Back to the substance of this discussion, if you don't mind.
If a code change gives a performance improvement on the test suite of say
10%, that seems like a pretty good argument for it. After all, it seems to
be a *global* improvement that benefits everyone. But it may actually make
no difference at all in real usage. Speeding up 1000 cases of 0.5 sec each
by 10% is probably not noticeable by users if it's just taking a fixed 50mS
off the read-eval-print loop.
On the other hand, a code change may give *zero* measurable performance on
the test suite and still be worthwhile. After all, Maxima has many useful
capabilities, each important to particular users, and most of them not
exercised very heavily by the test suite. Would timing the test suite even
notice, say, a 10x improvement in the speed of factorization of large
sparse multivariate polynomials? or of the multiplication of large
bigfloats?
Now, would I recommend spending effort on a 10% improvement in those cases if
I were King of the World (or at least King of Maxima)? Probably not -- I'd
want to see 50% or more, since there are so many *functional* areas I'd
rather improve, especially if the change made the code harder to maintain.
But neither I nor anyone else on this list is King of Maxima. And the
particular proposed change is simple and local, and easily documented by a
1 or 2 line comment. Not a maintenance problem.
-s