Help me to understand FLOAT strange behavior

anatol · Posted: Thu May 20, 2010 6:52 pm

Hi All,

Could you please help me to understand FLOAT strange behavior. Here is code:
DCL AF FLOAT(15);
DCL AFL FLOAT(18);
AF = 6.66666666666;
AFL= 6.66666666666;
PUT SKIP DATA(AF,AFL);
IF AF = AFL THEN
PUT SKIP EDIT('ARE EQUAL')(A);
ELSE
PUT SKIP EDIT('ARE DIFFERENT')(A);

The output is for MVS PL/I:
AF= 6.66666666665999E+00 AFL= 6.66666666665999999E+00;
ARE DIFFERENT

The output for PLI Enterpize:
AF= 6.66666666666000E+00 AFL= 6.66666666666000000E+00;
ARE DIFFERENT

The questions are:
1.Why the AF and AFL are different - the value are the same
2. Is it possible to get the same result in PLI MVS and PLI Enterprize ( maybe some compile options)

Robert Sample · Posted: Thu May 20, 2010 6:59 pm

Floating point values are notorious for not comparing equally -- a single bit difference will keep the two values from being compared equally. Obviously, the difference between FLOAT(15) and FLOAT(18) is enough that the bits do not compare the same.

As far as I know, this behavior is true for all floating point variables -- whether in Assembler, PL/I, COBOL, or any other language you care to name. Equality of floating point variables is not usually tested for this reason.

anatol · Posted: Thu May 20, 2010 7:17 pm

Thank you, Robert.

is it possible to get the same result ( AF & AFL compare was for test only) for MVS PLI and Ent PLI? maybe use special compiler options... The differencies are:

MVS PLI :

AF= 6.66666666665999E+00 AFL= 6.66666666665999999E+00;

Enterprize PLI:

AF= 6.66666666666000E+00 AFL= 6.66666666666000000E+00;