How can we mix Python and Julia code without relying on finalizers?

[cjdoris]

e.g. we could have a macro which transforms ordinary Julia code to track which variables are Python objects and decrefs them as appropriate. These Python objects could just be pointers, as if we were writing CPython directly. We'd need to ensure that any Julia code mixed in could not raise uncaught exceptions.

[cjdoris]

e.g. maybe something like

@pyexpr (x::Py + y::Py)::Int

would convert x and y to Python objects (PyObject_From), add them (PyNumber_Add), and return the result as an Int.

Anything marked ::Py is coerced to a Python object. The semantics of an expressions NOT involving any Python objects are just ordinary Julia semantics. But anything involving Python objects can behave differently. e.g. in the above, + does not refer to the Julia function Base.+ but instead to PyNumber_Add. And expressions involving Python objects typically produce more Python objects. And the result can be coerced to a Julia type, such as with ::Int above.

This would require quite a bit of semantic analysis of the source code!

I guess it would pythrow() on any error (after freeing required resources) by default, but could have a onerror=(return PyNULL) argument.

[cjdoris]

In progress over on the dsl branch.

[cjdoris]

Made redundant by the rewrite now on master: all the benefit can be got using pydel! or @py.