Interesting seeing how HP calculator people argue about the merits of the size of the RPN stack. Here's my take.
I recently bought a DM-42. I've been aware of it for a long time, but I didn't even consider purchasing it until it had an n-stack. I learned RPN in the context of paper and pencil, not calculators, although I did use dc on Solaris and Linux back in my terminal-using days.
The first discrete RPN calculator I ever owned was an HP50g, which is an impenetrable monster that eats batteries. Still, the RPN was a delight (and I
know people here say it isn't RPN, it's RPL, but remember I learned RPN in a non-HP, non-calculator context). I could just work from left to right, pushing elements up the stack until I needed them again, without having to work "from the inside out". Also, I could just leave previous results in the stack, and start new calculations, leaving me with a chronological list of results in the stack.
My second RPN calculator was the DM16L. I love that calculator for its first-class bitwise functions, but I hate its tiny stack, where intermediate values get replaced with previous intermediate values and I get wrong results whenever I try anything more complex than "what's the state of a sixteen bit register when I mask out the most significant byte". Leaving results on the stack is impossible, of course - I either need to either get paper and a pen out, or just ditch the DM16L and use a computer. Of course, the lack of stack space leads to other annoyances, like functions taking parameters after the function, in prefix mode. I still find myself stumbling over the fact that I change the display mode by hitting FIX then the number of places I want to see, but change word size by pushing the number of bits I want onto the stack and then hitting WSIZE. Only the latter feels right.
I didn't need an HP50g. That thing's an awful experience (I still don't know how to graph with it). I did learn how to do my bitwise stuff on it, but it wasn't particularly pleasant. I didn't go for the other RPN calculators that were on the market (basically, the HP35s and the DM42) because a four level stack, to me, is
crippled. I have to do so much work myself keeping track of how deep I'm pushing it, and if I fail to do that I'm silently punished with wrong results. I dislike the experience intensely.
At university, in my programming lectures, I was taught to accommodate either zero, one, or an arbitrary number of things, and to avoid hard-coded limits. Four is such a limit, and since hardware is very much more capable now, attempts to rationalise it seem... irrational.
When Free42 got the NSTK mode, and the DM42 adopted it, I was suddenly very interested in acquiring a DM42 (especially with the BASE mode being made actually useful as well). I now very much enjoy using it, even though it still has all those annoying functions that take arguments
after the fact.
For the record, my favourite programming language is Tcl. Polish notation needs extra work when there are variable numbers of arguments, but it's still way more lovely than infix.