BUG: Make f2py respect kind specifications for real parameters

[zerothi]

Fortran sources with parameters having kind-specifiers
where not correctly intercepted in the crackfortran.py
source. I.e.

subroutine foo()
real(8), parameter :: Pi = 3.141592635_8
end subroutine

The reason was a restrictive check for only integer specifiers
which did not split real's into the correct number.

Furthermore, several tests has been added which tests the different
kind specifiers and their use in codes, also all of them together.

[charris]

LGTM modulo some small nits.

[zerothi]

I add the comment here again because the referenced section is "outdated".

I am sorry that this commit is now much more complex... Could you please review the changes made regarding the complex case.

I realize that it still does not correctly evaluate the complex numbers. But providing a complete parser for the complex numbers is a bit difficult (and besides, all this fixes are warnings).

[charris]

How about you just make this about the real case and deal with the complex case in another PR? I'm happy to put the simple case in 1.12.0, but more complicated things might need to wait. I'm curious about the warnings, are you saying that this isn't needed except to avoid warnings?

[zerothi]

Ok, I will then probably not make the PR for the complex case, my current fix only partly solves the problem. I will comment in this PR what needs to be done.

As for the warning, without this PR the compilation of the fortran code will produce this:

get_parameters: got "unexpected EOF while parsing (<string>, line 1)" on '0.529177_sp'

while with the PR the above message is not present.

[charris]

Could you squash the commits into one? Use git rebase -i HEAD~4 and follow directions. The bug part of the first commit message can also be capitalized as part of the process.

[zerothi]

Ok, done! :)

[zerothi]

@charris done. Thanks.

[charris]

Looks good, thanks @zerothi .