bpo-1635741: Port itertools module to multiphase initialization.

[corona10]

https://bugs.python.org/issue1635741

[corona10]

@shihai1991 Can you please take a look?

[vstinner]

LGTM. But I will wait for @shihai1991 review ;-)

[vstinner]

You forgot a typelist[i], I fixed it ;-)

[shihai1991]

LGTM ;)

[corona10]

You forgot a typelist[i], I fixed it ;-)

Thanks :)

[shihai1991]

there is one thing to be confirmed: race condition of identifier in typelists can be avoid too? no potential risk?

[corona10]

there is one thing to be confirmed: race condition of identifier in typelists can be avoid too? no potential risk?

Can you explain more concrete situation?

[vstinner]

there is one thing to be confirmed: race condition of identifier in typelists can be avoid too? no potential risk?

Currently, the GIL magically protects C extensions against race conditions. But the GIL can be released when executing arbitrary code. I don't think that it's the case here.

[vstinner]

Static types initialized by PyType_Ready() are bad. Types allocated on the heap using PyType_FromSpec() are better:
https://pythoncapi.readthedocs.io/type_object.html

But this should be addressed in a separated change ;-)