* regcomp and regexec -- regsub and regerror are elsewhere
* Copyright (c) 1986 by University of Toronto.
* Written by Henry Spencer. Not derived from licensed software.
* Permission is granted to anyone to use this software for any
* purpose on any computer system, and to redistribute it freely,
* subject to the following restrictions:
* 1. The author is not responsible for the consequences of use of
* this software, no matter how awful, even if they arise
* 2. The origin of this software must not be misrepresented, either
* by explicit claim or by omission.
* 3. Altered versions must be plainly marked as such, and must not
* be misrepresented as being the original software.
*** THIS IS AN ALTERED VERSION. It was altered by John Gilmore,
*** hoptoad!gnu, on 27 Dec 1986, to add \n as an alternative to |
*** to assist in implementing egrep.
*** THIS IS AN ALTERED VERSION. It was altered by John Gilmore,
*** hoptoad!gnu, on 27 Dec 1986, to add \< and \> for word-matching
*** as in BSD grep and ex.
*** THIS IS AN ALTERED VERSION. It was altered by John Gilmore,
*** hoptoad!gnu, on 28 Dec 1986, to optimize characters quoted with \.
*** THIS IS AN ALTERED VERSION. It was altered by James A. Woods,
*** ames!jaw, on 19 June 1987, to quash a regcomp() redundancy.
* Beware that some of this code is subtly aware of the way operator
* precedence is structured in regular expressions. Serious changes in
* regular-expression syntax might require a total rethink.
* The "internal use only" fields in regexp.h are present to pass info from
* compile to execute that permits the execute phase to run lots faster on
* simple cases. They are:
* regstart char that must begin a match; '\0' if none obvious
* reganch is the match anchored (at beginning-of-line only)?
* regmust string (pointer into program) that match must include, or NULL
* regmlen length of regmust string
* Regstart and reganch permit very fast decisions on suitable starting points
* for a match, cutting down the work a lot. Regmust permits fast rejection
* of lines that cannot possibly match. The regmust tests are costly enough
* that regcomp() supplies a regmust only if the r.e. contains something
* potentially expensive (at present, the only such thing detected is * or +
* at the start of the r.e., which can involve a lot of backup). Regmlen is
* supplied because the test in regexec() needs it and regcomp() is computing
* Structure for regexp "program". This is essentially a linear encoding
* of a nondeterministic finite-state machine (aka syntax charts or
* "railroad normal form" in parsing technology). Each node is an opcode
* plus a "next" pointer, possibly plus an operand. "Next" pointers of
* all nodes except BRANCH implement concatenation; a "next" pointer with
* a BRANCH on both ends of it is connecting two alternatives. (Here we
* have one of the subtle syntax dependencies: an individual BRANCH (as
* opposed to a collection of them) is never concatenated with anything
* because of operator precedence.) The operand of some types of node is
* a literal string; for others, it is a node leading into a sub-FSM. In
* particular, the operand of a BRANCH node is the first node of the branch.
* (NB this is *not* a tree structure: the tail of the branch connects
* to the thing following the set of BRANCHes.) The opcodes are:
/* definition number opnd? meaning */
#define END 0 /* no End of program. */
#define BOL 1 /* no Match "" at beginning of line. */
#define EOL 2 /* no Match "" at end of line. */
#define ANY 3 /* no Match any one character. */
#define ANYOF 4 /* str Match any character in this string. */
#define ANYBUT 5 /* str Match any character not in this string. */
#define BRANCH 6 /* node Match this alternative, or the next... */
#define BACK 7 /* no Match "", "next" ptr points backward. */
#define EXACTLY 8 /* str Match this string. */
#define NOTHING 9 /* no Match empty string. */
#define STAR 10 /* node Match this (simple) thing 0 or more times. */
#define PLUS 11 /* node Match this (simple) thing 1 or more times. */
#define WORDA 12 /* no Match "" at wordchar, where prev is nonword */
#define WORDZ 13 /* no Match "" at nonwordchar, where prev is word */
#define OPEN 20 /* no Mark this point in input as start of #n. */
/* OPEN+1 is number 1, etc. */
#define CLOSE 30 /* no Analogous to OPEN. */
* BRANCH The set of branches constituting a single choice are hooked
* together with their "next" pointers, since precedence prevents
* anything being concatenated to any individual branch. The
* "next" pointer of the last BRANCH in a choice points to the
* thing following the whole choice. This is also where the
* final "next" pointer of each individual branch points; each
* branch starts with the operand node of a BRANCH node.
* BACK Normal "next" pointers all implicitly point forward; BACK
* exists to make loop structures possible.
* STAR,PLUS '?', and complex '*' and '+', are implemented as circular
* BRANCH structures using BACK. Simple cases (one character
* per match) are implemented with STAR and PLUS for speed
* and to minimize recursive plunges.
* OPEN,CLOSE ...are numbered at compile time.
* A node is one char of opcode followed by two chars of "next" pointer.
* "Next" pointers are stored as two 8-bit pieces, high order first. The
* value is a positive offset from the opcode of the node containing it.
* An operand, if any, simply follows the node. (Note that much of the
* code generation knows about this implicit relationship.)
* Using two bytes for the "next" pointer is vast overkill for most things,
* but allows patterns to get big without disasters.
#define NEXT(p) (((*((p)+1)&0377)<<8) + (*((p)+2)&0377))
#define OPERAND(p) ((p) + 3)
* See regmagic.h for one further detail of program structure.
#define UCHARAT(p) ((int)*(unsigned char *)(p))
#define UCHARAT(p) ((int)*(p)&CHARBITS)
#define FAIL(m) { regerror(m); return(NULL); }
#define ISMULT(c) ((c) == '*' || (c) == '+' || (c) == '?')
* Flags to be passed up and down.
#define HASWIDTH 01 /* Known never to match null string. */
#define SIMPLE 02 /* Simple enough to be STAR/PLUS operand. */
#define SPSTART 04 /* Starts with * or +. */
#define WORST 0 /* Worst case. */
* Global work variables for regcomp().
static char *regparse
; /* Input-scan pointer. */
static int regnpar
; /* () count. */
static char *regcode
; /* Code-emit pointer; ®dummy = don't. */
static long regsize
; /* Code size. */
* Forward declarations for regcomp()'s friends.
STATIC
char *regbranch();
- regcomp - compile a regular expression into internal code
* We can't allocate space until we know how big the compiled form will be,
* but we can't compile it (and thus know how big it is) until we've got a
* place to put the code. So we cheat: we compile it twice, once with code
* generation turned off and size counting turned on, and once "for real".
* This also means that we don't allocate space until we are sure that the
* thing really will compile successfully, and we never have to move the
* code and thus invalidate pointers into it. (Note that it has to be in
* one piece because free() must be able to free it all.)
* Beware that the optimization-preparation code in here knows about some
* of the structure of the compiled regexp.
/* First pass: determine size, legality. */
if (exp
[0] == '.' && exp
[1] == '*') exp
+= 2; /* aid grep */
if (reg(0, &flags
) == NULL
)
/* Small enough for pointer-storage convention? */
if (regsize
>= 32767L) /* Probably could be 65535L. */
r
= (regexp
*)malloc(sizeof(regexp
) + (unsigned)regsize
);
/* Second pass: emit code. */
if (reg(0, &flags
) == NULL
)
/* Dig out information for optimizations. */
r
->regstart
= '\0'; /* Worst-case defaults. */
scan
= r
->program
+1; /* First BRANCH. */
if (OP(regnext(scan
)) == END
) { /* Only one top-level choice. */
/* Starting-point info. */
r
->regstart
= *OPERAND(scan
);
else if (OP(scan
) == BOL
)
* If there's something expensive in the r.e., find the
* longest literal string that must appear and make it the
* regmust. Resolve ties in favor of later strings, since
* the regstart check works with the beginning of the r.e.
* and avoiding duplication strengthens checking. Not a
* strong reason, but sufficient in the absence of others.
for (; scan
!= NULL
; scan
= regnext(scan
))
if (OP(scan
) == EXACTLY
&& strlen(OPERAND(scan
)) >= len
) {
len
= strlen(OPERAND(scan
));
- reg - regular expression, i.e. main body or parenthesized thing
* Caller must absorb opening parenthesis.
* Combining parenthesis handling with the base level of regular expression
* is a trifle forced, but the need to tie the tails of the branches to what
* follows makes it hard to avoid.
int paren
; /* Parenthesized? */
*flagp
= HASWIDTH
; /* Tentatively. */
/* Make an OPEN node, if parenthesized. */
ret
= regnode(OPEN
+parno
);
/* Pick up the branches, linking them together. */
regtail(ret
, br
); /* OPEN -> first. */
while (*regparse
== '|' || *regparse
== '\n') {
regtail(ret
, br
); /* BRANCH -> BRANCH. */
/* Make a closing node, and hook it on the end. */
ender
= regnode((paren
) ? CLOSE
+parno
: END
);
/* Hook the tails of the branches to the closing node. */
for (br
= ret
; br
!= NULL
; br
= regnext(br
))
/* Check for proper termination. */
if (paren
&& *regparse
++ != ')') {
} else if (!paren
&& *regparse
!= '\0') {
FAIL("junk on end"); /* "Can't happen". */
- regbranch - one alternative of an | operator
* Implements the concatenation operator.
*flagp
= WORST
; /* Tentatively. */
while (*regparse
!= '\0' && *regparse
!= ')' &&
*regparse
!= '\n' && *regparse
!= '|') {
latest
= regpiece(&flags
);
*flagp
|= flags
&HASWIDTH
;
if (chain
== NULL
) /* First piece. */
if (chain
== NULL
) /* Loop ran zero times. */
- regpiece - something followed by possible [*+?]
* Note that the branching code sequences used for ? and the general cases
* of * and + are somewhat optimized: they use the same NOTHING node as
* both the endmarker for their branch list and the body of the last branch.
* It might seem that this node could be dispensed with entirely, but the
* endmarker role is not redundant.
if (!(flags
&HASWIDTH
) && op
!= '?')
FAIL("*+ operand could be empty");
*flagp
= (op
!= '+') ? (WORST
|SPSTART
) : (WORST
|HASWIDTH
);
if (op
== '*' && (flags
&SIMPLE
))
/* Emit x* as (x&|), where & means "self". */
reginsert(BRANCH
, ret
); /* Either x */
regoptail(ret
, regnode(BACK
)); /* and loop */
regoptail(ret
, ret
); /* back */
regtail(ret
, regnode(BRANCH
)); /* or */
regtail(ret
, regnode(NOTHING
)); /* null. */
} else if (op
== '+' && (flags
&SIMPLE
))
/* Emit x+ as x(&|), where & means "self". */
next
= regnode(BRANCH
); /* Either */
regtail(regnode(BACK
), ret
); /* loop back */
regtail(next
, regnode(BRANCH
)); /* or */
regtail(ret
, regnode(NOTHING
)); /* null. */
reginsert(BRANCH
, ret
); /* Either x */
regtail(ret
, regnode(BRANCH
)); /* or */
next
= regnode(NOTHING
); /* null. */
- regatom - the lowest level
* Optimization: gobbles an entire sequence of ordinary characters so that
* it can turn them into a single node, which is smaller to store and
* faster to run. Backslashed characters are exceptions, each becoming a
* separate node; the code is simpler that way and it's not worth fixing.
*flagp
= WORST
; /* Tentatively. */
/* FIXME: these chars only have meaning at beg/end of pat? */
*flagp
|= HASWIDTH
|SIMPLE
;
if (*regparse
== '^') { /* Complement of range. */
if (*regparse
== ']' || *regparse
== '-')
while (*regparse
!= '\0' && *regparse
!= ']') {
if (*regparse
== ']' || *regparse
== '\0')
class = UCHARAT(regparse
-2)+1;
classend
= UCHARAT(regparse
);
FAIL("invalid [] range");
for (; class <= classend
; class++)
*flagp
|= HASWIDTH
|SIMPLE
;
*flagp
|= flags
&(HASWIDTH
|SPSTART
);
FAIL("internal urp"); /* Supposed to be caught earlier. */
FAIL("?+* follows nothing");
/* FIXME: Someday handle \1, \2, ... */
/* Handle general quoted chars in exact-match routine */
* Encode a string of characters to be matched exactly.
* This is a bit tricky due to quoted chars and due to
* '*', '+', and '?' taking the SINGLE char previous
* On entry, the char at regparse[-1] is going to go
* into the string, no matter what it is. (It could be
* following a \ if we are entered from the '\' case.)
* Basic idea is to pick up a good char in ch and
* examine the next char. If it's *+? then we twiddle.
* If it's \ then we frozzle. If it's other magic char
* we push ch and terminate the string. If none of the
* above, we push ch on the string and go around again.
* regprev is used to remember where "the current char"
* starts in the string, if due to a *+? we need to back
* up and put the current char in a separate, 1-char, string.
* When regprev is NULL, ch is the only char in the
* string; this is used in *+? handling, and in setting
* flags |= SIMPLE at the end.
regparse
--; /* Look at cur char */
for ( regprev
= 0 ; ; ) {
ch
= *regparse
++; /* Get current char */
switch (*regparse
) { /* look at next one */
regc(ch
); /* Add cur to string */
case '.': case '[': case '(':
case ')': case '|': case '\n':
/* FIXME, $ and ^ should not always be magic */
regc(ch
); /* dump cur char */
goto done
; /* and we are done */
case '?': case '+': case '*':
if (!regprev
) /* If just ch in str, */
/* End mult-char string one early */
regparse
= regprev
; /* Back up parse */
regc(ch
); /* Cur char OK */
switch (regparse
[1]){ /* Look after \ */
/* FIXME: Someday handle \1, \2, ... */
goto done
; /* Not quoted */
/* Backup point is \, scan * point is after it. */
continue; /* NOT break; */
regprev
= regparse
; /* Set backup point */
if (!regprev
) /* One char? */
static char * /* Location. */
*ptr
++ = '\0'; /* Null "next" pointer. */
- regc - emit (if appropriate) a byte of code
if (regcode
!= ®dummy
)
- reginsert - insert an operator in front of already-emitted operand
* Means relocating the operand.
if (regcode
== ®dummy
) {
place
= opnd
; /* Op node, where operand used to be. */
- regtail - set the next-pointer at the end of a node chain
*(scan
+1) = (offset
>>8)&0377;
- regoptail - regtail on operand of first argument; nop if operandless
/* "Operandless" and "op != BRANCH" are synonymous in practice. */
if (p
== NULL
|| p
== ®dummy
|| OP(p
) != BRANCH
)
regtail(OPERAND(p
), val
);
* Global work variables for regexec().
static char *reginput
; /* String-input pointer. */
static char *regbol
; /* Beginning of input, for ^ check. */
static char **regstartp
; /* Pointer to startp array. */
static char **regendp
; /* Ditto for endp. */
- regexec - match a regexp against a string
register const regexp
*prog
;
register const char *string
;
if (prog
== NULL
|| string
== NULL
) {
regerror("NULL parameter");
/* Check validity of program. */
if (UCHARAT(prog
->program
) != MAGIC
) {
regerror("corrupted program");
/* If there is a "must appear" string, look for it. */
if (prog
->regmust
!= NULL
) {
while ((s
= strchr(s
, prog
->regmust
[0])) != NULL
) {
if (strncmp(s
, prog
->regmust
, prog
->regmlen
) == 0)
if (s
== NULL
) /* Not present. */
/* Mark beginning of line for ^ . */
/* Simplest case: anchored match need be tried only once. */
return(regtry(prog
, string
));
/* Messy cases: unanchored match. */
if (prog
->regstart
!= '\0')
/* We know what char it must start with. */
while ((s
= strchr(s
, prog
->regstart
)) != NULL
) {
/* We don't -- general case. */
- regtry - try match at specific point
static int /* 0 failure, 1 success */
regstartp
= prog
->startp
;
for (i
= NSUBEXP
; i
> 0; i
--) {
if (regmatch(prog
->program
+ 1)) {
prog
->startp
[0] = string
;
prog
->endp
[0] = reginput
;
- regmatch - main matching routine
* Conceptually the strategy is simple: check to see whether the current
* node matches, call self recursively to see whether the rest matches,
* and then act accordingly. In practice we make some effort to avoid
* recursion, in particular by going through "ordinary" nodes (that don't
* need to know whether the rest of the match failed) by a loop instead of
static int /* 0 failure, 1 success */
register char *scan
; /* Current node. */
char *next
; /* Next node. */
if (scan
!= NULL
&& regnarrate
)
fprintf(stderr
, "%s(\n", regprop(scan
));
fprintf(stderr
, "%s...\n", regprop(scan
));
/* Must be looking at a letter, digit, or _ */
if ((!isalnum(*reginput
)) && *reginput
!= '_')
/* Prev must be BOL or nonword */
(isalnum(reginput
[-1]) || reginput
[-1] == '_'))
/* Must be looking at non letter, digit, or _ */
if (isalnum(*reginput
) || *reginput
== '_')
/* We don't care what the previous char was */
/* Inline the first character, for speed. */
if (len
> 1 && strncmp(opnd
, reginput
, len
) != 0)
if (*reginput
== '\0' || strchr(OPERAND(scan
), *reginput
) == NULL
)
if (*reginput
== '\0' || strchr(OPERAND(scan
), *reginput
) != NULL
)
* Don't set startp if some later
* invocation of the same parentheses
if (regstartp
[no
] == NULL
)
* Don't set endp if some later
* invocation of the same parentheses
if (OP(next
) != BRANCH
) /* No choice. */
next
= OPERAND(scan
); /* Avoid recursion. */
if (regmatch(OPERAND(scan
)))
} while (scan
!= NULL
&& OP(scan
) == BRANCH
);
* Lookahead to avoid useless match attempts
* when we know what character comes next.
min
= (OP(scan
) == STAR
) ? 0 : 1;
no
= regrepeat(OPERAND(scan
));
/* If it could work, try it. */
if (nextch
== '\0' || *reginput
== nextch
)
/* Couldn't or didn't -- back up. */
return(1); /* Success! */
regerror("memory corruption");
* We get here only if there's trouble -- normally "case END" is
regerror("corrupted pointers");
- regrepeat - repeatedly match something simple, report how many
while (*scan
!= '\0' && strchr(opnd
, *scan
) != NULL
) {
while (*scan
!= '\0' && strchr(opnd
, *scan
) == NULL
) {
default: /* Oh dear. Called inappropriately. */
regerror("internal foulup");
count
= 0; /* Best compromise. */
- regnext - dig the "next" pointer out of a node
- regdump - dump a regexp onto stdout in vaguely comprehensible form
register char op
= EXACTLY
; /* Arbitrary non-END op. */
while (op
!= END
) { /* While that wasn't END last time... */
printf("%2d%s", s
-r
->program
, regprop(s
)); /* Where, what. */
if (next
== NULL
) /* Next ptr. */
printf("(%d)", (s
-r
->program
)+(next
-s
));
if (op
== ANYOF
|| op
== ANYBUT
|| op
== EXACTLY
) {
/* Literal string, where present. */
/* Header fields of interest. */
printf("start `%c' ", r
->regstart
);
printf("must have \"%s\"", r
->regmust
);
- regprop - printable representation of opcode
sprintf(buf
+strlen(buf
), "OPEN%d", OP(op
)-OPEN
);
sprintf(buf
+strlen(buf
), "CLOSE%d", OP(op
)-CLOSE
);
regerror("corrupted opcode");
* The following is provided for those people who do not have strcspn() in
* their C libraries. They should get off their butts and do something
* about it; at least one public-domain implementation of those (highly
* useful) string routines has been published on Usenet.
* strcspn - find length of initial segment of s1 consisting entirely
* of characters not from s2
for (scan1
= s1
; *scan1
!= '\0'; scan1
++) {
for (scan2
= s2
; *scan2
!= '\0';) /* ++ moved down. */