[AD] MID revisited

["Matthew Leverton" <meffer@xxxxxxxxxx>]

See bug #1640516 for reference and earlier threads for past discussions.

In short, the Allegro MID is a misnomer, as it really clamps between x
and z with the assumption that x is not greater than z. So, for
example, MID(5,y,0) will give unpredictable results.

The examples make use of MID, so it really is a public macro, even if
it isn't publicly documented (yet).

There are two choices:

1) Ignore it, but document it.
2) Rename MID to CLAMP and provide a proper MID.

I prefer #2, as it actually addresses the problem.

The two drawbacks of #2 are:

1) MID() is bound to be slightly slower than CLAMP(). So people
currently using MID() in add-ons, etc should change their code
accordingly. The macro I'm proposing only uses two comparisons for the
"in order" where x < y < z, which is the same as MID(). It uses three
comparisons for all other cases (except x > y > z, which takes two).

2) If such 3rd party code is changed to CLAMP(), the code would need
to maintain a version check and #define CLAMP MID if the code is meant
to work with 4.2 and previous. (Mainly affects libraries.)

Attached is a patch against 4.3.10plus and a test program. You can see
the output of the test program here: <http://pastebin.com/f5a19542>

So it would be nice to make a definite decision one way or the other
regarding this.

--
Matthew Leverton

#include <allegro.h>
#include <stdio.h>

int main()
{
	for (int x=1; x<=4; ++x)
	{
		for (int z=1; z<=4; z++)
		{
			for (int y=1; y<=4; y++)
			{
				printf("MID(%2d,%2d,%2d) = %2d\tCLAMP(%2d,%2d,%2d) = ", x,y,z,MID(x,y,z), x,y,z);
				if (x <= z)
				{
					printf("%2d", CLAMP(x,y,z));
					if (MID(x,y,z) != CLAMP(x,y,z)) printf(" - ERROR");
				}
				else
					printf(" undefined");
				printf("\n");
			}
		}
	}

	return 0;
}
END_OF_MAIN()

Attachment: clamp.diff.gz
Description: GNU Zip compressed data