windows (mingw64) fixes

[pdumon]

• if the output of sines generates an empty last line, a -1 char is returned somehow. read_input_data couldn't handle that
• multiplying inf with the min_err (e.g. 0) gives a NaN on windows/mingw64 (-1#INDe000) that won't compare with errk

[stevengj]

LGTM with some minor stylistic tweaks.

[stevengj]

Actually, I'm confused. What was the problem, exactly?

-1 is EOF, the end-of-file marker. In that case, fscanf in read_input_data should return nread == EOF, and the loop should exit.

[pdumon]

Actually, I'm confused. What was the problem, exactly?

-1 is EOF, the end-of-file marker. In that case, fscanf in read_input_data should return nread == EOF, and the loop should exit.

Ah, my bad. I couldn't find that -1 is actually EOF, plus I thought fscanf was actually reading a wrong value from the empty line output by sines. But you're right, the algorithm is fine as it is. The problem was downstream: just the NaN issue needs to be fixed. Just tested without the changes in eat_whitespace and read_input_data and works fine indeed.

[pdumon]

reverted the unnecessary 'fix'

[stevengj]

Tagged as version 1.4.2