11 Mar
2016
11 Mar
'16
12:09 a.m.
From the standpoint of a compiler writer, the increment
operators (and
later, similar operators like &=) were surprisingly painful, on
two
levels.
The first problem is that there were lots of nasty interactions between
these operators and function calls. Consider gems like:
f( ++i, ++i );
f( ++i, i++ );
f( i++, ++i );
f( i++, i++ );
The "natural" behavior depended critically on whether the stack grew
upwards or downwards. Add to this the delights that arose if i was an
external variable that was also referenced (or changed) within f. This
may seem like an unlikely event, but it happened all time when the macro
implementation of getchar was introduced.
The second problem was that most victims of the above had a lot of
difficulty, at least the first time, in understanding what the bug was and
tracking it down... And many considered the behavior they saw a compiler
bug (no mater what the documentation and/or the Standard said about
"undefined")...