Fix indexing of self.args when passing Python object as varargs parameter

[Baekalfen]

This avoids a "Compiler crash in AnalyseExpressionsTransform", although still produces a compile error as intended.

The reason for the issue is that i comes from the length of args, where there is an additional object first in the list. The self.args is therefore offset by 1.

Bonus question: Why is this error here? I found this, as I was monkeypatching this error out. Seemingly, there's no issue passing PyObject* as a vararg parameter.

[da-woods]

Bonus question: Why is this error here? I found this, as I was monkeypatching this error out. Seemingly, there's no issue passing PyObject* as a vararg parameter.

I think the idea is that we treat object and PyObject* as different - the former has its reference counting managed by Cython while the later is just an opaque pointer type. So you could always cast to <PyObject*> before passing. It's probably reasonable - it at least forces the caller to think about the reference counting implications of passing a Python object argument.

[Baekalfen]

Understandable. I'm using Cython only in Pure Python Mode with pxd's, so I can't just cast to PyObject*. I'll probably keep my workaround for now, as just use it for logging, and don't really need to ref-count.

[da-woods]

This looks good to me - thanks.

I suspect this should go in the next 3.0 bugfix release. But we're currently in the process of branching off the new development branch so I'm not 100% sure where it should go, so I'll wait a little bit to see if there's any objections.

[da-woods]

Thanks

[da-woods]

Backported to 3.0.x in 101b492