/* dfa.c - deterministic extended regexp routines for GNU
Copyright (C) 1988 Free Software Foundation, Inc.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2, or (at your option)
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
/* Written June, 1988 by Mike Haertel
Modified July, 1988 by Arthur David Olson to assist BMG speedups */
extern char *calloc(), *malloc(), *realloc();
#if defined(HAVE_STRING_H) || defined(STDC_HEADERS)
#define isgraph(C) (isprint(C) && !isspace(C))
#define ISALPHA(C) (isascii(C) && isalpha(C))
#define ISUPPER(C) (isascii(C) && isupper(C))
#define ISLOWER(C) (isascii(C) && islower(C))
#define ISDIGIT(C) (isascii(C) && isdigit(C))
#define ISXDIGIT(C) (isascii(C) && isxdigit(C))
#define ISSPACE(C) (isascii(C) && isspace(C))
#define ISPUNCT(C) (isascii(C) && ispunct(C))
#define ISALNUM(C) (isascii(C) && isalnum(C))
#define ISPRINT(C) (isascii(C) && isprint(C))
#define ISGRAPH(C) (isascii(C) && isgraph(C))
#define ISCNTRL(C) (isascii(C) && iscntrl(C))
#define ISALPHA(C) isalpha(C)
#define ISUPPER(C) isupper(C)
#define ISLOWER(C) islower(C)
#define ISDIGIT(C) isdigit(C)
#define ISXDIGIT(C) isxdigit(C)
#define ISSPACE(C) isspace(C)
#define ISPUNCT(C) ispunct(C)
#define ISALNUM(C) isalnum(C)
#define ISPRINT(C) isprint(C)
#define ISGRAPH(C) isgraph(C)
#define ISCNTRL(C) iscntrl(C)
dfaerror("Memory exhausted");
dfaerror("Memory exhausted");
dfaerror("Memory exhausted");
#define CALLOC(p, t, n) ((p) = (t *) xcalloc((n), sizeof (t)))
#define MALLOC(p, t, n) ((p) = (t *) xmalloc((n) * sizeof (t)))
#define REALLOC(p, t, n) ((p) = (t *) xrealloc((ptr_t) (p), (n) * sizeof (t)))
/* Reallocate an array of type t if nalloc is too small for index. */
#define REALLOC_IF_NECESSARY(p, t, nalloc, index) \
if ((index) >= (nalloc)) \
while ((index) >= (nalloc)) \
fprintf(stderr
, "%c", t
);
case EMPTY
: s
= "EMPTY"; break;
case BACKREF
: s
= "BACKREF"; break;
case BEGLINE
: s
= "BEGLINE"; break;
case ENDLINE
: s
= "ENDLINE"; break;
case BEGWORD
: s
= "BEGWORD"; break;
case ENDWORD
: s
= "ENDWORD"; break;
case LIMWORD
: s
= "LIMWORD"; break;
case NOTLIMWORD
: s
= "NOTLIMWORD"; break;
case QMARK
: s
= "QMARK"; break;
case STAR
: s
= "STAR"; break;
case PLUS
: s
= "PLUS"; break;
case CAT
: s
= "CAT"; break;
case OR
: s
= "OR"; break;
case ORTOP
: s
= "ORTOP"; break;
case LPAREN
: s
= "LPAREN"; break;
case RPAREN
: s
= "RPAREN"; break;
default: s
= "CSET"; break;
fprintf(stderr
, "%s", s
);
/* Stuff pertaining to charclasses. */
return c
[b
/ INTBITS
] & 1 << b
% INTBITS
;
c
[b
/ INTBITS
] |= 1 << b
% INTBITS
;
c
[b
/ INTBITS
] &= ~(1 << b
% INTBITS
);
for (i
= 0; i
< CHARCLASS_INTS
; ++i
)
for (i
= 0; i
< CHARCLASS_INTS
; ++i
)
for (i
= 0; i
< CHARCLASS_INTS
; ++i
)
for (i
= 0; i
< CHARCLASS_INTS
; ++i
)
/* A pointer to the current dfa is kept here during parsing. */
/* Find the index of charclass s in dfa->charclasses, or allocate a new charclass. */
for (i
= 0; i
< dfa
->cindex
; ++i
)
if (equal(s
, dfa
->charclasses
[i
]))
REALLOC_IF_NECESSARY(dfa
->charclasses
, charclass
, dfa
->calloc
, dfa
->cindex
);
copyset(s
, dfa
->charclasses
[i
]);
/* Syntax bits controlling the behavior of the lexical analyzer. */
static int syntax_bits
, syntax_bits_set
;
/* Flag for case-folding letters into sets. */
/* Entry point to set syntax options. */
/* Lexical analyzer. All the dross that deals with the obnoxious
GNU Regex syntax bits is located here. The poor, suffering
reader is referred to the GNU Regex documentation for the
meaning of the @#%!@#%^!@ syntax bits. */
static char *lexstart
; /* Pointer to beginning of input string. */
static char *lexptr
; /* Pointer to next input character. */
static lexleft
; /* Number of characters remaining. */
static token lasttok
; /* Previous token returned; initially END. */
static int laststart
; /* True if we're separated from beginning or (, |
only by zero-width characters. */
static int parens
; /* Count of outstanding left parens. */
static int minrep
, maxrep
; /* Repeat counts for {m,n}. */
/* Note that characters become unsigned here. */
#define FETCH(c, eoferr) \
(c) = (unsigned char) *lexptr++; \
#define FUNC(F, P) static int F(c) int c; { return P(c); }
FUNC(is_xdigit
, ISXDIGIT
)
/* The following list maps the names of the Posix named character classes
to predicate functions that determine whether a given character is in
the class. The leading [ has already been eaten by the lexical analyzer. */
return strncmp(s
, lexptr
, len
) == 0;
int backslash
= 0, invert
;
/* Basic plan: We fetch a character. If it's a backslash,
we set the backslash flag and go through the loop again.
On the plus side, this avoids having a duplicate of the
main switch inside the backslash case. On the minus side,
it means that just about every case begins with
dfaerror("Unfinished \\ escape");
if (syntax_bits
& RE_CONTEXT_INDEP_ANCHORS
return lasttok
= BEGLINE
;
if (syntax_bits
& RE_CONTEXT_INDEP_ANCHORS
|| (syntax_bits
& RE_NO_BK_PARENS
? lexleft
> 0 && *lexptr
== ')'
: lexleft
> 1 && lexptr
[0] == '\\' && lexptr
[1] == ')')
|| (syntax_bits
& RE_NO_BK_VBAR
? lexleft
> 0 && *lexptr
== '|'
: lexleft
> 1 && lexptr
[0] == '\\' && lexptr
[1] == '|')
|| ((syntax_bits
& RE_NEWLINE_ALT
)
&& lexleft
> 0 && *lexptr
== '\n'))
return lasttok
= ENDLINE
;
if (backslash
&& !(syntax_bits
& RE_NO_BK_REFS
))
return lasttok
= BACKREF
;
return lasttok
= BEGWORD
;
return lasttok
= ENDWORD
;
return lasttok
= LIMWORD
;
return lasttok
= NOTLIMWORD
;
if (syntax_bits
& RE_LIMITED_OPS
)
if (backslash
!= ((syntax_bits
& RE_BK_PLUS_QM
) != 0))
if (!(syntax_bits
& RE_CONTEXT_INDEP_OPS
) && laststart
)
if (!(syntax_bits
& RE_CONTEXT_INDEP_OPS
) && laststart
)
if (syntax_bits
& RE_LIMITED_OPS
)
if (backslash
!= ((syntax_bits
& RE_BK_PLUS_QM
) != 0))
if (!(syntax_bits
& RE_CONTEXT_INDEP_OPS
) && laststart
)
if (!(syntax_bits
& RE_INTERVALS
))
if (backslash
!= ((syntax_bits
& RE_NO_BK_BRACES
) == 0))
{M,} - minimum count, maximum is infinity
FETCH(c
, "unfinished repeat count");
FETCH(c
, "unfinished repeat count");
minrep
= 10 * minrep
+ c
- '0';
dfaerror("malformed repeat count");
FETCH(c
, "unfinished repeat count");
maxrep
= 10 * maxrep
+ c
- '0';
if (!(syntax_bits
& RE_NO_BK_BRACES
))
dfaerror("malformed repeat count");
FETCH(c
, "unfinished repeat count");
dfaerror("malformed repeat count");
if (syntax_bits
& RE_LIMITED_OPS
)
if (backslash
!= ((syntax_bits
& RE_NO_BK_VBAR
) == 0))
if (syntax_bits
& RE_LIMITED_OPS
|| !(syntax_bits
& RE_NEWLINE_ALT
))
if (backslash
!= ((syntax_bits
& RE_NO_BK_PARENS
) == 0))
if (backslash
!= ((syntax_bits
& RE_NO_BK_PARENS
) == 0))
if (parens
== 0 && syntax_bits
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
if (!(syntax_bits
& RE_DOT_NEWLINE
))
if (syntax_bits
& RE_DOT_NOT_NULL
)
return lasttok
= CSET
+ charclass_index(ccl
);
for (c2
= 0; c2
< NOTCHAR
; ++c2
)
return lasttok
= CSET
+ charclass_index(ccl
);
FETCH(c
, "Unbalanced [");
FETCH(c
, "Unbalanced [");
/* Nobody ever said this had to be fast. :-)
Note that if we're looking at some other [:...:]
construct, we just treat it as a bunch of ordinary
characters. We can do this because we assume
regex has checked for syntax errors before
if (c
== '[' && (syntax_bits
& RE_CHAR_CLASSES
))
for (c1
= 0; prednames
[c1
].name
; ++c1
)
if (looking_at(prednames
[c1
].name
))
for (c2
= 0; c2
< NOTCHAR
; ++c2
)
if ((*prednames
[c1
].pred
)(c2
))
lexptr
+= strlen(prednames
[c1
].name
);
lexleft
-= strlen(prednames
[c1
].name
);
FETCH(c1
, "Unbalanced [");
if (c
== '\\' && (syntax_bits
& RE_BACKSLASH_ESCAPE_IN_LISTS
))
FETCH(c
, "Unbalanced [");
FETCH(c1
, "Unbalanced [");
FETCH(c2
, "Unbalanced [");
/* In the case [x-], the - is an ordinary hyphen,
which is left in c1, the lookahead character. */
&& (syntax_bits
& RE_BACKSLASH_ESCAPE_IN_LISTS
))
FETCH(c2
, "Unbalanced [");
FETCH(c1
, "Unbalanced [");
if (syntax_bits
& RE_HAT_LISTS_NOT_NEWLINE
)
return lasttok
= CSET
+ charclass_index(ccl
);
if (case_fold
&& ISALPHA(c
))
return lasttok
= CSET
+ charclass_index(ccl
);
/* The above loop should consume at most a backslash
and some other character. */
/* Recursive descent parser for regular expressions. */
static token tok
; /* Lookahead token. */
static depth
; /* Current depth of a hypothetical stack
holding deferred productions. This is
used to determine the depth that will be
required of the real stack later on in
/* Add the given token to the parse tree, maintaining the depth count and
updating the maximum depth if necessary. */
REALLOC_IF_NECESSARY(dfa
->tokens
, token
, dfa
->talloc
, dfa
->tindex
);
dfa
->tokens
[dfa
->tindex
++] = t
;
/* The grammar understood by the parser is as follows.
The parser builds a parse tree in postfix form in an array of tokens. */
if ((tok
>= 0 && tok
< NOTCHAR
) || tok
>= CSET
|| tok
== BACKREF
|| tok
== BEGLINE
|| tok
== ENDLINE
|| tok
== BEGWORD
|| tok
== ENDWORD
|| tok
== LIMWORD
|| tok
== NOTLIMWORD
)
dfaerror("Unbalanced (");
/* Return the number of tokens in the given subexpression. */
switch (dfa
->tokens
[tindex
- 1])
return 1 + nsubtoks(tindex
- 1);
ntoks1
= nsubtoks(tindex
- 1);
return 1 + ntoks1
+ nsubtoks(tindex
- 1 - ntoks1
);
/* Copy the given subexpression to the top of the tree. */
copytoks(tindex
, ntokens
)
for (i
= 0; i
< ntokens
; ++i
)
addtok(dfa
->tokens
[tindex
+ i
]);
while (tok
== QMARK
|| tok
== STAR
|| tok
== PLUS
|| tok
== REPMN
)
ntokens
= nsubtoks(dfa
->tindex
);
tindex
= dfa
->tindex
- ntokens
;
for (i
= 1; i
< minrep
; ++i
)
copytoks(tindex
, ntokens
);
copytoks(tindex
, ntokens
);
while (tok
!= RPAREN
&& tok
!= OR
&& tok
>= 0)
/* Main entry point for the parser. S is a string to be parsed, len is the
length of the string, so s can include NUL characters. D is a pointer to
the struct dfa to parse into. */
dfaerror("No syntax specified");
dfaerror("Unbalanced )");
addtok(END
- d
->nregexps
);
/* Some primitives for operating on sets of positions. */
/* Copy one set to another; the destination must be large enough. */
for (i
= 0; i
< src
->nelem
; ++i
)
dst
->elems
[i
] = src
->elems
[i
];
/* Insert a position in a set. Position sets are maintained in sorted
order according to index. If position already exists in the set with
the same index then their constraints are logically or'd together.
S->elems must point to an array large enough to hold the resulting set. */
for (i
= 0; i
< s
->nelem
&& p
.index
< s
->elems
[i
].index
; ++i
)
if (i
< s
->nelem
&& p
.index
== s
->elems
[i
].index
)
s
->elems
[i
].constraint
|= p
.constraint
;
/* Merge two sets of positions into a third. The result is exactly as if
the positions of both sets were inserted into an initially empty set. */
while (i
< s1
->nelem
&& j
< s2
->nelem
)
if (s1
->elems
[i
].index
> s2
->elems
[j
].index
)
m
->elems
[m
->nelem
++] = s1
->elems
[i
++];
else if (s1
->elems
[i
].index
< s2
->elems
[j
].index
)
m
->elems
[m
->nelem
++] = s2
->elems
[j
++];
m
->elems
[m
->nelem
] = s1
->elems
[i
++];
m
->elems
[m
->nelem
++].constraint
|= s2
->elems
[j
++].constraint
;
m
->elems
[m
->nelem
++] = s1
->elems
[i
++];
m
->elems
[m
->nelem
++] = s2
->elems
[j
++];
/* Delete a position from a set. */
for (i
= 0; i
< s
->nelem
; ++i
)
if (p
.index
== s
->elems
[i
].index
)
for (--s
->nelem
; i
< s
->nelem
; ++i
)
s
->elems
[i
] = s
->elems
[i
+ 1];
/* Find the index of the state corresponding to the given position set with
the given preceding context, or create a new state if there is no such
state. Newline and letter tell whether we got here on a newline or
state_index(d
, s
, newline
, letter
)
newline
= newline
? 1 : 0;
for (i
= 0; i
< s
->nelem
; ++i
)
hash
^= s
->elems
[i
].index
+ s
->elems
[i
].constraint
;
/* Try to find a state that exactly matches the proposed one. */
for (i
= 0; i
< d
->sindex
; ++i
)
if (hash
!= d
->states
[i
].hash
|| s
->nelem
!= d
->states
[i
].elems
.nelem
|| newline
!= d
->states
[i
].newline
|| letter
!= d
->states
[i
].letter
)
for (j
= 0; j
< s
->nelem
; ++j
)
if (s
->elems
[j
].constraint
!= d
->states
[i
].elems
.elems
[j
].constraint
|| s
->elems
[j
].index
!= d
->states
[i
].elems
.elems
[j
].index
)
/* We'll have to create a new state. */
REALLOC_IF_NECESSARY(d
->states
, dfa_state
, d
->salloc
, d
->sindex
);
d
->states
[i
].hash
= hash
;
MALLOC(d
->states
[i
].elems
.elems
, position
, s
->nelem
);
copy(s
, &d
->states
[i
].elems
);
d
->states
[i
].newline
= newline
;
d
->states
[i
].letter
= letter
;
d
->states
[i
].backref
= 0;
d
->states
[i
].constraint
= 0;
d
->states
[i
].first_end
= 0;
for (j
= 0; j
< s
->nelem
; ++j
)
if (d
->tokens
[s
->elems
[j
].index
] < 0)
constraint
= s
->elems
[j
].constraint
;
if (SUCCEEDS_IN_CONTEXT(constraint
, newline
, 0, letter
, 0)
|| SUCCEEDS_IN_CONTEXT(constraint
, newline
, 0, letter
, 1)
|| SUCCEEDS_IN_CONTEXT(constraint
, newline
, 1, letter
, 0)
|| SUCCEEDS_IN_CONTEXT(constraint
, newline
, 1, letter
, 1))
d
->states
[i
].constraint
|= constraint
;
if (! d
->states
[i
].first_end
)
d
->states
[i
].first_end
= d
->tokens
[s
->elems
[j
].index
];
else if (d
->tokens
[s
->elems
[j
].index
] == BACKREF
)
d
->states
[i
].constraint
= NO_CONSTRAINT
;
d
->states
[i
].backref
= 1;
/* Find the epsilon closure of a set of positions. If any position of the set
contains a symbol that matches the empty string in some context, replace
that position with the elements of its follow labeled with an appropriate
constraint. Repeat exhaustively until no funny positions are left.
S->elems must be large enough to hold the result. */
MALLOC(visited
, int, d
->tindex
);
for (i
= 0; i
< d
->tindex
; ++i
)
for (i
= 0; i
< s
->nelem
; ++i
)
if (d
->tokens
[s
->elems
[i
].index
] >= NOTCHAR
&& d
->tokens
[s
->elems
[i
].index
] != BACKREF
&& d
->tokens
[s
->elems
[i
].index
] < CSET
)
p
.constraint
= old
.constraint
;
switch (d
->tokens
[old
.index
])
p
.constraint
&= BEGLINE_CONSTRAINT
;
p
.constraint
&= ENDLINE_CONSTRAINT
;
p
.constraint
&= BEGWORD_CONSTRAINT
;
p
.constraint
&= ENDWORD_CONSTRAINT
;
p
.constraint
&= LIMWORD_CONSTRAINT
;
p
.constraint
&= NOTLIMWORD_CONSTRAINT
;
for (j
= 0; j
< d
->follows
[old
.index
].nelem
; ++j
)
p
.index
= d
->follows
[old
.index
].elems
[j
].index
;
/* Force rescan to start at the beginning. */
/* Perform bottom-up analysis on the parse tree, computing various functions.
Note that at this point, we're pretending constructs like \< are real
characters rather than constraints on what can follow them.
Nullable: A node is nullable if it is at the root of a regexp that can
* EMPTY leaves are nullable.
* No other leaf is nullable.
* A QMARK or STAR node is nullable.
* A PLUS node is nullable if its argument is nullable.
* A CAT node is nullable if both its arguments are nullable.
* An OR node is nullable if either argument is nullable.
Firstpos: The firstpos of a node is the set of positions (nonempty leaves)
that could correspond to the first character of a string matching the
regexp rooted at the given node.
* EMPTY leaves have empty firstpos.
* The firstpos of a nonempty leaf is that leaf itself.
* The firstpos of a QMARK, STAR, or PLUS node is the firstpos of its
* The firstpos of a CAT node is the firstpos of the left argument, union
the firstpos of the right if the left argument is nullable.
* The firstpos of an OR node is the union of firstpos of each argument.
Lastpos: The lastpos of a node is the set of positions that could
correspond to the last character of a string matching the regexp at
* EMPTY leaves have empty lastpos.
* The lastpos of a nonempty leaf is that leaf itself.
* The lastpos of a QMARK, STAR, or PLUS node is the lastpos of its
* The lastpos of a CAT node is the lastpos of its right argument, union
the lastpos of the left if the right argument is nullable.
* The lastpos of an OR node is the union of the lastpos of each argument.
Follow: The follow of a position is the set of positions that could
correspond to the character following a character matching the node in
a string matching the regexp. At this point we consider special symbols
that match the empty string in some context to be just normal characters.
Later, if we find that a special symbol is in a follow set, we will
replace it with the elements of its follow, labeled with an appropriate
* Every node in the firstpos of the argument of a STAR or PLUS node is in
the follow of every node in the lastpos.
* Every node in the firstpos of the second argument of a CAT node is in
the follow of every node in the lastpos of the first argument.
Because of the postfix representation of the parse tree, the depth-first
analysis is conveniently done by a linear scan with the aid of a stack.
Sets are stored as arrays of the elements, obeying a stack-like allocation
scheme; the number of elements in each set deeper in the stack can be
used to determine the address of a particular set's array. */
dfaanalyze(d
, searchflag
)
int *nullable
; /* Nullable stack. */
int *nfirstpos
; /* Element count stack for firstpos sets. */
position
*firstpos
; /* Array where firstpos elements are stored. */
int *nlastpos
; /* Element count stack for lastpos sets. */
position
*lastpos
; /* Array where lastpos elements are stored. */
int *nalloc
; /* Sizes of arrays allocated to follow sets. */
position_set tmp
; /* Temporary set for merging sets. */
position_set merged
; /* Result of merging sets. */
int wants_newline
; /* True if some position wants newline info. */
position
*o_firstpos
, *o_lastpos
;
fprintf(stderr
, "dfaanalyze:\n");
for (i
= 0; i
< d
->tindex
; ++i
)
fprintf(stderr
, " %d:", i
);
d
->searchflag
= searchflag
;
MALLOC(nullable
, int, d
->depth
);
MALLOC(nfirstpos
, int, d
->depth
);
MALLOC(firstpos
, position
, d
->nleaves
);
o_firstpos
= firstpos
, firstpos
+= d
->nleaves
;
MALLOC(nlastpos
, int, d
->depth
);
MALLOC(lastpos
, position
, d
->nleaves
);
o_lastpos
= lastpos
, lastpos
+= d
->nleaves
;
MALLOC(nalloc
, int, d
->tindex
);
for (i
= 0; i
< d
->tindex
; ++i
)
MALLOC(merged
.elems
, position
, d
->nleaves
);
CALLOC(d
->follows
, position_set
, d
->tindex
);
for (i
= 0; i
< d
->tindex
; ++i
)
{ /* Nonsyntactic #ifdef goo... */
/* The empty set is nullable. */
/* The firstpos and lastpos of the empty leaf are both empty. */
*nfirstpos
++ = *nlastpos
++ = 0;
/* Every element in the firstpos of the argument is in the follow
of every element in the lastpos. */
tmp
.nelem
= nfirstpos
[-1];
for (j
= 0; j
< nlastpos
[-1]; ++j
)
merge(&tmp
, &d
->follows
[pos
[j
].index
], &merged
);
REALLOC_IF_NECESSARY(d
->follows
[pos
[j
].index
].elems
, position
,
nalloc
[pos
[j
].index
], merged
.nelem
- 1);
copy(&merged
, &d
->follows
[pos
[j
].index
]);
/* A QMARK or STAR node is automatically nullable. */
if (d
->tokens
[i
] != PLUS
)
/* Every element in the firstpos of the second argument is in the
follow of every element in the lastpos of the first argument. */
tmp
.nelem
= nfirstpos
[-1];
pos
= lastpos
+ nlastpos
[-1];
for (j
= 0; j
< nlastpos
[-2]; ++j
)
merge(&tmp
, &d
->follows
[pos
[j
].index
], &merged
);
REALLOC_IF_NECESSARY(d
->follows
[pos
[j
].index
].elems
, position
,
nalloc
[pos
[j
].index
], merged
.nelem
- 1);
copy(&merged
, &d
->follows
[pos
[j
].index
]);
/* The firstpos of a CAT node is the firstpos of the first argument,
union that of the second argument if the first is nullable. */
nfirstpos
[-2] += nfirstpos
[-1];
firstpos
+= nfirstpos
[-1];
/* The lastpos of a CAT node is the lastpos of the second argument,
union that of the first argument if the second is nullable. */
nlastpos
[-2] += nlastpos
[-1];
pos
= lastpos
+ nlastpos
[-2];
for (j
= nlastpos
[-1] - 1; j
>= 0; --j
)
nlastpos
[-2] = nlastpos
[-1];
/* A CAT node is nullable if both arguments are nullable. */
nullable
[-2] = nullable
[-1] && nullable
[-2];
/* The firstpos is the union of the firstpos of each argument. */
nfirstpos
[-2] += nfirstpos
[-1];
/* The lastpos is the union of the lastpos of each argument. */
nlastpos
[-2] += nlastpos
[-1];
/* An OR node is nullable if either argument is nullable. */
nullable
[-2] = nullable
[-1] || nullable
[-2];
/* Anything else is a nonempty position. (Note that special
constructs like \< are treated as nonempty strings here;
an "epsilon closure" effectively makes them nullable later.
Backreferences have to get a real position so we can detect
transitions on them later. But they are nullable. */
*nullable
++ = d
->tokens
[i
] == BACKREF
;
/* This position is in its own firstpos and lastpos. */
*nfirstpos
++ = *nlastpos
++ = 1;
firstpos
->index
= lastpos
->index
= i
;
firstpos
->constraint
= lastpos
->constraint
= NO_CONSTRAINT
;
/* Allocate the follow set for this position. */
MALLOC(d
->follows
[i
].elems
, position
, nalloc
[i
]);
/* ... balance the above nonsyntactic #ifdef goo... */
fprintf(stderr
, "node %d:", i
);
fprintf(stderr
, nullable
[-1] ? " nullable: yes\n" : " nullable: no\n");
fprintf(stderr
, " firstpos:");
for (j
= nfirstpos
[-1] - 1; j
>= 0; --j
)
fprintf(stderr
, " %d:", firstpos
[j
].index
);
prtok(d
->tokens
[firstpos
[j
].index
]);
fprintf(stderr
, "\n lastpos:");
for (j
= nlastpos
[-1] - 1; j
>= 0; --j
)
fprintf(stderr
, " %d:", lastpos
[j
].index
);
prtok(d
->tokens
[lastpos
[j
].index
]);
/* For each follow set that is the follow set of a real position, replace
it with its epsilon closure. */
for (i
= 0; i
< d
->tindex
; ++i
)
if (d
->tokens
[i
] < NOTCHAR
|| d
->tokens
[i
] == BACKREF
fprintf(stderr
, "follows(%d:", i
);
for (j
= d
->follows
[i
].nelem
- 1; j
>= 0; --j
)
fprintf(stderr
, " %d:", d
->follows
[i
].elems
[j
].index
);
prtok(d
->tokens
[d
->follows
[i
].elems
[j
].index
]);
copy(&d
->follows
[i
], &merged
);
if (d
->follows
[i
].nelem
< merged
.nelem
)
REALLOC(d
->follows
[i
].elems
, position
, merged
.nelem
);
copy(&merged
, &d
->follows
[i
]);
/* Get the epsilon closure of the firstpos of the regexp. The result will
be the set of positions of state 0. */
for (i
= 0; i
< nfirstpos
[-1]; ++i
)
insert(firstpos
[i
], &merged
);
/* Check if any of the positions of state 0 will want newline context. */
for (i
= 0; i
< merged
.nelem
; ++i
)
if (PREV_NEWLINE_DEPENDENT(merged
.elems
[i
].constraint
))
/* Build the initial state. */
MALLOC(d
->states
, dfa_state
, d
->salloc
);
state_index(d
, &merged
, wants_newline
, 0);
/* Find, for each character, the transition out of state s of d, and store
it in the appropriate slot of trans.
We divide the positions of s into groups (positions can appear in more
than one group). Each group is labeled with a set of characters that
every position in the group matches (taking into account, if necessary,
preceding context information of s). For each group, find the union
of the its elements' follows. This set is the set of positions of the
new state. For each character in the group's label, set the transition
on this character to be to a state corresponding to the set's positions,
and its associated backward context information, if necessary.
If we are building a searching matcher, we include the positions of state
The collection of groups is constructed by building an equivalence-class
partition of the positions of s.
For each position, find the set of characters C that it matches. Eliminate
any characters from C that fail on grounds of backward context.
Search through the groups, looking for a group whose label L has nonempty
intersection with C. If L - C is nonempty, create a new group labeled
L - C and having the same positions as the current group, and set L to
the intersection of L and C. Insert the position in this group, set
C = C - L, and resume scanning.
If after comparing with every group there are characters remaining in C,
create a new group labeled with the characters of C and insert this
position in that group. */
position_set grps
[NOTCHAR
]; /* As many as will ever be needed. */
charclass labels
[NOTCHAR
]; /* Labels corresponding to the groups. */
int ngrps
= 0; /* Number of groups actually used. */
position pos
; /* Current position being considered. */
charclass matches
; /* Set of matching characters. */
int matchesf
; /* True if matches is nonempty. */
charclass intersect
; /* Intersection with some label set. */
int intersectf
; /* True if intersect is nonempty. */
charclass leftovers
; /* Stuff in the label that didn't match. */
int leftoversf
; /* True if leftovers is nonempty. */
static charclass letters
; /* Set of characters considered letters. */
static charclass newline
; /* Set of characters that aren't newline. */
position_set follows
; /* Union of the follows of some group. */
position_set tmp
; /* Temporary space for merging sets. */
int state
; /* New state. */
int wants_newline
; /* New state wants to know newline context. */
int state_newline
; /* New state on a newline transition. */
int wants_letter
; /* New state wants to know letter context. */
int state_letter
; /* New state on a letter transition. */
static initialized
; /* Flag for static initialization. */
/* Initialize the set of letters, if necessary. */
for (i
= 0; i
< NOTCHAR
; ++i
)
for (i
= 0; i
< d
->states
[s
].elems
.nelem
; ++i
)
pos
= d
->states
[s
].elems
.elems
[i
];
if (d
->tokens
[pos
.index
] >= 0 && d
->tokens
[pos
.index
] < NOTCHAR
)
setbit(d
->tokens
[pos
.index
], matches
);
else if (d
->tokens
[pos
.index
] >= CSET
)
copyset(d
->charclasses
[d
->tokens
[pos
.index
] - CSET
], matches
);
/* Some characters may need to be eliminated from matches because
they fail in the current context. */
if (pos
.constraint
!= 0xFF)
if (! MATCHES_NEWLINE_CONTEXT(pos
.constraint
,
d
->states
[s
].newline
, 1))
if (! MATCHES_NEWLINE_CONTEXT(pos
.constraint
,
d
->states
[s
].newline
, 0))
for (j
= 0; j
< CHARCLASS_INTS
; ++j
)
matches
[j
] &= newline
[j
];
if (! MATCHES_LETTER_CONTEXT(pos
.constraint
,
for (j
= 0; j
< CHARCLASS_INTS
; ++j
)
matches
[j
] &= ~letters
[j
];
if (! MATCHES_LETTER_CONTEXT(pos
.constraint
,
for (j
= 0; j
< CHARCLASS_INTS
; ++j
)
matches
[j
] &= letters
[j
];
/* If there are no characters left, there's no point in going on. */
for (j
= 0; j
< CHARCLASS_INTS
&& !matches
[j
]; ++j
)
for (j
= 0; j
< ngrps
; ++j
)
/* If matches contains a single character only, and the current
group's label doesn't contain that character, go on to the
if (d
->tokens
[pos
.index
] >= 0 && d
->tokens
[pos
.index
] < NOTCHAR
&& !tstbit(d
->tokens
[pos
.index
], labels
[j
]))
/* Check if this group's label has a nonempty intersection with
for (k
= 0; k
< CHARCLASS_INTS
; ++k
)
(intersect
[k
] = matches
[k
] & labels
[j
][k
]) ? intersectf
= 1 : 0;
/* It does; now find the set differences both ways. */
leftoversf
= matchesf
= 0;
for (k
= 0; k
< CHARCLASS_INTS
; ++k
)
/* Even an optimizing compiler can't know this for sure. */
int match
= matches
[k
], label
= labels
[j
][k
];
(leftovers
[k
] = ~match
& label
) ? leftoversf
= 1 : 0;
(matches
[k
] = match
& ~label
) ? matchesf
= 1 : 0;
/* If there were leftovers, create a new group labeled with them. */
copyset(leftovers
, labels
[ngrps
]);
copyset(intersect
, labels
[j
]);
MALLOC(grps
[ngrps
].elems
, position
, d
->nleaves
);
copy(&grps
[j
], &grps
[ngrps
]);
/* Put the position in the current group. Note that there is no
reason to call insert() here. */
grps
[j
].elems
[grps
[j
].nelem
++] = pos
;
/* If every character matching the current position has been
accounted for, we're done. */
/* If we've passed the last group, and there are still characters
unaccounted for, then we'll have to create a new group. */
copyset(matches
, labels
[ngrps
]);
MALLOC(grps
[ngrps
].elems
, position
, d
->nleaves
);
grps
[ngrps
].elems
[0] = pos
;
MALLOC(follows
.elems
, position
, d
->nleaves
);
MALLOC(tmp
.elems
, position
, d
->nleaves
);
/* If we are a searching matcher, the default transition is to a state
containing the positions of state 0, otherwise the default transition
for (i
= 0; i
< d
->states
[0].elems
.nelem
; ++i
)
if (PREV_NEWLINE_DEPENDENT(d
->states
[0].elems
.elems
[i
].constraint
))
if (PREV_LETTER_DEPENDENT(d
->states
[0].elems
.elems
[i
].constraint
))
copy(&d
->states
[0].elems
, &follows
);
state
= state_index(d
, &follows
, 0, 0);
state_newline
= state_index(d
, &follows
, 1, 0);
state_letter
= state_index(d
, &follows
, 0, 1);
for (i
= 0; i
< NOTCHAR
; ++i
)
trans
[i
] = state_newline
;
for (i
= 0; i
< NOTCHAR
; ++i
)
for (i
= 0; i
< ngrps
; ++i
)
/* Find the union of the follows of the positions of the group.
This is a hideously inefficient loop. Fix it someday. */
for (j
= 0; j
< grps
[i
].nelem
; ++j
)
for (k
= 0; k
< d
->follows
[grps
[i
].elems
[j
].index
].nelem
; ++k
)
insert(d
->follows
[grps
[i
].elems
[j
].index
].elems
[k
], &follows
);
/* If we are building a searching matcher, throw in the positions
for (j
= 0; j
< d
->states
[0].elems
.nelem
; ++j
)
insert(d
->states
[0].elems
.elems
[j
], &follows
);
/* Find out if the new state will want any context information. */
if (tstbit('\n', labels
[i
]))
for (j
= 0; j
< follows
.nelem
; ++j
)
if (PREV_NEWLINE_DEPENDENT(follows
.elems
[j
].constraint
))
for (j
= 0; j
< CHARCLASS_INTS
; ++j
)
if (labels
[i
][j
] & letters
[j
])
for (j
= 0; j
< follows
.nelem
; ++j
)
if (PREV_LETTER_DEPENDENT(follows
.elems
[j
].constraint
))
/* Find the state(s) corresponding to the union of the follows. */
state
= state_index(d
, &follows
, 0, 0);
state_newline
= state_index(d
, &follows
, 1, 0);
state_letter
= state_index(d
, &follows
, 0, 1);
/* Set the transitions for each character in the current label. */
for (j
= 0; j
< CHARCLASS_INTS
; ++j
)
for (k
= 0; k
< INTBITS
; ++k
)
if (labels
[i
][j
] & 1 << k
)
trans
[c
] = state_newline
;
for (i
= 0; i
< ngrps
; ++i
)
/* Some routines for manipulating a compiled dfa's transition tables.
Each state may or may not have a transition table; if it does, and it
is a non-accepting state, then d->trans[state] points to its table.
If it is an accepting state then d->fails[state] points to its table.
If it has no table at all, then d->trans[state] is NULL.
TODO: Improve this comment, get rid of the unnecessary redundancy. */
int *trans
; /* The new transition table. */
/* Set an upper limit on the number of transition tables that will ever
exist at once. 1024 is arbitrary. The idea is that the frequently
used transition tables will be quickly rebuilt, whereas the ones that
were only needed once or twice will be cleared away. */
for (i
= 0; i
< d
->tralloc
; ++i
)
free((ptr_t
) d
->trans
[i
]);
free((ptr_t
) d
->fails
[i
]);
/* Set up the success bits for this state. */
if (ACCEPTS_IN_CONTEXT(d
->states
[s
].newline
, 1, d
->states
[s
].letter
, 0,
if (ACCEPTS_IN_CONTEXT(d
->states
[s
].newline
, 0, d
->states
[s
].letter
, 1,
if (ACCEPTS_IN_CONTEXT(d
->states
[s
].newline
, 0, d
->states
[s
].letter
, 0,
MALLOC(trans
, int, NOTCHAR
);
/* Now go through the new transition table, and make sure that the trans
and fail arrays are allocated large enough to hold a pointer for the
largest state mentioned in the table. */
for (i
= 0; i
< NOTCHAR
; ++i
)
if (trans
[i
] >= d
->tralloc
)
int oldalloc
= d
->tralloc
;
while (trans
[i
] >= d
->tralloc
)
REALLOC(d
->realtrans
, int *, d
->tralloc
+ 1);
d
->trans
= d
->realtrans
+ 1;
REALLOC(d
->fails
, int *, d
->tralloc
);
REALLOC(d
->success
, int, d
->tralloc
);
REALLOC(d
->newlines
, int, d
->tralloc
);
while (oldalloc
< d
->tralloc
)
d
->trans
[oldalloc
] = NULL
;
d
->fails
[oldalloc
++] = NULL
;
/* Keep the newline transition in a special place so we can use it as
d
->newlines
[s
] = trans
['\n'];
CALLOC(d
->realtrans
, int *, d
->tralloc
+ 1);
d
->trans
= d
->realtrans
+ 1;
CALLOC(d
->fails
, int *, d
->tralloc
);
MALLOC(d
->success
, int, d
->tralloc
);
MALLOC(d
->newlines
, int, d
->tralloc
);
/* Search through a buffer looking for a match to the given struct dfa.
Find the first occurrence of a string matching the regexp in the buffer,
and the shortest possible version thereof. Return a pointer to the first
character after the match, or NULL if none is found. Begin points to
the beginning of the buffer, and end points to the first character after
its end. We store a newline in *end to act as a sentinel, so end had
better point somewhere valid. Newline is a flag indicating whether to
allow newlines to be in the matching string. If count is non-
NULL it points to a place we're supposed to increment every time we
see a newline. Finally, if backref is non-NULL it points to a place
where we're supposed to store a 1 if backreferencing happened and the
match needs to be verified by a backtracking matcher. Otherwise
we store a 0 in *backref. */
dfaexec(d
, begin
, end
, newline
, count
, backref
)
register s
, s1
, tmp
; /* Current state. */
register unsigned char *p
; /* Current input character. */
register **trans
, *t
; /* Copy of d->trans so it can be optimized
static sbit
[NOTCHAR
]; /* Table for anding with d->success. */
for (i
= 0; i
< NOTCHAR
; ++i
)
p
= (unsigned char *) begin
;
/* The dreaded inner loop. */
while ((t
= trans
[s
]) != 0);
tmp
= s
, s
= s1
, s1
= tmp
;
if (s
>= 0 && p
<= (unsigned char *) end
&& d
->fails
[s
])
if (d
->success
[s
] & sbit
[*p
])
if (d
->states
[s
].backref
)
/* If the previous character was a newline, count it. */
if (count
&& (char *) p
<= end
&& p
[-1] == '\n')
/* Check if we've run off the end of the buffer. */
if (p
[-1] == '\n' && newline
)
/* Initialize the components of a dfa that the other routines don't
initialize for themselves. */
MALLOC(d
->charclasses
, charclass
, d
->calloc
);
MALLOC(d
->tokens
, token
, d
->talloc
);
d
->tindex
= d
->depth
= d
->nleaves
= d
->nregexps
= 0;
/* Parse and analyze a single string of the given length. */
dfacomp(s
, len
, d
, searchflag
)
if (case_fold
) /* dummy folding in service of dfamust() */
dfaerror("out of memory");
for (i
= 0; i
< len
; ++i
)
d
->cindex
= d
->tindex
= d
->depth
= d
->nleaves
= d
->nregexps
= 0;
dfaanalyze(d
, searchflag
);
dfaanalyze(d
, searchflag
);
/* Free the storage held by the components of a dfa. */
struct dfamust
*dm
, *ndm
;
free((ptr_t
) d
->charclasses
);
for (i
= 0; i
< d
->sindex
; ++i
)
free((ptr_t
) d
->states
[i
].elems
.elems
);
for (i
= 0; i
< d
->tindex
; ++i
)
free((ptr_t
) d
->follows
[i
].elems
);
free((ptr_t
) d
->follows
);
for (i
= 0; i
< d
->tralloc
; ++i
)
free((ptr_t
) d
->trans
[i
]);
free((ptr_t
) d
->fails
[i
]);
free((ptr_t
) d
->realtrans
);
free((ptr_t
) d
->newlines
);
for (dm
= d
->musts
; dm
; dm
= ndm
)
/* Having found the postfix representation of the regular expression,
try to find a long sequence of characters that must appear in any line
Finding a "longest" sequence is beyond the scope here;
we take an easy way out and hope for the best.
(Take "(ab|a)b"--please.)
We do a bottom-up calculation of sequences of characters that must appear
in matches of r.e.'s represented by trees rooted at the nodes of the postfix
sequences that must appear at the left of the match ("left")
sequences that must appear at the right of the match ("right")
lists of sequences that must appear somewhere in the match ("in")
sequences that must constitute the match ("is")
When we get to the root of the tree, we use one of the longest of its
calculated "in" sequences as our answer. The sequence we find is returned in
d->must (where "d" is the single argument passed to "dfamust");
the length of the sequence is returned in d->mustn.
The sequences calculated for the various types of node (in pseudo ANSI c)
are shown below. "p" is the operand of unary operators (and the left-hand
operand of binary operators); "q" is the right-hand operand of binary
"ZERO" means "a zero-length sequence" below.
QMARK ZERO ZERO ZERO ZERO
PLUS p->left p->right ZERO p->in
CAT (p->is==ZERO)? (q->is==ZERO)? (p->is!=ZERO && p->in plus
p->left : q->right : q->is!=ZERO) ? q->in plus
p->is##q->left p->right##q->is p->is##q->is : p->right##q->left
OR longest common longest common (do p->is and substrings common to
leading trailing q->is have same p->in and q->in
(sub)sequence (sub)sequence length and
of p->left of p->right content) ?
and q->left and q->right p->is : NULL
If there's anything else we recognize in the tree, all four sequences get set
to zero-length sequences. If there's something we don't recognize in the tree,
we just return a zero-length sequence.
Break ties in favor of infrequent letters (choosing 'zzz' in preference to
And. . .is it here or someplace that we might ponder "optimizations" such as
egrep 'psi|epsilon' -> egrep 'psi'
egrep 'pepsi|epsilon' -> egrep 'epsi'
(Yes, we now find "epsi" as a "string
that must occur", but we might also
simplify the *entire* r.e. being sought)
grep '(ab|a)b' -> grep 'ab'
There are several issues:
Is optimization easy (enough)?
Does optimization actually accomplish anything,
or is the automaton you get from "psi|epsilon" (for example)
the same as the one you get from "psi" (for example)?
Are optimizable r.e.'s likely to be used in real-life situations
(something like 'ab*' is probably unlikely; something like is
'psi|epsilon' is likelier)? */
newsize
= (new == NULL
) ? 0 : strlen(new);
else oldsize
= strlen(old
);
result
= (char *) malloc(newsize
+ 1);
result
= (char *) realloc((void *) old
, oldsize
+ newsize
+ 1);
if (result
!= NULL
&& new != NULL
)
(void) strcpy(result
+ oldsize
, new);
return icatalloc((char *) NULL
, string
);
for (cp
= lookin
; *cp
!= '\0'; ++cp
)
if (strncmp(cp
, lookfor
, len
) == 0)
for (i
= 0; cpp
[i
] != NULL
; ++i
)
if ((new = icpyalloc(new)) == NULL
)
/* Is there already something in the list that's new (or longer)? */
for (i
= 0; cpp
[i
] != NULL
; ++i
)
if (istrstr(cpp
[i
], new) != NULL
)
/* Eliminate any obsoleted strings. */
if (istrstr(new, cpp
[j
]) == NULL
)
/* Add the new string. */
cpp
= (char **) realloc((char *) cpp
, (i
+ 2) * sizeof *cpp
);
/* Given pointers to two strings, return a pointer to an allocated
list of their distinct common substrings. Return NULL if something
if (left
== NULL
|| right
== NULL
)
cpp
= (char **) malloc(sizeof *cpp
);
for (lcp
= left
; *lcp
!= '\0'; ++lcp
)
rcp
= index(right
, *lcp
);
for (i
= 1; lcp
[i
] != '\0' && lcp
[i
] == rcp
[i
]; ++i
)
rcp
= index(rcp
+ 1, *lcp
);
if ((cpp
= enlist(cpp
, lcp
, len
)) == NULL
)
if (old
== NULL
|| new == NULL
)
for (i
= 0; new[i
] != NULL
; ++i
)
old
= enlist(old
, new[i
], strlen(new[i
]));
/* Given two lists of substrings, return a new list giving substrings
if (left
== NULL
|| right
== NULL
)
both
= (char **) malloc(sizeof *both
);
for (lnum
= 0; left
[lnum
] != NULL
; ++lnum
)
for (rnum
= 0; right
[rnum
] != NULL
; ++rnum
)
temp
= comsubs(left
[lnum
], right
[rnum
]);
both
= addlists(both
, temp
);
mp
->left
[0] = mp
->right
[0] = mp
->is
[0] = '\0';
musts
= (must
*) malloc((dfa
->tindex
+ 1) * sizeof *musts
);
for (i
= 0; i
<= dfa
->tindex
; ++i
)
for (i
= 0; i
<= dfa
->tindex
; ++i
)
mp
[i
].in
= (char **) malloc(sizeof *mp
[i
].in
);
if (mp
[i
].in
== NULL
|| mp
[i
].left
== NULL
||
mp
[i
].right
== NULL
|| mp
[i
].is
== NULL
)
mp
[i
].left
[0] = mp
[i
].right
[0] = mp
[i
].is
[0] = '\0';
fprintf(stderr
, "dfamust:\n");
for (i
= 0; i
< dfa
->tindex
; ++i
)
fprintf(stderr
, " %d:", i
);
for (ri
= 0; ri
< dfa
->tindex
; ++ri
)
switch (t
= dfa
->tokens
[ri
])
goto done
; /* "cannot happen" */
goto done
; /* "cannot happen" */
goto done
; /* "cannot happen" */
/* Guaranteed to be. Unlikely, but. . . */
if (strcmp(lmp
->is
, rmp
->is
) != 0)
while (lmp
->left
[i
] != '\0' && lmp
->left
[i
] == rmp
->left
[i
])
if (lmp
->right
[ln
- i
- 1] != rmp
->right
[rn
- i
- 1])
lmp
->right
[j
] = lmp
->right
[(ln
- i
) + j
];
new = inboth(lmp
->in
, rmp
->in
);
goto done
; /* "cannot happen" */
goto done
; /* "cannot happen" */
for (i
= 0; musts
[0].in
[i
] != NULL
; ++i
)
if (strlen(musts
[0].in
[i
]) > strlen(result
))
if (strcmp(result
, musts
[0].is
) == 0)
goto done
; /* "cannot happen" */
/* In. Everything in left, plus everything in
right, plus catenation of
left's right and right's left. */
lmp
->in
= addlists(lmp
->in
, rmp
->in
);
if (lmp
->right
[0] != '\0' &&
tp
= icpyalloc(lmp
->right
);
tp
= icatalloc(tp
, rmp
->left
);
lmp
->in
= enlist(lmp
->in
, tp
,
lmp
->left
= icatalloc(lmp
->left
,
lmp
->right
= icatalloc(lmp
->right
, rmp
->right
);
if (lmp
->is
[0] != '\0' && rmp
->is
[0] != '\0')
lmp
->is
= icatalloc(lmp
->is
, rmp
->is
);
mp
->is
[0] = mp
->left
[0] = mp
->right
[0] = t
;
mp
->is
[1] = mp
->left
[1] = mp
->right
[1] = '\0';
mp
->in
= enlist(mp
->in
, mp
->is
, 1);
fprintf(stderr
, " node: %d:", ri
);
fprintf(stderr
, "\n in:");
for (i
= 0; mp
->in
[i
]; ++i
)
fprintf(stderr
, " \"%s\"", mp
->in
[i
]);
fprintf(stderr
, "\n is: \"%s\"\n", mp
->is
);
fprintf(stderr
, " left: \"%s\"\n", mp
->left
);
fprintf(stderr
, " right: \"%s\"\n", mp
->right
);
dm
= (struct dfamust
*) malloc(sizeof (struct dfamust
));
dm
->must
= malloc(strlen(result
) + 1);
strcpy(dm
->must
, result
);
for (i
= 0; i
<= dfa
->tindex
; ++i
)
ifree((char *) mp
[i
].in
);