pwh@nimbus.UUCP (Phil Hutto) (02/28/85)
Are people aware of the bugs in the definition of the above FP primitives in
the Berkeley implementation distributed with 4.2bsd? They are all very simple to
fix but cause functions using them to behave incorrectly.
distl,distr,tlr - All have a "fix point" of phi (<>).
All should return bottom (?) when given a null argument, yet they return null.
In other words, tlr:<> is <>; distl:<> is <>; distr:<> is <>.
distr - Doesn't check whether its argument is a "pair", a sequence containing
two elements. distr should be undefined for non-pair arguments yet if given
three or more arguments it simply returns whatever the result would be if
given only the first two arguments.
For example, distr:< <a> b c > is < <a b> >.
The "definitions" of each function for the above arguments are all clearly
given in both the FP manual and Backus' original paper. The functions as
implemented simply don't behave as advertised.
The fixes are simple. One possible set of fixes is:
{NEWtlr (null->%?;tlr) }
{NEWdistl (null->%?;distl) }
{NEWdistr (eq@[length,%2]->distr;%?) }
If anyone knows of other bugs, i'd like to know. We use FP in our graduate
programming language course and have had fair success yet bugs in the
implementation can be disheartening to a student trying to grasp the concept
of the language.
--
Phil Hutto
School of Information & Computer Science, Georgia Tech, Atlanta GA 30332
uucp: ...!{akgua,allegra,hplabs,ihnp4,seismo,ulysses}!gatech!nimbus!pwh