The ANS Forth provides an extra wordset dedicated to LOCALS which happens to be critized by about every forth developer around. The problem is the inverse order of declarations compared with the normal stack comment. So, if your word expects three values as input parameters, you would usually declare it with a stack comment like this:

 : MY-WORD ( a b c -- x )
     whatever instructions
 ;

however if you want these three values to end up in named places on the return-stack, you have to write this one with the following LOCALS syntax:

 : MY-WORD LOCALS| c b a |  ( a b c -- x )
   whatever instructions
 ;

To understand this scheme, we have to look at the underlying word (LOCAL) which is also provided in the ANS Forth LOCALS EXT wordset. This word is not quite useful for the user, it is meant to be used by compiling words like LVALUE to declare new LOCALS to the LOCALS machine in the system. The PFE has such an LVALUE word that declares the following name to be a new LOCAL, and it initializes it in the course. To declare three values, we would have to write this:

 : MY-WORD ( a b c -- x )
   LVALUE c
   LVALUE b
   LVALUE a
   whatever instructions
 ;

and of course, this order is necessary since the first value is the one on top of the stack being the rightmost in the stack comment declaration. The ANS Forth LOCALS|-word does simply follow this order, making the name "c" to be in the first place and therefore leftmost in its simple syntax.

Please note that PFE does not follow the solution that some other forth systems implement - using another syntax extension that does the locals declaration in the "correct" order. However, a scheme like the John Hopkins Scheme adds a lot of complexity - the declaration must be parsed and reordered to match the internal order of initializations, and in order to provide LOCALS types other than single-value items, they must extend the parsing code even further. Instead, just skip that and follow the forthish style, using the simplest syntax that is still best to read - declare each local-value seperately.

One of the explicit locals has been describe above - called LVALUE to remind you of VALUE which has the same stack execution in declaring global values. The same has been done for the other local declarators - their names shall match the declarators that do the same thing for global items with a single "L" prepended. Currently, the following two local declarators are available in PFE

  LVALUE ( a [name] -- ) : [name] ( -- a )
  LBUFFER: ( size [name] -- ) : [name] ( -- addr-of-buffer )

feel free add more. I know one project that had a good use for LFCONSTANT but there is no floating extensions module so far - although I would like to include one if you have one written.

One nice side effect of explicit Locals is the availability of init computations that you can not have with any of the compound declarations. Have a look at this example to get an idea:

  : SAVE-BUFFER-TO-FILE ( buffer-addr buffer-len file-name -- done? )
     COUNT R/W OPEN-FILE IF DROP FALSE EXIT THEN
     LVALUE FILE-ID
     LVALUE BUFFER-LEN
     LVALUE BUFFER-ADDR
     whatever needs to be done else
     BUFFER-ADDR BUFFER-LEN FILE-ID WRITE-FILE
     and so on
  ;

well, this program does nothing so complex that it would require the use of locals to enhance readibility and maintainability. Just get the idea that we did more that the pure ANS Forth locals can provide for. However, be aware that PFE does not do sanity checks so far - putting an LVALUE inside a LOOP would not be that wrong (it would just reinitialize the value at that point over and over again) but the size argument to LBUFFER: is taken at runtime, and the needed space is carved from the return stack just then. Running this one inside a LOOP or mixing such a declarator with words that affect the return-stack (like >R or R>) could lead to undesired effects that are not warned for at compile time.

Looking close at LBUFFER: a C programmer could compare the functionality to that of alloca(). The declared name is actually a simpel LVALUE that will be assigned space carved dynamically form the return-stack - and automatically FREEd on any EXIT point in the function since the LOCALS frame is given back. This can be used to change the habit of many programmers to write non-reentrant forth words whenever a local buffer is needed, for instance to combine a file name with a prefix in a local string buffer.

Looking at the ANS Forth standard, one will find another reason for this habit since the standard does not require the return-stack of forth to be very large - that was done so the complete return-stack could be left in some hardware on special Forth CPUs - the threading model of forth does benefit a lot from call/return optimizations on the CPU, making it a lot faster. And if you look through the ANS Forth documents, you will see that there is no requirement for RP@ to exist, or that any item on the return-stack has an address. All this is purely optional. However - all the forth system running inside a host OS on general-purpose CPUs will provide a proper return-stack plus and the availability to take references of the values put there.

There is another way to help around the problem of a temporary string buffer to do some concatenations and splitting - it is traditionally called a POCKET. The first forth systems had just one such global string buffer that was reused in every word. Of course, that could easily to lead to problems with a word using it and calling a word thereafter that uses the POCKET too. Plus you can not have two string buffers with a single POCKET.

That's why quite some forth systems started to provide more than one POCKET, and a developped system like PFE provides a word to automatically access each temporary buffer in a round-robin fashion to lower the potential problems of called-words to overwrite the current buffer. Be aware that this can still happen here, and that many of the PFE system words do actually use the dynamic POCKETs internally to lower the return-stack pressure.

The word to get a new POCKET is simply called POCKET-PAD to remined you that the returned pocket buffer is not exclusive to the function requiring it. And it is sometimes a good idea to put this address directly into a LOCAL value to have name for it. It will look like this:

  : MY-WORD ( filename -- )
    POCKET-PAD LVALUE filename
    S" /prefix/path" filename PLACE
    count filename +PLACE
    (USE-WORD)
  ;

and please notice how used the pocket-pad for just a short time to construct an argument to (USE-WORD). This is the recommended use, since it is hard to guess what some of the called words in MY-WORD will actually do, and whether they request POCKETs too, and finally getting hold of the same one that was used here. In most of the forth programs one can come up with a good expected value of POCKET allocations in lower words however, and a pocket-pad does lower memory pressure quite a bit. And a POCKET method scheme can be easily ported to systems that have a non-addressable hardware return-stack.