I'm currently implementing SET-ORDER in Tali Forth 2. The ANS 2012 standard says at https://forth-standard.org/standard/search/SET-ORDER the following:


For -1 as the input, I'm just setting FORTH-WORDLIST as the 1 wordlist, as it contains both FORTH-WORDLIST and SET-ORDER as the standard dictates. This is my "minimum search order".

My question has to do with passing 0 for n (with no wordlist IDs). The standard is saying to empty the search order, but wouldn't that make all the words disappear? My thought is that, in this particular case, the minimum search order would also make sense... but the comment for the minimum search order looks like it only applies to -1 on the stack.

The example implementation (available at the above link) seems to simply empty the entire search order, and that's currently what my code does as well. That seems like a mean thing to do to a user, but Forth is very much a no-training-wheels kind of language, and technically they asked for it if they typed "0 SET-ORDER".

I'm curious what others think of this situation and what the best behavior might be.

I think that the distinction between a "minimal" search order and an "empty" search order is, and must be, significant. If the search order is "minimal", there's still a wordlist there, since it must include at least FORTH-WORDLIST and SET-ORDER. If there are no wordlists in the search order, no words can be found.

I shall also take the position that -1 SET-ORDER (to set a minimal search order) should leave an entry on the stack, preferably -1. Justification: ( widn ... wid1 n -- ) absorbs n wordlists, and n. If n is zero, it absorbs zero wordlists and n. If n is -1, it clearly _produces_ a stack entry after removing n, and that stack entry might as well be n. Counterarguments appreciated.

I like your thought that -1 should "remove" -1 items from the stack, thereby leaving one. It took me a couple of minutes to figure out what you were talking about.

The example implementation they give is:

It looks like they are just using the -1 as a flag, replacing it with whatever wordlists (more than 1 seems to be allowed) and an appropriate n before the code continues processing the normal way. #order is a variable holding the number of wordlists in the search order, and context is an array that holds the wids (Wordlist IDentifiers).

If you do set 0 wordlists in the search order, then every word you type after that point is undefined, right? There's no way to fix that except to reset the system. That's the part that I'm having trouble with, but it seems like the standard calls for that behavior (if the user asks for it, of course).

It might not be particularly useful to leave the system with no wordlists in the search order, but that doesn't mean that it might not be useful for the system to be in such a state temporarily, especially in the context of implementing some higher-level vocabulary mechanism. And the SEARCH-ORDER word set was defined (per the dpans94 rationale) as a primitive set from which various other schemes could be defined.

That makes a little more sense to me, nyef. A compiled word has already looked up all of it's words; it could temporarily remove all of the wordlists from the search order while it's running, as long as it put something back before it went back to interactive mode (if it ever goes back to interactive mode). Thanks!

That's the way I read it too.

I you accidentally end up with an empty word search order, you're pretty much out of luck and need to restart your system. In that state, all you can theoretically do is put numbers on the stack.

I'm sorry to say this, and I recognize there was lots of historical baggage that the ANS people had to deal with, but the whole wordlist structure of ANS Forth is too complicated and downright nuts. It's no wonder people still use vocabulary. Gforth makes it a little better with >order (https://www.complang.tuwien.ac.at/forth ... Lists.html) which puts a wid at the top of the search order.

Now admittedly, the fact that you can crash your system with the suicide sequence 0 set-order is very useful if you have a nuclear missile silo programmed in ANS Forth and Dr. No is about to break into the control room. I'm at a loss for other uses, though. And don't get me started on the fact that order displays the wordlist order in the exact opposite way of the normal stack notation, that is, top wordlist on the left.

Thank you, I feel better now. It's probably best that Sam is taking care of the wordlist stuff for Tali Forth, not me.

One of the reasons I still use Forth-83.