commercial macsyma, with newsimp loaded, gives the answer
((2 * x)/(sqrt(%pi))) - ((2 * x^3)/(sqrt(%pi) * 3)) + ((x^5)/(sqrt(%pi)
* 5)) + . . .
without newsimp loaded it gives the answer
((2 * sqrt(%pi) * x)/(%pi)) - ((2 * sqrt(%pi) * x^3)/(3 * %pi)) +
((sqrt(%pi) * x^5)/(5 * %pi)) + . . .
they are equivalent since sqrt(pi)/pi - 1/sqrt(pi).
The internal forms differ in that one has 1/2 the other ((rat) 1 2)..
So my conclusion here is that we have a bug in GCL in its hashing. I
cannot find a simple
way of reproducing it, but somehow there is a hash table which maphash
says has
a value -4
but (gethash ..) says has a value (1 . -4).