newRPL - build 1255 released! [updated to 1299]

[3298]

I just typed your example into a 50g (well, technically x49gp since I'm also currently developing a patch series for it, but who cares), and found that my memory failed me: an unquoted variable does not automatically EVAL a contained symbolic.
On the other hand, EVALing the symbolic does recurse, i.e. 'E' RCL EVAL returns 1771566 (and it doesn't matter if A and B are reals or integers). If I store 1 into A, the result of 'E' RCL EVAL will be 117650. Nothing was stored back on the previous evaluation of E when A was still 5.
Even better, if I purge A, then 'E' RCL EVAL returns the symbolic 'A^6+36*A^5+540*A^4+4320*A^3+19440*A^2+46657*A+46656'. When this would be encountered in a form, it would be quite obvious from this that A is missing.
Circular references are detected too: I stored '5+E' into A and tried to do 'E' RCL EVAL. Result: "EVAL Error: Circular Reference"
(Edit: all of this was done on the stock firmware, but it's so basic that I would be surprised if newRPL was incompatible.)

I get the impression you're thinking of the new EVAL1, but EVAL is still there.

What I'm trying to say is this: There's no need to invent a new system, all the pieces are already there. In a solver form, show the result of <variable> EVAL instead of the variable itself (and preferably catch errors such as the circular reference one, giving them special treatment such as displaying the error text in place of the result). Job done.