gh-93820: Pickle enum.Flag by name

[serhiy-storchaka]

#93820.

[tiran]

A Flag instance with multiple members is serialized like this:

>>> (re.A | re.I | re.DOTALL | 0x1000).__reduce_ex__(1)
(<built-in function or_>, (re.ASCII, re.IGNORECASE|re.DOTALL|0x1000))

It looks like only the first member is reduced by name.

[serhiy-storchaka]

The rest is reduced recursively.

>>> pickletools.dis(pickletools.optimize(pickle.dumps(re.A | re.I | re.DOTALL | 0x1000, protocol=3)))
    0: \x80 PROTO      3
    2: c    GLOBAL     '_operator or_'
   17: q    BINPUT     0
   19: c    GLOBAL     'builtins getattr'
   37: q    BINPUT     1
   39: c    GLOBAL     're RegexFlag'
   53: q    BINPUT     2
   55: X    BINUNICODE 'ASCII'
   65: \x86 TUPLE2
   66: R    REDUCE
   67: h    BINGET     0
   69: h    BINGET     1
   71: h    BINGET     2
   73: X    BINUNICODE 'IGNORECASE'
   88: \x86 TUPLE2
   89: R    REDUCE
   90: h    BINGET     0
   92: h    BINGET     1
   94: h    BINGET     2
   96: X    BINUNICODE 'DOTALL'
  107: \x86 TUPLE2
  108: R    REDUCE
  109: h    BINGET     2
  111: M    BININT2    4096
  114: \x85 TUPLE1
  115: R    REDUCE
  116: \x86 TUPLE2
  117: R    REDUCE
  118: \x86 TUPLE2
  119: R    REDUCE
  120: \x86 TUPLE2
  121: R    REDUCE
  122: .    STOP
highest protocol among opcodes = 2

[tiran]

copy.copy does not recurse. Is that ok?

>>> copy.copy(re.A | re.I | re.DOTALL | 0x1000)
re.ASCII 4114
re.ASCII|re.IGNORECASE|re.DOTALL|0x1000
>>> copy.deepcopy(re.A | re.I | re.DOTALL | 0x1000)
re.ASCII 4114
re.ASCII 0
re.IGNORECASE 4112
re.IGNORECASE 0
re.DOTALL 4096
re.DOTALL 0
re.ASCII|re.IGNORECASE|re.DOTALL|0x1000

output with print(m, rest) in __reduce_ex__.

[serhiy-storchaka]

I think that it is okay. Enums are considered singletons (it allows to pickle them by name). It would be even better to add the __copy__ method which returns self.

This PR is not ideal:

1. The residual member is reduced as re.RegexFlag(0x1000) instead of just 0x1000.
2. Named flags are reduced as getattr(re.RegexFlag, 'ASCII') instead of global re.RegexFlag.ASCII or even re.ASCII.

It is a matter of the following optimizations.

[miss-islington]

Thanks @serhiy-storchaka for the PR 🌮🎉.. I'm working now to backport this PR to: 3.11.
🐍🍒⛏🤖 I'm not a witch! I'm not a witch!

[miss-islington]

Sorry, @serhiy-storchaka, I could not cleanly backport this to 3.11 due to a conflict.
Please backport using cherry_picker on command line.
cherry_picker 536985814a7116f14c9bc90aa1b3e3d36d5b2367 3.11

[bedevere-bot]

GH-94288 is a backport of this pull request to the 3.11 branch.