LongCalc 0.5.30

• • Mpir.NET (>= 0.4.0)

Unsigned Assembly

Something prompted me to determine exactly what happens in the more current versions when 'growing' an mpz_t or bf object above the limit of 2 ^ 37 bits. I had always assumed it would cause a program crash much as what happened when 'growing' numbers in versions before 0.5.0. I found that what happens is, if possible, worse. The sign bit on an mpz_t item appears to occur at bit# (2 ^ 37) - 64, and gets overwritten in an unprotected shift left, add, or multiply operation. Basically, the earlier 0.5.xx versions returned inaccurate results when performing operations that resulted in output longer than (2 ^ 37) - 64 bits.

This version attempts to fix that problem. The Add, Subtract, Multiply, Divide, and shiftLeft functions have been modified to restrict bf and bd objects to have an mpz_t integer component shorter than (2 ^ 37) - 64, partly by restricting the decimal Precision parameters to be always less than or equal 20686000000 (which is a decimal equivalent to nearly 1/2 of (2 ^ 37) - 64 bits). This should result in accurate outcomes of objects, even when squaring an item or multiplying any two valid bf or bd items together.

This version also includes an updated and enhanced LongCalcDemoApp executable.

Checks for arithmetic overflow or underflow are again compiler-included in this release.