There are a big variety of Forth systems, and some of the differences make no difference to code. Which ones do? Here's a first attempt at that.

Some of the biggest differences between Forth systems:

  • HostedForth – target system has plenty of memory, already has a host operating system, and stores standard ASCII text files on mass storage using that target's file system.
  • StandAloneForth – all necessary OS functions, such as the file system, implemented in Forth. Since no underlying OS is needed, can run on systems a bit too small to support (what's this called?).
  • UmbilicalForth – Some functions (the dictionary, some compiler words, ... all the source code) run on a host operating system. Other functions run on the target system. Typically used when the target system has no mass storage, and too little RAM to support a full StandAloneForth. Allows interactive development on systems that don't have the RAM to support any other kind of Forth.
  • CrossForth – All interative development runs on the host. When the application is "finished", an executable image file is produced and burned into the ROM of the target system.

Other differences between Forth implementations:

Usually I figure if you need a specific word then you can use it if it's present. Sometimes there's more than one common usage with the same name, or it's worth pointing out the use.

  • 0 Memory
    • 00 Do I need a particular size of address unit?
      • 000 don't care
      • 001 8-bit
      • 002 same as a char
    • 01 What size char do I need?
      • 010 at least 8 bits
      • 011 at least 16 bits
      • 012 at least 32 bits
    • 02 What size cell do I need?
      • 020 at least 16 bits
      • 021at least 32 bits
      • 022 64-bit
    • 03 What about big-endian, little-endian?
\ matters when doing C@ C! etc on cell-size data
\ and when reading cell-data that comes in from outside
      • 030 don't care
      • 031 big-endian
      • 032 little-endian
    • 04 Do I need a system that doesn't require aligned addresses?
      • 040 don't care
      • 041 yes
\ I assume that a system that aligns addresses *will* align floats 
\ and vice versa
\ So the question need only be asked once.  Is this right?
\ I'm not sure. Ouch.
    • 05 Do I need the data in dataspace to be contiguous? 
\large -ALLOT etc
\yes, -ALLOT bigger than allotted since last time code was compiled.
      • 050 no
      • 051 yes
      • 052 sometimes more than an ANS Forth data region
    • 06 Do I need a nonHarvard architecture with cell-size calls containing xt's? 
\ Is hotpatching the only advantage to a nonharvard architecture?
\ Some people lay down code as data and then execute it. Some can't.  
\ CREATE ] DOES> R> 2>R ;
      • 060 no
      • 061 yes
    • 07 How big is my biggest counted string?
      • 070 no counted strings
      • 071 <256 chars
      • 072 <65536 chars
    • 08 Do I need to access Forth83-style headers?
\ header magic, printing wordnames etc
\ Decompiling, whatever. ANS Forth doesn't provide anything like that.
\ Forth83 gives standard words to do it.
      • 080 no
      • 081 yes
      • 082 I need another kind of header
    • 09 I/O buffers
      • 090 How big a keybord input buffer do I need?
\ words like S" that don't do REFILL can be limited by buffer, long paths or urls etc
        • 0900 don't care
        • 0901 <80 chars
        • 0902 <128 chars
        • 0903 <1024 chars
        • 0904 <32767 chars
      • 091 How big is the pictured numeric input buffer?
\ 2*cellsize+2 for biggest double number, but with HOLD could be bigger
\ If you don't need binary, you'll need less. Or if you don't do doubles or floating.
\ Some people use that buffer with special characters, no telling how big they want.
        • 0920 none, no character numeric output or <#
        • 0921 < 2*cellsize+3 chars
        • 0922 <129 chars
        • 0923 <1025 chars
        • 0924 < 32768 chars
    • 0A Other buffers
      • 0A0 How big an area do I need at PAD ?
        • 0A00 none
        • 0A01 <84 chars
        • 0A02 <1024 chars
        • 0A03 <32K
        • 0A04 other
      • 0A1 Do I need blocks?
        • 0A10 no
        • 0A11 1 block buffer
        • 0A12 I need 2 valid block buffers at once
  • 1 Arithmetic
    • 10 Do I care about the form of numbers? 
\ bitwise operations on negative numbers
      • 100 don't care
      • 101 2's-complement
      • 102 1's-complement
      • 103 sign-magnitude
      • 104 other
    • 11 Do I care about symmetric or floored division?
\ matters for division with negative numbers
      • 110 no
      • 111 symmetric
      • 112 floored
  • 2 Return stack
    • 20 Do I want to manipulate the return stack with Michael Gassanenko's words?
\ might be emulated even when there's a dummy return stack for >R to use.
      • 200 no
      • 201 yes
    • 21 Do I want to pass parameters on the return stack into a DO LOOP ?
\ usually can't because loop-sys parameters are in the way.
      • 210 no
      • 211 yes
    • 22 Do I use POSTPONE ; 
\ are there other things that tend to fail when this does?
\ Two styles of compiler. defer GO ' interpret is GO ' compile is GO
\ change deferred or variable to stop compiling.
\ : interpret begin interpret-stuff again ; : ] begin compile-stuff again ;
\ execute EXIT to stop compiling.
      • 220 no
      • 222 yes
    • 23 Do I need the compiler to be one level of nesting below the interpreter?
\ #22 | #23. You can code so it doesn't matter or you can depend on one way.
      • 230 no
      • 231 yes
  • 3 Data stack ( which system behaviors interfere with user data stack?)
    • 30 Do I pass data on the data stack through control structures while compiling?
\  ... [ COMPUTE-VALUE DUP ] LITERAL DO DO-STUFF LITERAL ...
      • 300 no
      • 301 yes
    • 31 Do I pass data on the data stack through defining words?
\ 5 : FOO LITERAL ...
      • 310 no
      • 311 yes
  • 4 State-smartness
    • 40 Do I use Anton Ertl's alternative to state-smart words?
      • 400 no
      • 401 yes
    • 41 Do I define state-smart words?
      • 410 no
      • 411 yes
  • 5 Particular words
    • 50 Do I use BEGIN with multiple WHILE ?
      • 500 no
      • 501 yes
    • 51 Do I ever use DOES> when the latest-defined word was not by CREATE ?
      • 510 no
      • 511 yes, and I expect DOES> to work on the last word defined by CREATE
      • 512 yes, and I expect DOES> to work on colon definitions
      • 513 yes, and I expect DOES> to work on colon definitions and :NONAME
    • 52 Do I ever use IMMEDIATE after DOES> or :NONAME ?
      • 520 no
      • 521 yes, and I expect it to act on the latest named definition.
      • 522 yes, and I expect it to act on each DOES> child.
      • 523 yes, and I expect it to have no effect.
    • 53 Do I use NOT ?
      • 530 no
      • 531 yes, and I expect it to behave like 0=
      • 532 yes, and I expect it to behave like INVERT
      • 533 yes, and I expect some other action
  • 6 What true flag do I expect words like 0= to return?
    • 60 don't care
    • 61 low bit set
    • 62 all bits set
  • 7 Do I use floating point?
    • 70 no
    • 71 yes, and I need floats on the data stack.
    • 72 yes, and I don't care where the floats are.
    • 73 yes, and I need a floating stack of 8 items or less.
    • 74 yes, and I need a deep floating point stack.
  • 8 Do I need mass storage?
    • 80 no
    • 81 direct Access by Blocks
    • 82 file System
    • 83 other
  • 9 Do I need graphic character output?
    • 90 no
    • 91 ascii7
    • 92 ISO8
    • 93 standard 16-bit
    • 94 other
  • A Number representation
    • A0 What is the maximum BASE I use?
      • A00 no numeric string output
      • A01 maximum BASE < 37
      • A02 maximum BASE < 63
      • A03 other
    • A1 Do I ever nest <# ?
\ mostly used in debugging, print one number while another is being formatted
      • A10 no
      • A11 yes
      • A12 nested recursively

See ForthImplementation

Kwiki Logo