Compiler should fall back to higher pickle protocols in some cases

[gilch]

There are cases where protocol 0 doesn't work at all, but a higher protocol might:

#> |class Box:__slots__='item'|
>>> class Box:__slots__='item'
#> .#(Box)
>>> # CompileError

(>   >  > >><__main__.Box object at 0x000001B3E0C344C0><< <  <   <)
# Compiler.pickle() TypeError:
#  cannot pickle 'Box' object: a class that defines __slots__ without defining __getstate__ cannot be pickled with protocol 0

#> (pickle..dumps (Box))
>>> __import__('pickle').dumps(
...   Box())
b'\x80\x04\x95\x17\x00\x00\x00\x00\x00\x00\x00\x8c\x08__main__\x94\x8c\x03Box\x94\x93\x94)\x81\x94.'

The compiler already checks if the highest protocol is shorter and uses it in some cases. It just needs to not give up if protocol 0 fails.

[gilch]

I think fix_imports should also be False. This is complicating the pickles for no reason. The default behavior is for backwards compatibility. Python 2 is way past EOL. Hissp's compiled output is simple enough that it can run on older versions than required by the compiler, but sometimes you need to be careful to avoid certain newer features. Pickles can be one of those features. Going earlier than 3.5 is not recommended. Python 2 is really not recommended.

I'm not using protocol 0 for Python 2 compatibility. It's mainly because it often uses fewer characters in the output, but also, it's at least somewhat human readable in the REPL for simple cases, which could make debugging easier. fix_imports is counterproductive there. Higher protocols typically add a header, consume four characters per byte, and can't easily be read short of disassembly. You'd have to memorize a lot more opcodes, and they're not mnemonic like protocol 0.