~drizzle-trunk/drizzle/development

.TH REGEX 3 "17 May 1993"

.BY "Henry Spencer"

.de ZR

.\" one other place knows this name:  the SEE ALSO section

.IR regex (7) \\$1

..

.SH NAME

regcomp, regexec, regerror, regfree \- regular-expression library

.SH SYNOPSIS

.ft B

.\".na

#include <sys/types.h>

.br

#include <regex.h>

.HP 10

int regcomp(regex_t\ *preg, const\ char\ *pattern, int\ cflags);

.HP

int\ regexec(const\ regex_t\ *preg, const\ char\ *string,

size_t\ nmatch, regmatch_t\ pmatch[], int\ eflags);

.HP

size_t\ regerror(int\ errcode, const\ regex_t\ *preg,

char\ *errbuf, size_t\ errbuf_size);

.HP

void\ regfree(regex_t\ *preg);

.\".ad

.ft

.SH DESCRIPTION

These routines implement POSIX 1003.2 regular expressions (``RE''s);

see

.ZR .

.I Regcomp

compiles an RE written as a string into an internal form,

.I regexec

matches that internal form against a string and reports results,

.I regerror

transforms error codes from either into human-readable messages,

and

.I regfree

frees any dynamically-allocated storage used by the internal form

of an RE.

.PP

The header

.I <regex.h>

declares two structure types,

.I regex_t

and

.IR regmatch_t ,

the former for compiled internal forms and the latter for match reporting.

It also declares the four functions,

a type

.IR regoff_t ,

and a number of constants with names starting with ``REG_''.

.PP

.I Regcomp

compiles the regular expression contained in the

.I pattern

string,

subject to the flags in

.IR cflags ,

and places the results in the

.I regex_t

structure pointed to by

.IR preg .

.I Cflags

is the bitwise OR of zero or more of the following flags:

.IP REG_EXTENDED \w'REG_EXTENDED'u+2n

Compile modern (``extended'') REs,

rather than the obsolete (``basic'') REs that

are the default.

.IP REG_BASIC

This is a synonym for 0,

provided as a counterpart to REG_EXTENDED to improve readability.

.IP REG_NOSPEC

Compile with recognition of all special characters turned off.

All characters are thus considered ordinary,

so the ``RE'' is a literal string.

This is an extension,

compatible with but not specified by POSIX 1003.2,

and should be used with

caution in software intended to be portable to other systems.

REG_EXTENDED and REG_NOSPEC may not be used

in the same call to

.IR regcomp .

.IP REG_ICASE

Compile for matching that ignores upper/lower case distinctions.

See

.ZR .

.IP REG_NOSUB

Compile for matching that need only report success or failure,

not what was matched.

.IP REG_NEWLINE

Compile for newline-sensitive matching.

By default, newline is a completely ordinary character with no special

meaning in either REs or strings.

With this flag,

`[^' bracket expressions and `.' never match newline,

a `^' anchor matches the null string after any newline in the string

in addition to its normal function,

and the `$' anchor matches the null string before any newline in the

string in addition to its normal function.

.IP REG_PEND

The regular expression ends,

not at the first NUL,

but just before the character pointed to by the

.I re_endp

member of the structure pointed to by

.IR preg .

The

.I re_endp

member is of type

.IR const\ char\ * .

This flag permits inclusion of NULs in the RE;

they are considered ordinary characters.

This is an extension,

compatible with but not specified by POSIX 1003.2,

and should be used with

caution in software intended to be portable to other systems.

.PP

When successful,

.I regcomp

returns 0 and fills in the structure pointed to by

.IR preg .

One member of that structure

(other than

.IR re_endp )

is publicized:

.IR re_nsub ,

of type

.IR size_t ,

contains the number of parenthesized subexpressions within the RE

(except that the value of this member is undefined if the

REG_NOSUB flag was used).

If

.I regcomp

fails, it returns a non-zero error code;

see DIAGNOSTICS.

.PP

.I Regexec

matches the compiled RE pointed to by

.I preg

against the

.IR string ,

subject to the flags in

.IR eflags ,

and reports results using

.IR nmatch ,

.IR pmatch ,

and the returned value.

The RE must have been compiled by a previous invocation of

.IR regcomp .

The compiled form is not altered during execution of

.IR regexec ,

so a single compiled RE can be used simultaneously by multiple threads.

.PP

By default,

the NUL-terminated string pointed to by

.I string

is considered to be the text of an entire line, minus any terminating

newline.

The

.I eflags

argument is the bitwise OR of zero or more of the following flags:

.IP REG_NOTBOL \w'REG_STARTEND'u+2n

The first character of

the string

is not the beginning of a line, so the `^' anchor should not match before it.

This does not affect the behavior of newlines under REG_NEWLINE.

.IP REG_NOTEOL

The NUL terminating

the string

does not end a line, so the `$' anchor should not match before it.

This does not affect the behavior of newlines under REG_NEWLINE.

.IP REG_STARTEND

The string is considered to start at

\fIstring\fR\ + \fIpmatch\fR[0].\fIrm_so\fR

and to have a terminating NUL located at

\fIstring\fR\ + \fIpmatch\fR[0].\fIrm_eo\fR

(there need not actually be a NUL at that location),

regardless of the value of

.IR nmatch .

See below for the definition of

.IR pmatch

and

.IR nmatch .

This is an extension,

compatible with but not specified by POSIX 1003.2,

and should be used with

caution in software intended to be portable to other systems.

Note that a non-zero \fIrm_so\fR does not imply REG_NOTBOL;

REG_STARTEND affects only the location of the string,

not how it is matched.

.PP

See

.ZR

for a discussion of what is matched in situations where an RE or a

portion thereof could match any of several substrings of

.IR string .

.PP

Normally,

.I regexec

returns 0 for success and the non-zero code REG_NOMATCH for failure.

Other non-zero error codes may be returned in exceptional situations;

see DIAGNOSTICS.

.PP

If REG_NOSUB was specified in the compilation of the RE,

or if

.I nmatch

is 0,

.I regexec

ignores the

.I pmatch

argument (but see below for the case where REG_STARTEND is specified).

Otherwise,

.I pmatch

points to an array of

.I nmatch

structures of type

.IR regmatch_t .

Such a structure has at least the members

.I rm_so

and

.IR rm_eo ,

both of type

.I regoff_t

(a signed arithmetic type at least as large as an

.I off_t

and a

.IR ssize_t ),

containing respectively the offset of the first character of a substring

and the offset of the first character after the end of the substring.

Offsets are measured from the beginning of the

.I string

argument given to

.IR regexec .

An empty substring is denoted by equal offsets,

both indicating the character following the empty substring.

.PP

The 0th member of the

.I pmatch

array is filled in to indicate what substring of

.I string

was matched by the entire RE.

Remaining members report what substring was matched by parenthesized

subexpressions within the RE;

member

.I i

reports subexpression

.IR i ,

with subexpressions counted (starting at 1) by the order of their opening

parentheses in the RE, left to right.

Unused entries in the array\(emcorresponding either to subexpressions that

did not participate in the match at all, or to subexpressions that do not

exist in the RE (that is, \fIi\fR\ > \fIpreg\fR\->\fIre_nsub\fR)\(emhave both

.I rm_so

and

.I rm_eo

set to \-1.

If a subexpression participated in the match several times,

the reported substring is the last one it matched.

(Note, as an example in particular, that when the RE `(b*)+' matches `bbb',

the parenthesized subexpression matches each of the three `b's and then

an infinite number of empty strings following the last `b',

so the reported substring is one of the empties.)

.PP

If REG_STARTEND is specified,

.I pmatch

must point to at least one

.I regmatch_t

(even if

.I nmatch

is 0 or REG_NOSUB was specified),

to hold the input offsets for REG_STARTEND.

Use for output is still entirely controlled by

.IR nmatch ;

if

.I nmatch

is 0 or REG_NOSUB was specified,

the value of

.IR pmatch [0]

will not be changed by a successful

.IR regexec .

.PP

.I Regerror

maps a non-zero

.I errcode

from either

.I regcomp

or

.I regexec

to a human-readable, printable message.

If

.I preg

is non-NULL,

the error code should have arisen from use of

the

.I regex_t

pointed to by

.IR preg ,

and if the error code came from

.IR regcomp ,

it should have been the result from the most recent

.I regcomp

using that

.IR regex_t .

.RI ( Regerror

may be able to supply a more detailed message using information

from the

.IR regex_t .)

.I Regerror

places the NUL-terminated message into the buffer pointed to by

.IR errbuf ,

limiting the length (including the NUL) to at most

.I errbuf_size

bytes.

If the whole message won't fit,

as much of it as will fit before the terminating NUL is supplied.

In any case,

the returned value is the size of buffer needed to hold the whole

message (including terminating NUL).

If

.I errbuf_size

is 0,

.I errbuf

is ignored but the return value is still correct.

.PP

If the

.I errcode

given to

.I regerror

is first ORed with REG_ITOA,

the ``message'' that results is the printable name of the error code,

e.g. ``REG_NOMATCH'',

rather than an explanation thereof.

If

.I errcode

is REG_ATOI,

then

.I preg

shall be non-NULL and the

.I re_endp

member of the structure it points to

must point to the printable name of an error code;

in this case, the result in

.I errbuf

is the decimal digits of

the numeric value of the error code

(0 if the name is not recognized).

REG_ITOA and REG_ATOI are intended primarily as debugging facilities;

they are extensions,

compatible with but not specified by POSIX 1003.2,

and should be used with

caution in software intended to be portable to other systems.

Be warned also that they are considered experimental and changes are possible.

.PP

.I Regfree

frees any dynamically-allocated storage associated with the compiled RE

pointed to by

.IR preg .

The remaining

.I regex_t

is no longer a valid compiled RE

and the effect of supplying it to

.I regexec

or

.I regerror

is undefined.

.PP

None of these functions references global variables except for tables

of constants;

all are safe for use from multiple threads if the arguments are safe.

.SH IMPLEMENTATION CHOICES

There are a number of decisions that 1003.2 leaves up to the implementor,

either by explicitly saying ``undefined'' or by virtue of them being

forbidden by the RE grammar.

This implementation treats them as follows.

.PP

See

.ZR

for a discussion of the definition of case-independent matching.

.PP

There is no particular limit on the length of REs,

except insofar as memory is limited.

Memory usage is approximately linear in RE size, and largely insensitive

to RE complexity, except for bounded repetitions.

See BUGS for one short RE using them

that will run almost any system out of memory.

.PP

A backslashed character other than one specifically given a magic meaning

by 1003.2 (such magic meanings occur only in obsolete [``basic''] REs)

is taken as an ordinary character.

.PP

Any unmatched [ is a REG_EBRACK error.

.PP

Equivalence classes cannot begin or end bracket-expression ranges.

The endpoint of one range cannot begin another.

.PP

RE_DUP_MAX, the limit on repetition counts in bounded repetitions, is 255.

.PP

A repetition operator (?, *, +, or bounds) cannot follow another

repetition operator.

A repetition operator cannot begin an expression or subexpression

or follow `^' or `|'.

.PP

`|' cannot appear first or last in a (sub)expression or after another `|',

i.e. an operand of `|' cannot be an empty subexpression.

An empty parenthesized subexpression, `()', is legal and matches an

empty (sub)string.

An empty string is not a legal RE.

.PP

A `{' followed by a digit is considered the beginning of bounds for a

bounded repetition, which must then follow the syntax for bounds.

A `{' \fInot\fR followed by a digit is considered an ordinary character.

.PP

`^' and `$' beginning and ending subexpressions in obsolete (``basic'')

REs are anchors, not ordinary characters.

.SH SEE ALSO

grep(1), regex(7)

.PP

POSIX 1003.2, sections 2.8 (Regular Expression Notation)

and

B.5 (C Binding for Regular Expression Matching).

.SH DIAGNOSTICS

Non-zero error codes from

.I regcomp

and

.I regexec

include the following:

.PP

.nf

.ta \w'REG_ECOLLATE'u+3n

REG_NOMATCH	regexec() failed to match

REG_BADPAT	invalid regular expression

REG_ECOLLATE	invalid collating element

REG_ECTYPE	invalid character class

REG_EESCAPE	\e applied to unescapable character

REG_ESUBREG	invalid backreference number

REG_EBRACK	brackets [ ] not balanced

REG_EPAREN	parentheses ( ) not balanced

REG_EBRACE	braces { } not balanced

REG_BADBR	invalid repetition count(s) in { }

REG_ERANGE	invalid character range in [ ]

REG_ESPACE	ran out of memory

REG_BADRPT	?, *, or + operand invalid

REG_EMPTY	empty (sub)expression

REG_ASSERT	``can't happen''\(emyou found a bug

REG_INVARG	invalid argument, e.g. negative-length string

.fi

.SH HISTORY

Written by Henry Spencer at University of Toronto,

henry@zoo.toronto.edu.

.SH BUGS

This is an alpha release with known defects.

Please report problems.

.PP

There is one known functionality bug.

The implementation of internationalization is incomplete:

the locale is always assumed to be the default one of 1003.2,

and only the collating elements etc. of that locale are available.

.PP

The back-reference code is subtle and doubts linger about its correctness

in complex cases.

.PP

.I Regexec

performance is poor.

This will improve with later releases.

.I Nmatch

exceeding 0 is expensive;

.I nmatch

exceeding 1 is worse.

.I Regexec

is largely insensitive to RE complexity \fIexcept\fR that back

references are massively expensive.

RE length does matter; in particular, there is a strong speed bonus

for keeping RE length under about 30 characters,

with most special characters counting roughly double.

.PP

.I Regcomp

implements bounded repetitions by macro expansion,

which is costly in time and space if counts are large

or bounded repetitions are nested.

An RE like, say,

`((((a{1,100}){1,100}){1,100}){1,100}){1,100}'

will (eventually) run almost any existing machine out of swap space.

.PP

There are suspected problems with response to obscure error conditions.

Notably,

certain kinds of internal overflow,

produced only by truly enormous REs or by multiply nested bounded repetitions,

are probably not handled well.

.PP

Due to a mistake in 1003.2, things like `a)b' are legal REs because `)' is

a special character only in the presence of a previous unmatched `('.

This can't be fixed until the spec is fixed.

.PP

The standard's definition of back references is vague.

For example, does

`a\e(\e(b\e)*\e2\e)*d' match `abbbd'?

Until the standard is clarified,

behavior in such cases should not be relied on.

.PP

The implementation of word-boundary matching is a bit of a kludge,

and bugs may lurk in combinations of word-boundary matching and anchoring.