This program: ```pascal const NegInfinity: single = -1.0 / 0.0; begin writeln(-0.0); writeln(1.0 / (-1.0 / 0.0)); writeln(1.0 / NegInfinity); end. ``` fed to the trunk compiler prints ``` 0.000000000E+00 0.000000000E+00 -0.000000000E+00 ``` Third minus zero is thanks to the division not being folded; with true (untyped) `const NegInfinity = -1.0 / 0.0` it is folded completely into a positive zero, as in the `1.0 / (-1.0 / 0.0)` case. There was a time when all of the three were giving minus zeros, see e.g. output with FPC 3.0.4: https://ideone.com/m17u8F. Programming languages generally respect minus zeros: https://ideone.com/1H4ptw (C++), https://ideone.com/qZHdIg (Rust). Mainly because dividing 1.0 by +0.0 and by −0.0 gives two extremes, +∞ and −∞.