[Python-Dev] Another case for frozendict

[R. David Murray]

On Wed, 16 Jul 2014 10:10:07 -0700, Devin Jeanpierre <jeanpierreda at gmail.com> wrote:
> On Wed, Jul 16, 2014 at 6:37 AM, R. David Murray <rdmurray at bitdance.com> wrote:
> > IMO, preventing someone from shooting themselves in the foot by modifying
> > something they shouldn't modify according to the API is not a Python
> > use case ("consenting adults").
> 
> Then why have immutable objects at all? Why do you have to put tuples
> and frozensets inside sets, instead of lists and sets? Compare with
> Java, which really is "consenting adults" here -- you can add a
> mutable object to a set, just don't mutate it, or you might not be
> able to find it in the set again.
> 
> Several people seem to act as if the Pythonic way is to not allow for
> any sort of immutable types at all. ISTM people are trying to
> retroactively claim some standard of Pythonicity that never existed.
> Python can and does protect you from shooting yourself in the foot by
> making objects immutable. Or do you have another explanation for the
> proliferation of immutable types, and the inability to add mutable
> types to sets and dicts?
> 
> Using a frozendict to protect and enforce an invariant in the re
> module is entirely reasonable. So is creating a new dict each time.
> The intermediate -- reusing a mutable dict and failing in
> incomprehensible ways if you mutate it, and potentially even crashing
> due to memory safety issues -- is not Pythonic at all.

You'll note I ended up agreeing with you there: when mutation breaks an
invariant of the object it came from, that's an issue.  Which would be
the case if you could use mutable objects as keys.

--David