In another context someone here once cited the Irish washer woman who
feared that she "would grow bored with praising the Lord" for all
eternity, and I have similar reactions to praise of regression
testing.  (She was the same washer woman who hoped, since she  had no
voice, that she would get out of ther singing.)

Regression testing is too often simplistic.  An example will help me
to make clear what I mean..  During Y2K remediation testing in one
shop I was advising some hundreds of discrepancies, all suspiciously
of the same sign, arose.

Traced to their source---They were all traced!---they all stemmed from
use of the same 'guilty' subroutine, which I had written.  There was
of course some glee that I was the culprit.  Its defect was that it
determined leap years correctly using the Gregorian rule.  The code it
replaced had instead used the Julian leap-year rule, obsolete since
1582 or thereabouts.

What I found interesting was that some senior people argued
vociferously for retention of the old rule for the sake of consistency
with the past.  It was, they said, too late to use the correct rule.
My response was that it was too late to use the incorrect rule.
Fortunately, the lawyers intervened.  They said that retention of an
obviously incorrect rule, publicly known to be so, would have
horrendous legal consequences; and the new leap-year rule was adopted.

Something akin to regression testing is of course inescapable; but
comprehensive unit testing of new subroutines is better and, like
liquor, quicker.  One needs to be alerted to discrepancies, but one
also needs to remember that many things are done wrong in old code.

Emerson long ago observed that "a foolish consistency is the hobgoblin
of little minds".

John Gilmore, Ashland, MA 01721 - USA

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to