I think that [1,2] - [1,2] -> 0 is clearly a bug (i.e., I had no doubts),
since it creates an object that no longer has the same dimensions that it
would if the result were non-zero.
The concerns (not doubts) that I had was that it might break existing code,
something that always concerns me greatly.
By "regression testing", I meant I tested against both the testsuite and the
share test suite, under four different compilers, checking both that a) we
didn't introduce any new errors, and that b) with listarith=false, the
testsuite gives the same errors as before. Even as per Wikipedia's
definition, it does seem like "regression testing" to me. And this is
something I am indeed quite concerned about, as there have been more than a
few changes (including wholesale changes for the sake of cosmetics) that
broke things like the tensor algebra code in rather non-trivial ways, giving
me many hours of headaches, and I do not wish to return that favor to
others.
Anyhow, had I realized that by making this change, I am opening such a huge
and unpleasant-smelling can of worms, I'd never have done it. Life is too
short.
Viktor
-----Original Message-----
From: Barton Willis [mailto:willisb at unk.edu]
Sent: Monday, November 24, 2008 7:42 AM
To: Viktor T. Toth
Cc: 'Robert Dodier'; 'Raymond Toy'; 'Maxima List'; 'Viktor Toth'
Subject: Re: [Maxima] [Maxima-commits] CVS: maxima/src simp.lisp,1.60,1.61
-----maxima-bounces at math.utexas.edu wrote: -----
>Actually,?the?change?is?a?trivial?bug?fix,
Maybe this change is good, but it isn't trivial (insignificant
or self-evident (mathematics)). I think the change could break some
code.
The user-documentation for listarith isn't clear about this case, I
think. Has this bug been reported? I wasn't able to find it (I
searched the bug list for "listarith").
Suggestion (for everybody): If you have even the least doubt about a
bug fix: (1) file a bug report with a proposed fix (2) allow the
maxima community time to comment before making the fix (3) append a
regression test to the testsuite (4) reference the bug when you commit
the change and clearly state what the change does.
>which?I?regression?tested?with all?four?compilers?that?I?have?installed.
I'm guessing that Maxima has done [1,2] - [1,2] --> 0 for a few decades, so
this is testing, not regression testing. See:
http://en.wikipedia.org/wiki/Regression_testing
Barton