I don't know, to me this is a bit like taking two c programs that print "hello world" and "goodbye world", looking at the assembly that is compiled then remarking "pretty sad and very unmaintainable."
Point being, maintain the code generator and not the generated code.
Right. But here the generated code is checked into the source tree and the code generator is kept proprietary, which is why it's sad and unmaintainable.
They could still separate the extraction of the data (keeping it proprietary) from the code generation. The latter could work off of the extracted data kept in a more sensible format.
Point being, maintain the code generator and not the generated code.