#1932 expand/taylor fail for noncommutative math

[David Ronis]

Consider the following two tests:

f1:1-epsilon*H1+epsilon^2/2*H1^^2;
f2:1-epsilon*H2+epsilon^2/2*H2^^2;
f3:1-epsilon*H3+epsilon^2/2*H3^^2;

a:f1.f2.f2.f1;

expand(a);

taylor(a, epsilon, 0, 2);
map(factor,%);

a: f1.f2.f3.f1.f3.f2.f2.f3.f1.f2.f1.f3.f3.f1.f2.f3.f2.f1;

expand(a);
taylor(a, epsilon, 0, 2);
map(factor,%);

The first works. The second hangs at the second expand.

I'm running the CVS head version