libpascalreal - C APIs for using Turbo Pascal 48-bit floating-point numbers

This is a (very) small library meant to provide interfaces for converting Turbo Pascal real into C double and back.

Information about this format is scarce, and it appears to never have been officially documented, so this implementation (like any other) cannot be reliably treated as a reference one.

Compilation

You're encouraged to integrate this library directly into your project. The only thing that's needed is a C99 compiler with a math library.

Known issues / caveats

• The library was not tested with a compiler or platform that doesn't use the double-precision IEEE 754 floating-point format for representing the double type.
• Denormals, NaNs and infinities are converted to negative zero.
• Numbers too large to be expressed as a real are converted to the maximum legal value.
• Similarly, numbers too small are converted to zero.

The format

8-bit exponent with an exponent offset of 129, 39-bit mantissa and a sign bit.

According to Delphi in a Nutshell by Ray Lischner released in March 2000, this type is (was?) supported by Delphi as well under the name Real48.

Additionally, it straight out states that

It cannot store denormalized values, infinity, or not-a-number. If the exponent is zero, the number is zero.

(source)

Memory layout

 0        1        2        3        4        5        
 76543210 76543210 76543210 76543210 76543210 76543210 
+--------+--------+--------+--------+--------+--------+
|EEEEEEEE|MMMMMMMM|MMMMMMMM|MMMMMMMM|MMMMMMMM|SMMMMMMM|
+--------+--------+--------+--------+--------+--------+

Real-world usage

According to Modding Wiki, this binary format is used by save game files by some games written in Turbo Pascal.

It is also used in a Polish industrial automation protocol called GAZ-MODEM (later renamed GAZ-MODEM 1), standardised by PGNiG in ZN-G-4007:1995 Appendix A.

Legal

The licence of the library itself is available in LICENCE.BSD.