This is a description of Gazelle's bytecode file format.
The file format itself is "Bitcode," a generic bytecode format
developed as part of the LLVM project. It is documented here:
http://llvm.cs.uiuc.edu/docs/BitCodeFormat.html
This file describes the meaning of the information that Gazelle
puts into the bytecode.
Bitcode file, application magic number = "GH"
All offsets are 0-based unless otherwise specified.
STRINGS -- all strings are put into the STRINGS block, and referenced
by offset in other parts of the file
[STRING, <ascii array of chars>]
[STRING, <ascii array of chars>]
...
INTFAS -- the list of all IntFAs (lexing DFAs)
INTFA -- a single DFA
[INTFA_STATE, # of transitions] <-- first state is assumed to be start state
[INTFA_STATE, # of transitions]
[INTFA_FINAL_STATE, # of transitions, terminal_name_int]
...
(states are identified by their offset in this list)
[INTFA_TRANSITION, edge val, dest state offset]
[INTFA_TRANSITION_RANGE, edge val low, edge val high, dest state offset]
...
(transitions are identified by their offset in this list)
GLAS
GLA
[GLA_STATE, intfa #, # of transitions]
[GLA_FINAL_STATE, 1-based rtn alt offset (0 is return from final)]
...
[GLA_TRANSITION, 1-based terminal_name_int (0 is EOF), dest GLA state offset]
...
RTNS -- a list of all the RTNs
RTN - a single RTN
[RTN_INFO, nonterm_name_int, # of slots]
[RTN_STATE_WITH_INTFA, # of transitions, is final, intfa#]
[RTN_STATE_WITH_GLA, # of transitions, is final, gla#]
[RTN_TRIVIAL_STATE, # of transitions, is final]
...
-- slotnums are 0-based in the runtime, but 1-based in the bytecode and in the
-- compiler. We use 0 in the bytecode to represent "no slotnum" (-1 elsewhere),
-- as used by subparsers.
[RTN_TRANSITION_TERMINAL, terminal_name_int, dest RTN state offset, slotname_int, slotnum+1]
[RTN_TRANSITION_NONTERM, rtn_offset, dest RTN state offset, slotname_int, slotnum+1]
...
