While reading Andrew's Forth for the 265SXB, I noticed that ROLL was "TODO" (https://github.com/andrew-jacobs/w65c81 ... -forth.asm). That's more than I did in Tali Forth, where I took one look at that word and decided to skip it, though it is CORE EXT (http://www.forth200x.org/documents/html ... #core:ROLL). This got got me thinking that I should man up and figure out how to implement it. I've come up with this algorithm as a suggestion.

We start with this model of what the Data Stack looks like:
Now, the Forth specs say that 0 ROLL is a null operation (which isn't really true, because n is dropped), 1 ROLL is SWAP and 2 ROLL is ROT. This gives us:The two steps of the algrorithm are now:This works because any entries between positions 2 and (n+2) retain their stack positions, which is slightly counter-intuitive for how you usually think of stacks but works if you think about the memory positions.

I haven't gotten around to any code yet (stupid glorious sunshine means I'm outside in the garden), but I wanted to put that up here for discussion. Will add an implementation for testing as soon as I can.

Yeah, ROLL is a cycle hog because of the move ... PICK is so much more desirable, and can be implemented in only two machine instructions plus a one machine instruction NEXT on my 65m32 DTC forth. When I get home, I'll see if I got around to coding ROLL in any of my unfinished forth projects, but I get the feeling that I put it off, due to the yucky feeling I get from dealing with something so inherently inefficient. Many hard-core Forthers don't like PICK or ROLL, but I think that PICK is a nice little hack inside a word that should probably be re-factored, but isn't, for whatever reason.

Mike B.

[PS: You're not really suggesting that the deepest stack items be moved to close the gap left during the ROLL, are you?]

This seems to work and isn't too monstrous

_________________
Andrew Jacobs
6502 & PIC Stuff - http://www.obelisk.me.uk/
Cross-Platform 6502/65C02/65816 Macro Assembler - http://www.obelisk.me.uk/dev65/
Open Source Projects - https://github.com/andrew-jacobs

Here's the one from PETTIL, that weird Forth with the split stack and separate TOS. Garth has been suggesting it probably isn't very good from day one, but I just wouldn't listen and now I am really beginning to rethink this split stack with separate TOS vs. the traditional FIG stack with all that (zp,X) addressing vs just a split stack. Evidence is hard to come by.

Still, this code today is 6 bytes smaller than the 34 bytes it took up yesterday. Thanks for nudging me to look at it.

The specs do not say that ROLL is a null operation when the top of the stack is 0 ( zero) but that 0 ROLL is a null operation. The stack before 0 ROLL is the same as after 0 ROLL. The 0 and ROLL have to be considered together. ROT is not a null operation but ROT ROT ROT certainly is.
Oh and here is Fleet Forth's implementation of ROLL

The branch to POP.JMP is a branch into ON

Since this source is for a metacompiler, as opposed to a regular Forth assembler, using a label does not increase the target size ( it only exists on the host.)

Cheers,
Jim