Issue #38. Fixed rounding errors.

[kubsztal]

Corrected rounding errors by evaluating the number of significant digits in the real number and applying the .16g format only when more than 16 significant digits are present.

[jpgrayson]

Hi @kubsztal. Thank you for this PR, but I am not going to accept it.

From the VCD spec, IEEE 1364-2005, 18.2.1 Syntax of four-state VCD file:

A real number is dumped using a %.16g printf() format. This preserves the precision of that number by
outputting all 53 bits in the mantissa of a 64-bit IEEE 754 double-precision number. Application programs
can read a real number using a %g format to scanf().

The implementation in pyvcd is guided by that spec.

From #38:

the example 99.588 is generated incorrectly as r99.58799999999999

I don't think the formatted representation is incorrect; it is an artifact of the input and output being a base-10 representation, but the underlying floating point having a base-2 representation. A pretty good explanation of this can be found in the Python docs, here.

In short, 99.588 and 99.58799999999999 are both valid decimal representations of the same floating point value.

[kubsztal]

Thanks for the feedback @jpgrayson! You're right, it doesn't make sense to implement something that goes against the spec - I didn't realize the precision was explicitly defined there. I'll reach out to the GTKWave team to see if they can loosen up their interpretation of real values, since it looks like they're sticking to %.16g as well instead of just %g (or %lg).