Changing this isn't really feasible with the current design of the initialisation code - we call Py_FatalError in various places because we don't have the infrastructure set up to do anything else.

PEP 432 should help (and the basic design there still seems sound), since we'll have the core interpreter up and running much earlier, and hence be able to make use of the non-fatal exception machinery. However, it would still mean carefully reviewing everywhere we call Py_FatalError and ensuring there's an error returning path back up to the embedding application.

Unfortunately, I don't know when I'm going to find time to get back to that, and the startup process is sufficiently arcane that it wouldn't be a quick process for someone else to come up to speed on the current state of the draft implementation :(