µEforth (micro-Eforth) simplifies EForth even futher, by building just enough of the core of the system in C to allow the rest to be be built in proper Forth.
A handful of "tricky" words that involve internal loops or many steps are built in their own functions:
FIND ( a n -- xt | 0 ) PARSE ( ch -- a n ) S>NUMBER? ( a n -- n f | 0 ) CREATE ( "name" -- ) EVALUATE1 ( -- )
This includes EVALUATE1 which parses a single word and
interprets or compiles it (reusing PARSE,
FIND, and S>NUMBER?).
See core.h.
A few global variables connect parsing and compilation state between
C and Forth (by way of a memory region accessed via 'SYS):
'TIB --- Pointer to the Translation Input Buffer #TIB --- Length of the Translation Input Buffer >IN --- Number of characters consumed from TIB BASE --- Numeric base for printing and parsing STATE --- State of compiling, -1 for compiling, 0 for interpreting CURRENT --- Pointer to pointer to last word of current vocabulary CONTEXT --- Pointer to pointer to last word of context vocabulary 'NOTFOUND --- Execution token of a handler to call on word not found
Error handling is routed via a deferred callback in 'NOTFOUND
used when a word is absent from the dictionary.
This is eventually directed to an error routing that prints
a proper error, once I/O and exceptions are available.
X-Macros are then used to build up a small set of core opcodes defined in 1-3 lines each:
0= 0< + U/MOD */MOD AND OR XOR LSHIFT RSHIFT DUP SWAP OVER DROP @ SL@ SW@ C@ ! L! W! C! SP@ SP! RP@ RP! >R R> R@ : ; EXIT EXECUTE BRANCH 0BRANCH DONEXT DOLIT ALITERAL CELL DOES> IMMEDIATE 'SYS
NOTE: Later to reduce the use of the RAM heap and improve performance, additional non-essential extra opcodes were added in place of high-level words.
See tier0_opcodes.h.
I/O and access to systems outside Forth are connected via a few per platform words. Typically this set of words should be minimal, while still allowing relevant parts of the host system to be available to Forth.