[fpc-pascal] Loss of precision with the overloaded Max() function

[Alan Krause]

An issue recently popped up that I thought I'd bring up as a potential
issue. In short, if you call the Max() function with an integer and a
double parameter (such as d := Max( 0, d ) ), then fpc appears to be using
the Max( single, single ) overloaded function instead of Max( double,
double ). This is causing a loss of precision once the double values get
large enough. Here is a sample program which illustrates the issue:

program project1;

uses
  Math
  , SysUtils
  ;

var
  d : Double;
  s : Single;

begin
  d := 159279.49;
  writeln( '159279.49 stored in Double d = ' + FloatToStr( d ) );
  d := Max( 0, d );
  writeln( 'Max( 0, d ) stored in Double d = ' + FloatToStr( d ) );
  writeln();

  s := 159279.49;
  writeln( '159279.49 stored in Single s = ' + FloatToStr( s ) );
  s := Max( 0, s );
  writeln( 'Max( 0, s ) stored in Single s = ' + FloatToStr( s ) );
  writeln();

  d := s;
  writeln( 'Assigning the value of s to Double d = ' + FloatToStr( d ) );
end.

Doing a bit of research on this issue, I see that it was brought up on
Stack Overflow back in 2018 here
<https://stackoverflow.com/questions/52020129/unexpected-result-from-max-function-in-freepascal>.
We can work around the issue by sending in 0 as 0.0, which then calls the
correct Max( double, double ) function.

Cheers,
  Alan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freepascal.org/pipermail/fpc-pascal/attachments/20240329/0defd8c5/attachment.htm>