Fixed point type fractional digit limit

[axic]

Currently the limit is set at maximum 80 digits, that means the largest number representable would be
0.999999999999999999999999999999999999999999999999999999999999999999999999999999999 in the ufixed256x80 type, while the largest number fitting 256 bits is 115792089237316195423570985008687907853269984665640564039457584007913129639936.

999999999999999999999999999999999999999999999999999999999999999999999999999999999
is larger than
115792089237316195423570985008687907853269984665640564039457584007913129639936

It actually requires 266 bits (log2(pow(10, 80) - 1)). I think the proper limit is 77 digits.

Not to mention signed fixed point, where an extra bit must be reserved for the sign.

[randomnetcat]

Raised in the course of #3389 while discussing conversions.
Possible issue: this could overflow a u256 of 10^(abs(from - to)), causing issues with conversions.
Concerns raised in Gitter:

• Consider making it 76 instead instead of (possibly confusing) change based on signed or unsigned.
• Consider making max number of decimal places lower based on number of bits such that 1 is always in range.

[chriseth]

Could you please clarify what "it" mean? Is this about limiting the number of fractional digits in general, or depending on the bit size?

[randomnetcat]

Sorry, yes I mean the limit on fractional digits by it.