While investigating https://bitbucket.org/pypa/setuptools/issue/388/install-from-sdist-fails-on-python-350b1 I noticed that Grammar/Grammar changed in 3.5, but Lib/symbol.py wasn't updated. I'm not familiar with the CPython parser, but I suspect that adding/removing/splitting grammar rules causes the nonterminal symbol IDs to shift, which ought to require an update in symbol.py. Huh. Now I see a comment in the file says it is automatically generated, but in that case why wasn't that done when I did 'hg pull -u && make && make install'? Why is it checked into source control?
It is "automatically generated" in that it isn't hand-written. On the other hand, it isn't "automatically generated" in the sense of being part of the make process, ./python symbol.py is supposed to be run by hand when it is appropriate. A bit ago someone wrote tests for keyword.py that among other things made sure we didn't forget to update it when needed. Someone needs to write a similar test for symbol, it looks like. Whether or not one or both of these could be/should be incorporated into make (now that we have 'make touch' to deal with the consequences) is a separate question. As to why it is checked in, we check in almost all the build artifacts previous to the compile stage, so that there is no need to have an already-built python to build python from source.
