Avoid recursion during error handling

[tkf]

In the current master, self.sprint and self.showerror are used to handle errors:

pyjulia/julia/core.py

Lines 457 to 476 in e617ad9

	def check_exception(self, src=None):
	exoc = self.api.jl_exception_occurred()
	self._debug("exception occured? " + str(exoc))
	if not exoc:
	# self._debug("No Exception")
	self.api.jl_exception_clear()
	return
	# If, theoretically, an exception happens in early stage of
	# self.__init__, showerror and sprint as below does not work.
	# Let's use jl_typeof_str in such case.
	try:
	sprint = self.sprint
	showerror = self.showerror
	except AttributeError:
	res = None
	else:
	res = self.api.jl_call2(void_p(self.api.convert),
	void_p(self.api.PyObject),
	void_p(exoc))

They are set at the end of api.Julia.__init__

pyjulia/julia/core.py

Lines 432 to 433 in e617ad9

	self.sprint = self.eval('sprint')
	self.showerror = self.eval('showerror')

but self._call(u"using PyCall") has to be called before this to initialize pyjulia in the first place:

pyjulia/julia/core.py

Line 416 in e617ad9

self._call(u"using PyCall")

The error case in self._call(u"using PyCall") was actually handled in check_exception with the except AttributeError branch. However, my PR #162 just got pulled broke this horribly since accessing self.sprint invokes LegacyJulia.__getattr__ which then in turn tries to initialize api.Julia. There are several ways to fix it but the first thing we can do is to avoid inheritance and separate out the "magic" object api.LegacyJulia from the core api.Julia.

It fixes the error I mentioned in #175 (but not the issue itself).