Commit | Line | Data |
a687059c |
1 | /* $Header: regcomp.h,v 3.0 89/10/18 15:22:39 lwall Locked $ |
2 | * |
3 | * $Log: regcomp.h,v $ |
4 | * Revision 3.0 89/10/18 15:22:39 lwall |
5 | * 3.0 baseline |
6 | * |
7 | */ |
8 | |
9 | /* |
10 | * The "internal use only" fields in regexp.h are present to pass info from |
11 | * compile to execute that permits the execute phase to run lots faster on |
12 | * simple cases. They are: |
13 | * |
14 | * regstart str that must begin a match; Nullch if none obvious |
15 | * reganch is the match anchored (at beginning-of-line only)? |
16 | * regmust string (pointer into program) that match must include, or NULL |
17 | * [regmust changed to STR* for bminstr()--law] |
18 | * regmlen length of regmust string |
19 | * [regmlen not used currently] |
20 | * |
21 | * Regstart and reganch permit very fast decisions on suitable starting points |
22 | * for a match, cutting down the work a lot. Regmust permits fast rejection |
23 | * of lines that cannot possibly match. The regmust tests are costly enough |
24 | * that regcomp() supplies a regmust only if the r.e. contains something |
25 | * potentially expensive (at present, the only such thing detected is * or + |
26 | * at the start of the r.e., which can involve a lot of backup). Regmlen is |
27 | * supplied because the test in regexec() needs it and regcomp() is computing |
28 | * it anyway. |
29 | * [regmust is now supplied always. The tests that use regmust have a |
30 | * heuristic that disables the test if it usually matches.] |
31 | * |
32 | * [In fact, we now use regmust in many cases to locate where the search |
33 | * starts in the string, so if regback is >= 0, the regmust search is never |
34 | * wasted effort. The regback variable says how many characters back from |
35 | * where regmust matched is the earliest possible start of the match. |
36 | * For instance, /[a-z].foo/ has a regmust of 'foo' and a regback of 2.] |
37 | */ |
38 | |
39 | /* |
40 | * Structure for regexp "program". This is essentially a linear encoding |
41 | * of a nondeterministic finite-state machine (aka syntax charts or |
42 | * "railroad normal form" in parsing technology). Each node is an opcode |
43 | * plus a "next" pointer, possibly plus an operand. "Next" pointers of |
44 | * all nodes except BRANCH implement concatenation; a "next" pointer with |
45 | * a BRANCH on both ends of it is connecting two alternatives. (Here we |
46 | * have one of the subtle syntax dependencies: an individual BRANCH (as |
47 | * opposed to a collection of them) is never concatenated with anything |
48 | * because of operator precedence.) The operand of some types of node is |
49 | * a literal string; for others, it is a node leading into a sub-FSM. In |
50 | * particular, the operand of a BRANCH node is the first node of the branch. |
51 | * (NB this is *not* a tree structure: the tail of the branch connects |
52 | * to the thing following the set of BRANCHes.) The opcodes are: |
53 | */ |
54 | |
55 | /* definition number opnd? meaning */ |
56 | #define END 0 /* no End of program. */ |
57 | #define BOL 1 /* no Match "" at beginning of line. */ |
58 | #define EOL 2 /* no Match "" at end of line. */ |
59 | #define ANY 3 /* no Match any one character. */ |
60 | #define ANYOF 4 /* str Match any character in this string. */ |
61 | #define ANYBUT 5 /* str Match any character not in this string. */ |
62 | #define BRANCH 6 /* node Match this alternative, or the next... */ |
63 | #define BACK 7 /* no Match "", "next" ptr points backward. */ |
64 | #define EXACTLY 8 /* str Match this string (preceded by length). */ |
65 | #define NOTHING 9 /* no Match empty string. */ |
66 | #define STAR 10 /* node Match this (simple) thing 0 or more times. */ |
67 | #define PLUS 11 /* node Match this (simple) thing 1 or more times. */ |
68 | #define ALNUM 12 /* no Match any alphanumeric character */ |
69 | #define NALNUM 13 /* no Match any non-alphanumeric character */ |
70 | #define BOUND 14 /* no Match "" at any word boundary */ |
71 | #define NBOUND 15 /* no Match "" at any word non-boundary */ |
72 | #define SPACE 16 /* no Match any whitespace character */ |
73 | #define NSPACE 17 /* no Match any non-whitespace character */ |
74 | #define DIGIT 18 /* no Match any numeric character */ |
75 | #define NDIGIT 19 /* no Match any non-numeric character */ |
76 | #define REF 20 /* no Match some already matched string */ |
77 | #define OPEN 30 /* no Mark this point in input as start of #n. */ |
78 | /* OPEN+1 is number 1, etc. */ |
79 | #define CLOSE 40 /* no Analogous to OPEN. */ |
80 | /* CLOSE must be last one! see regmust finder */ |
81 | |
82 | /* |
83 | * Opcode notes: |
84 | * |
85 | * BRANCH The set of branches constituting a single choice are hooked |
86 | * together with their "next" pointers, since precedence prevents |
87 | * anything being concatenated to any individual branch. The |
88 | * "next" pointer of the last BRANCH in a choice points to the |
89 | * thing following the whole choice. This is also where the |
90 | * final "next" pointer of each individual branch points; each |
91 | * branch starts with the operand node of a BRANCH node. |
92 | * |
93 | * BACK Normal "next" pointers all implicitly point forward; BACK |
94 | * exists to make loop structures possible. |
95 | * |
96 | * STAR,PLUS '?', and complex '*' and '+', are implemented as circular |
97 | * BRANCH structures using BACK. Simple cases (one character |
98 | * per match) are implemented with STAR and PLUS for speed |
99 | * and to minimize recursive plunges. |
100 | * |
101 | * OPEN,CLOSE ...are numbered at compile time. |
102 | */ |
103 | |
104 | /* The following have no fixed length. */ |
105 | #ifndef DOINIT |
106 | extern char varies[]; |
107 | #else |
108 | char varies[] = {BRANCH,BACK,STAR,PLUS, |
109 | REF+1,REF+2,REF+3,REF+4,REF+5,REF+6,REF+7,REF+8,REF+9,0}; |
110 | #endif |
111 | |
112 | /* The following always have a length of 1. */ |
113 | #ifndef DOINIT |
114 | extern char simple[]; |
115 | #else |
116 | char simple[] = {ANY,ANYOF,ANYBUT,ALNUM,NALNUM,SPACE,NSPACE,DIGIT,NDIGIT,0}; |
117 | #endif |
118 | |
119 | EXT char regdummy; |
120 | |
121 | /* |
122 | * A node is one char of opcode followed by two chars of "next" pointer. |
123 | * "Next" pointers are stored as two 8-bit pieces, high order first. The |
124 | * value is a positive offset from the opcode of the node containing it. |
125 | * An operand, if any, simply follows the node. (Note that much of the |
126 | * code generation knows about this implicit relationship.) |
127 | * |
128 | * Using two bytes for the "next" pointer is vast overkill for most things, |
129 | * but allows patterns to get big without disasters. |
130 | * |
131 | * [If REGALIGN is defined, the "next" pointer is always aligned on an even |
132 | * boundary, and reads the offset directly as a short. Also, there is no |
133 | * special test to reverse the sign of BACK pointers since the offset is |
134 | * stored negative.] |
135 | */ |
136 | |
137 | #ifndef gould |
138 | #ifndef cray |
139 | #define REGALIGN |
140 | #endif |
141 | #endif |
142 | |
143 | #define OP(p) (*(p)) |
144 | |
145 | #ifndef lint |
146 | #ifdef REGALIGN |
147 | #define NEXT(p) (*(short*)(p+1)) |
148 | #else |
149 | #define NEXT(p) (((*((p)+1)&0377)<<8) + (*((p)+2)&0377)) |
150 | #endif |
151 | #else /* lint */ |
152 | #define NEXT(p) 0 |
153 | #endif /* lint */ |
154 | |
155 | #define OPERAND(p) ((p) + 3) |
156 | |
157 | #ifdef REGALIGN |
158 | #define NEXTOPER(p) ((p) + 4) |
159 | #else |
160 | #define NEXTOPER(p) ((p) + 3) |
161 | #endif |
162 | |
163 | #define MAGIC 0234 |
164 | |
165 | /* |
166 | * Utility definitions. |
167 | */ |
168 | #ifndef lint |
169 | #ifndef CHARBITS |
170 | #define UCHARAT(p) ((int)*(unsigned char *)(p)) |
171 | #else |
172 | #define UCHARAT(p) ((int)*(p)&CHARBITS) |
173 | #endif |
174 | #else /* lint */ |
175 | #define UCHARAT(p) regdummy |
176 | #endif /* lint */ |
177 | |
178 | #define FAIL(m) fatal("/%s/: %s",regprecomp,m) |
179 | |
180 | char *regnext(); |
181 | #ifdef DEBUGGING |
182 | void regdump(); |
183 | char *regprop(); |
184 | #endif |
185 | |