README file for PCRE (Perl-compatible regular expressions)
----------------------------------------------------------
*******************************************************************************
* IMPORTANT FOR THOSE UPGRADING FROM VERSIONS BEFORE 2.00 *
* *
* Please note that there has been a change in the API such that a larger *
* ovector is required at matching time, to provide some additional workspace. *
* The new man page has details. This change was necessary in order to support *
* some of the new functionality in Perl 5.005. *
* *
* IMPORTANT FOR THOSE UPGRADING FROM VERSION 2.00 *
* *
* Another (I hope this is the last!) change has been made to the API for the *
* pcre_compile() function. An additional argument has been added to make it *
* possible to pass over a pointer to character tables built in the current *
* locale by pcre_maketables(). To use the default tables, this new arguement *
* should be passed as NULL. *
* *
* IMPORTANT FOR THOSE UPGRADING FROM VERSION 2.05 *
* *
* Yet another (and again I hope this really is the last) change has been made *
* to the API for the pcre_exec() function. An additional argument has been *
* added to make it possible to start the match other than at the start of the *
* subject string. This is important if there are lookbehinds. The new man *
* page has the details, but you just want to convert existing programs, all *
* you need to do is to stick in a new fifth argument to pcre_exec(), with a *
* value of zero. For example, change *
* *
* pcre_exec(pattern, extra, subject, length, options, ovec, ovecsize) *
* to *
* pcre_exec(pattern, extra, subject, length, 0, options, ovec, ovecsize) *
*******************************************************************************
The distribution should contain the following files:
ChangeLog log of changes to the code
LICENCE conditions for the use of PCRE
Makefile for building PCRE in Unix systems
README this file
RunTest a Unix shell script for running tests
Tech.Notes notes on the encoding
pcre.3 man page source for the functions
pcre.3.txt plain text version
pcre.3.html HTML version
pcreposix.3 man page source for the POSIX wrapper API
pcreposix.3.txt plain text version
pcreposix.3.HTML HTML version
dftables.c auxiliary program for building chartables.c
get.c )
maketables.c )
study.c ) source of
pcre.c ) the functions
pcreposix.c )
pcre.h header for the external API
pcreposix.h header for the external POSIX wrapper API
internal.h header for internal use
pcretest.c test program
pgrep.1 man page source for pgrep
pgrep.1.txt plain text version
pgrep.1.HTML HTML version
pgrep.c source of a grep utility that uses PCRE
perltest Perl test program
testinput1 test data, compatible with Perl 5.004 and 5.005
testinput2 test data for error messages and non-Perl things
testinput3 test data, compatible with Perl 5.005
testinput4 test data for locale-specific tests
testoutput1 test results corresponding to testinput1
testoutput2 test results corresponding to testinput2
testoutput3 test results corresponding to testinput3
testoutput4 test results corresponding to testinput4
dll.mk for Win32 DLL
pcre.def ditto
To build PCRE on a Unix system, first edit Makefile for your system. It is a
fairly simple make file, and there are some comments near the top, after the
text "On a Unix system". Then run "make". It builds two libraries called
libpcre.a and libpcreposix.a, a test program called pcretest, and the pgrep
command. You can use "make install" to copy these, and the public header file
pcre.h, to appropriate live directories on your system. These installation
directories are defined at the top of the Makefile, and you should edit them if
necessary.
For a non-Unix system, read the comments at the top of Makefile, which give
some hints on what needs to be done. PCRE has been compiled on Windows systems
and on Macintoshes, but I don't know the details as I don't use those systems.
It should be straightforward to build PCRE on any system that has a Standard C
compiler.
Some help in building a Win32 DLL of PCRE in GnuWin32 environments was
contributed by Paul.Sokolovsky@technologist.com. These environments are
Mingw32 (http://www.xraylith.wisc.edu/~khan/software/gnu-win32/) and
CygWin (http://sourceware.cygnus.com/cygwin/). Paul comments:
For CygWin, set CFLAGS=-mno-cygwin, and do 'make dll'. You'll get
pcre.dll (containing pcreposix also), libpcre.dll.a, and dynamically
linked pgrep and pcretest. If you have /bin/sh, run RunTest (three
main test go ok, locale not supported).
To test PCRE, run the RunTest script in the pcre directory. This can also be
run by "make runtest". It runs the pcretest test program (which is documented
below) on each of the testinput files in turn, and compares the output with the
contents of the corresponding testoutput file. A file called testtry is used to
hold the output from pcretest. To run pcretest on just one of the test files,
give its number as an argument to RunTest, for example:
RunTest 3
The first and third test files can also be fed directly into the perltest
script to check that Perl gives the same results. The third file requires the
additional features of release 5.005, which is why it is kept separate from the
main test input, which needs only Perl 5.004. In the long run, when 5.005 is
widespread, these two test files may get amalgamated.
The second set of tests check pcre_info(), pcre_study(), pcre_copy_substring(),
pcre_get_substring(), pcre_get_substring_list(), error detection and run-time
flags that are specific to PCRE, as well as the POSIX wrapper API.
The fourth set of tests checks pcre_maketables(), the facility for building a
set of character tables for a specific locale and using them instead of the
default tables. The tests make use of the "fr" (French) locale. Before running
the test, the script checks for the presence of this locale by running the
"locale" command. If that command fails, or if it doesn't include "fr" in the
list of available locales, the fourth test cannot be run, and a comment is
output to say why. If running this test produces instances of the error
** Failed to set locale "fr"
in the comparison output, it means that locale is not available on your system,
despite being listed by "locale". This does not mean that PCRE is broken.
PCRE has its own native API, but a set of "wrapper" functions that are based on
the POSIX API are also supplied in the library libpcreposix.a. Note that this
just provides a POSIX calling interface to PCRE: the regular expressions
themselves still follow Perl syntax and semantics. The header file
for the POSIX-style functions is called pcreposix.h. The official POSIX name is
regex.h, but I didn't want to risk possible problems with existing files of
that name by distributing it that way. To use it with an existing program that
uses the POSIX API, it will have to be renamed or pointed at by a link.
Character tables
----------------
PCRE uses four tables for manipulating and identifying characters. The final
argument of the pcre_compile() function is a pointer to a block of memory
containing the concatenated tables. A call to pcre_maketables() can be used to
generate a set of tables in the current locale. If the final argument for
pcre_compile() is passed as NULL, a set of default tables that is built into
the binary is used.
The source file called chartables.c contains the default set of tables. This is
not supplied in the distribution, but is built by the program dftables
(compiled from dftables.c), which uses the ANSI C character handling functions
such as isalnum(), isalpha(), isupper(), islower(), etc. to build the table
sources. This means that the default C locale which is set for your system will
control the contents of these default tables. You can change the default tables
by editing chartables.c and then re-building PCRE. If you do this, you should
probably also edit Makefile to ensure that the file doesn't ever get
re-generated.
The first two 256-byte tables provide lower casing and case flipping functions,
respectively. The next table consists of three 32-byte bit maps which identify
digits, "word" characters, and white space, respectively. These are used when
building 32-byte bit maps that represent character classes.
The final 256-byte table has bits indicating various character types, as
follows:
1 white space character
2 letter
4 decimal digit
8 hexadecimal digit
16 alphanumeric or '_'
128 regular expression metacharacter or binary zero
You should not alter the set of characters that contain the 128 bit, as that
will cause PCRE to malfunction.
The pcretest program
--------------------
This program is intended for testing PCRE, but it can also be used for
experimenting with regular expressions.
If it is given two filename arguments, it reads from the first and writes to
the second. If it is given only one filename argument, it reads from that file
and writes to stdout. Otherwise, it reads from stdin and writes to stdout, and
prompts for each line of input.
The program handles any number of sets of input on a single input file. Each
set starts with a regular expression, and continues with any number of data
lines to be matched against the pattern. An empty line signals the end of the
set. The regular expressions are given enclosed in any non-alphameric
delimiters other than backslash, for example
/(a|bc)x+yz/
White space before the initial delimiter is ignored. A regular expression may
be continued over several input lines, in which case the newline characters are
included within it. See the testinput files for many examples. It is possible
to include the delimiter within the pattern by escaping it, for example
/abc\/def/
If you do so, the escape and the delimiter form part of the pattern, but since
delimiters are always non-alphameric, this does not affect its interpretation.
If the terminating delimiter is immediately followed by a backslash, for
example,
/abc/\
then a backslash is added to the end of the pattern. This is done to provide a
way of testing the error condition that arises if a pattern finishes with a
backslash, because
/abc\/
is interpreted as the first line of a pattern that starts with "abc/", causing
pcretest to read the next line as a continuation of the regular expression.
The pattern may be followed by i, m, s, or x to set the PCRE_CASELESS,
PCRE_MULTILINE, PCRE_DOTALL, or PCRE_EXTENDED options, respectively. For
example:
/caseless/i
These modifier letters have the same effect as they do in Perl. There are
others which set PCRE options that do not correspond to anything in Perl: /A,
/E, and /X set PCRE_ANCHORED, PCRE_DOLLAR_ENDONLY, and PCRE_EXTRA respectively.
Searching for all possible matches within each subject string can be requested
by the /g or /G modifier. After finding a match, PCRE is called again to search
the remainder of the subject string. The difference between /g and /G is that
the former uses the startoffset argument to pcre_exec() to start searching at
a new point within the entire string (which is in effect what Perl does),
whereas the latter passes over a shortened substring. This makes a difference
to the matching process if the pattern begins with a lookbehind assertion
(including \b or \B).
If any call to pcre_exec() in a /g or /G sequence matches an empty string, the
next call is done with the PCRE_NOTEMPTY flag set so that it cannot match an
empty string again. This imitates the way Perl handles such cases when using
the /g modifier or the split() function.
There are a number of other modifiers for controlling the way pcretest
operates.
The /+ modifier requests that as well as outputting the substring that matched
the entire pattern, pcretest should in addition output the remainder of the
subject string. This is useful for tests where the subject contains multiple
copies of the same substring.
The /L modifier must be followed directly by the name of a locale, for example,
/pattern/Lfr
For this reason, it must be the last modifier letter. The given locale is set,
pcre_maketables() is called to build a set of character tables for the locale,
and this is then passed to pcre_compile() when compiling the regular
expression. Without an /L modifier, NULL is passed as the tables pointer; that
is, /L applies only to the expression on which it appears.
The /I modifier requests that pcretest output information about the compiled
expression (whether it is anchored, has a fixed first character, and so on). It
does this by calling pcre_info() after compiling an expression, and outputting
the information it gets back. If the pattern is studied, the results of that
are also output.
The /D modifier is a PCRE debugging feature, which also assumes /I. It causes
the internal form of compiled regular expressions to be output after
compilation.
The /S modifier causes pcre_study() to be called after the expression has been
compiled, and the results used when the expression is matched.
The /M modifier causes the size of memory block used to hold the compiled
pattern to be output.
Finally, the /P modifier causes pcretest to call PCRE via the POSIX wrapper API
rather than its native API. When this is done, all other modifiers except /i,
/m, and /+ are ignored. REG_ICASE is set if /i is present, and REG_NEWLINE is
set if /m is present. The wrapper functions force PCRE_DOLLAR_ENDONLY always,
and PCRE_DOTALL unless REG_NEWLINE is set.
Before each data line is passed to pcre_exec(), leading and trailing whitespace
is removed, and it is then scanned for \ escapes. The following are recognized:
\a alarm (= BEL)
\b backspace
\e escape
\f formfeed
\n newline
\r carriage return
\t tab
\v vertical tab
\nnn octal character (up to 3 octal digits)
\xhh hexadecimal character (up to 2 hex digits)
\A pass the PCRE_ANCHORED option to pcre_exec()
\B pass the PCRE_NOTBOL option to pcre_exec()
\Cdd call pcre_copy_substring() for substring dd after a successful match
(any decimal number less than 32)
\Gdd call pcre_get_substring() for substring dd after a successful match
(any decimal number less than 32)
\L call pcre_get_substringlist() after a successful match
\N pass the PCRE_NOTEMPTY option to pcre_exec()
\Odd set the size of the output vector passed to pcre_exec() to dd
(any number of decimal digits)
\Z pass the PCRE_NOTEOL option to pcre_exec()
A backslash followed by anything else just escapes the anything else. If the
very last character is a backslash, it is ignored. This gives a way of passing
an empty line as data, since a real empty line terminates the data input.
If /P was present on the regex, causing the POSIX wrapper API to be used, only
\B, and \Z have any effect, causing REG_NOTBOL and REG_NOTEOL to be passed to
regexec() respectively.
When a match succeeds, pcretest outputs the list of captured substrings that
pcre_exec() returns, starting with number 0 for the string that matched the
whole pattern. Here is an example of an interactive pcretest run.
$ pcretest
PCRE version 2.06 08-Jun-1999
re> /^abc(\d+)/
data> abc123
0: abc123
1: 123
data> xyz
No match
If the strings contain any non-printing characters, they are output as \0x
escapes. If the pattern has the /+ modifier, then the output for substring 0 is
followed by the the rest of the subject string, identified by "0+" like this:
re> /cat/+
data> cataract
0: cat
0+ aract
If the pattern has the /g or /G modifier, the results of successive matching
attempts are output in sequence, like this:
re> /\Bi(\w\w)/g
data> Mississippi
0: iss
1: ss
0: iss
1: ss
0: ipp
1: pp
"No match" is output only if the first match attempt fails.
If any of \C, \G, or \L are present in a data line that is successfully
matched, the substrings extracted by the convenience functions are output with
C, G, or L after the string number instead of a colon. This is in addition to
the normal full list. The string length (that is, the return from the
extraction function) is given in parentheses after each string for \C and \G.
Note that while patterns can be continued over several lines (a plain ">"
prompt is used for continuations), data lines may not. However newlines can be
included in data by means of the \n escape.
If the -p option is given to pcretest, it is equivalent to adding /P to each
regular expression: the POSIX wrapper API is used to call PCRE. None of the
following flags has any effect in this case.
If the option -d is given to pcretest, it is equivalent to adding /D to each
regular expression: the internal form is output after compilation.
If the option -i is given to pcretest, it is equivalent to adding /I to each
regular expression: information about the compiled pattern is given after
compilation.
If the option -m is given to pcretest, it outputs the size of each compiled
pattern after it has been compiled. It is equivalent to adding /M to each
regular expression. For compatibility with earlier versions of pcretest, -s is
a synonym for -m.
If the -t option is given, each compile, study, and match is run 20000 times
while being timed, and the resulting time per compile or match is output in
milliseconds. Do not set -t with -s, because you will then get the size output
20000 times and the timing will be distorted. If you want to change the number
of repetitions used for timing, edit the definition of LOOPREPEAT at the top of
pcretest.c
The perltest program
--------------------
The perltest program tests Perl's regular expressions; it has the same
specification as pcretest, and so can be given identical input, except that
input patterns can be followed only by Perl's lower case modifiers. The
contents of testinput1 and testinput3 meet this condition.
The data lines are processed as Perl double-quoted strings, so if they contain
" \ $ or @ characters, these have to be escaped. For this reason, all such
characters in testinput1 and testinput3 are escaped so that they can be used
for perltest as well as for pcretest, and the special upper case modifiers such
as /A that pcretest recognizes are not used in these files. The output should
be identical, apart from the initial identifying banner.
The testinput2 and testinput4 files are not suitable for feeding to perltest,
since they do make use of the special upper case modifiers and escapes that
pcretest uses to test some features of PCRE. The first of these files also
contains malformed regular expressions, in order to check that PCRE diagnoses
them correctly.
Philip Hazel <ph10@cam.ac.uk>
July 1999