A bug?

The following generates an error:
phi(i,l,m,mu):= A[i,l,m]*%e^((l+1/2)*mu)+B[i,l,m]*%e^(-(l+1/2)*mu);

ans:[ phi(0,l,m,mu[0])-phi(1,l,m,mu[0])=0,
      phi(1,l,m,mu[1])-phi(2,l,m,mu[1])=0,
      phi(2,l,m,-mu[2])-phi(3,l,m,-mu[2])=0  ]$
  
for i:0 thru 2 step 1 do
  ap[i]:diff(ans,B[i,l,m]),
  am[i]:diff(ans,A[i,l,m]),
  for j:0 thru 2 step 1 do
      aap[j,i]:part(ap[i],j+1,2),
      aam[j,i]:part(am[i],j+1,2);
 
On the other hand, replacing the for loops with

for i:0 thru 2 step 1 do
  ap[i]:diff(ans,B[i,l,m]);

for i:0 thru 2 step 1 do 
  am[i]:diff(ans,A[i,l,m]);
  
for i:0 thru 2 step 1 do
  for j:0 thru 2 step 1 do
      aap[j,i]:part(ap[i],j+1,2);
           
for i:0 thru 2 step 1 do
  for j:0 thru 2 step 1 do
      aam[j,i]:part(am[i],j+1,2);
 
works as expected.

David