3.9
* Better support for typedefs
* Same code should compile for both 32-bit and 64-bit
* C++11 typed enum support
* Synthesize __setitem__ when operator [] returns reference
* Rewrite coerce system to have better performance
* Change semantics of __getattr__ and __setattr__ to match
Python's, add __getattribute__, __delattr__ and __delitem__
* Improve performance of slot functions a bit more
* Reduce memory overhead of type system
* Some support for wrapping arrays of numeric types
* C++11 character type support in cppparser
* Chars are handled as strings of length 1
* Template functions for CreatePyInstance that use runtime type map
* More functions from dtool as extension functions
* Code cleanup
rdb
Leandro (Cerberus1746) Benedet Garcia
Sam Edwards
tobspr
David Rose