Since after 1.7 the src/Tokenizer.cc code has been changed to use precomputed
tables for exponent, but now the compilation fails with
Tokenizer.cc:769:4: error: floating constant exceeds range of 'long double'
[-Werror=overflow]
exp_0_9(1E30)
This is because on powerpc, depending on libc, long double is either "quad
precision" double-double format (which has the same range as an ordinary
double, just higher precision) or simply the same type as a regular double.
Reverting to the older code fixes the problem.
- q66
