Compiling PCRE on non-Unix systems

 ----------------------------------

 This document contains the following sections:

   General

   Generic instructions for the PCRE C library

   The C++ wrapper functions

   Building for virtual Pascal

   Stack size in Windows environments

   Linking programs in Windows environments

   Comments about Win32 builds

   Building PCRE on Windows with CMake

   Use of relative paths with CMake on Windows

   Testing with runtest.bat

   Building under Windows with BCC5.5

   Building PCRE on OpenVMS

 GENERAL

 I (Philip Hazel) have no experience of Windows or VMS sytems and how their

 libraries work. The items in the PCRE distribution and Makefile that relate to

 anything other than Unix-like systems are untested by me.

 There are some other comments and files (including some documentation in CHM

 format) in the Contrib directory on the FTP site:

   ftp://ftp.csx.cam.ac.uk/pub/software/programming/pcre/Contrib

 If you want to compile PCRE for a non-Unix system (especially for a system that

 does not support "configure" and "make" files), note that the basic PCRE

 library consists entirely of code written in Standard C, and so should compile

 successfully on any system that has a Standard C compiler and library. The C++

 wrapper functions are a separate issue (see below).

 The PCRE distribution includes a "configure" file for use by the Configure/Make

 build system, as found in many Unix-like environments. There is also support

 support for CMake, which some users prefer, in particular in Windows

 environments. There are some instructions for CMake under Windows in the

 section entitled "Building PCRE with CMake" below. CMake can also be used to

 build PCRE in Unix-like systems.

 GENERIC INSTRUCTIONS FOR THE PCRE C LIBRARY

 The following are generic comments about building the PCRE C library "by hand".

  (1) Copy or rename the file config.h.generic as config.h, and edit the macro

      settings that it contains to whatever is appropriate for your environment.

      In particular, if you want to force a specific value for newline, you can

      define the NEWLINE macro. When you compile any of the PCRE modules, you

      must specify -DHAVE_CONFIG_H to your compiler so that config.h is included

      in the sources.

      An alternative approach is not to edit config.h, but to use -D on the

      compiler command line to make any changes that you need to the

      configuration options. In this case -DHAVE_CONFIG_H must not be set.

      NOTE: There have been occasions when the way in which certain parameters

      in config.h are used has changed between releases. (In the configure/make

      world, this is handled automatically.) When upgrading to a new release,

      you are strongly advised to review config.h.generic before re-using what

      you had previously.

  (2) Copy or rename the file pcre.h.generic as pcre.h.

  (3) EITHER:

        Copy or rename file pcre_chartables.c.dist as pcre_chartables.c.

      OR:

        Compile dftables.c as a stand-alone program (using -DHAVE_CONFIG_H if

        you have set up config.h), and then run it with the single argument

        "pcre_chartables.c". This generates a set of standard character tables

        and writes them to that file. The tables are generated using the default

        C locale for your system. If you want to use a locale that is specified

        by LC_xxx environment variables, add the -L option to the dftables

        command. You must use this method if you are building on a system that

        uses EBCDIC code.

      The tables in pcre_chartables.c are defaults. The caller of PCRE can

      specify alternative tables at run time.

  (4) Ensure that you have the following header files:

        pcre_internal.h

        ucp.h

  (5) Also ensure that you have the following file, which is #included as source

      when building a debugging version of PCRE, and is also used by pcretest.

        pcre_printint.src

  (6) Compile the following source files, setting -DHAVE_CONFIG_H as a compiler

      option if you have set up config.h with your configuration, or else use

      other -D settings to change the configuration as required.

        pcre_chartables.c

        pcre_compile.c

        pcre_config.c

        pcre_dfa_exec.c

        pcre_exec.c

        pcre_fullinfo.c

        pcre_get.c

        pcre_globals.c

        pcre_info.c

        pcre_maketables.c

        pcre_newline.c

        pcre_ord2utf8.c

        pcre_refcount.c

        pcre_study.c

        pcre_tables.c

        pcre_try_flipped.c

        pcre_ucd.c

        pcre_valid_utf8.c

        pcre_version.c

        pcre_xclass.c

      Make sure that you include -I. in the compiler command (or equivalent for

      an unusual compiler) so that all included PCRE header files are first

      sought in the current directory. Otherwise you run the risk of picking up

      a previously-installed file from somewhere else.

  (7) Now link all the compiled code into an object library in whichever form

      your system keeps such libraries. This is the basic PCRE C library. If

      your system has static and shared libraries, you may have to do this once

      for each type.

  (8) Similarly, compile pcreposix.c (remembering -DHAVE_CONFIG_H if necessary)

      and link the result (on its own) as the pcreposix library.

  (9) Compile the test program pcretest.c (again, don't forget -DHAVE_CONFIG_H).

      This needs the functions in the pcre and pcreposix libraries when linking.

      It also needs the pcre_printint.src source file, which it #includes.

 (10) Run pcretest on the testinput files in the testdata directory, and check

      that the output matches the corresponding testoutput files. Note that the

      supplied files are in Unix format, with just LF characters as line

      terminators. You may need to edit them to change this if your system uses

      a different convention. If you are using Windows, you probably should use

      the wintestinput3 file instead of testinput3 (and the corresponding output

      file). This is a locale test; wintestinput3 sets the locale to "french"

      rather than "fr_FR", and there some minor output differences.

 (11) If you want to use the pcregrep command, compile and link pcregrep.c; it

      uses only the basic PCRE library (it does not need the pcreposix library).

 THE C++ WRAPPER FUNCTIONS

 The PCRE distribution also contains some C++ wrapper functions and tests,

 contributed by Google Inc. On a system that can use "configure" and "make",

 the functions are automatically built into a library called pcrecpp. It should

 be straightforward to compile the .cc files manually on other systems. The

 files called xxx_unittest.cc are test programs for each of the corresponding

 xxx.cc files.

 BUILDING FOR VIRTUAL PASCAL

 A script for building PCRE using Borland's C++ compiler for use with VPASCAL

 was contributed by Alexander Tokarev. Stefan Weber updated the script and added

 additional files. The following files in the distribution are for building PCRE

 for use with VP/Borland: makevp_c.txt, makevp_l.txt, makevp.bat, pcregexp.pas.

 STACK SIZE IN WINDOWS ENVIRONMENTS

 The default processor stack size of 1Mb in some Windows environments is too

 small for matching patterns that need much recursion. In particular, test 2 may

 fail because of this. Normally, running out of stack causes a crash, but there

 have been cases where the test program has just died silently. See your linker

 documentation for how to increase stack size if you experience problems. The

 Linux default of 8Mb is a reasonable choice for the stack, though even that can

 be too small for some pattern/subject combinations.

 PCRE has a compile configuration option to disable the use of stack for

 recursion so that heap is used instead. However, pattern matching is

 significantly slower when this is done. There is more about stack usage in the

 "pcrestack" documentation.

 LINKING PROGRAMS IN WINDOWS ENVIRONMENTS

 If you want to statically link a program against a PCRE library in the form of

 a non-dll .a file, you must define PCRE_STATIC before including pcre.h,

 otherwise the pcre_malloc() and pcre_free() exported functions will be declared

 __declspec(dllimport), with unwanted results.

 CALLING CONVENTIONS IN WINDOWS ENVIRONMENTS

 It is possible to compile programs to use different calling conventions using

 MSVC. Search the web for "calling conventions" for more information. To make it

 easier to change the calling convention for the exported functions in the

 PCRE library, the macro PCRE_CALL_CONVENTION is present in all the external

 definitions. It can be set externally when compiling (e.g. in CFLAGS). If it is

 not set, it defaults to empty; the default calling convention is then used

 (which is what is wanted most of the time).

 COMMENTS ABOUT WIN32 BUILDS (see also "BUILDING PCRE WITH CMAKE" below)

 There are two ways of building PCRE using the "configure, make, make install"

 paradigm on Windows systems: using MinGW or using Cygwin. These are not at all

 the same thing; they are completely different from each other. There is also

 support for building using CMake, which some users find a more straightforward

 way of building PCRE under Windows. However, the tests are not run

 automatically when CMake is used.

 The MinGW home page (http://www.mingw.org/) says this:

   MinGW: A collection of freely available and freely distributable Windows

   specific header files and import libraries combined with GNU toolsets that

   allow one to produce native Windows programs that do not rely on any

   3rd-party C runtime DLLs.

 The Cygwin home page (http://www.cygwin.com/) says this:

   Cygwin is a Linux-like environment for Windows. It consists of two parts:

   . A DLL (cygwin1.dll) which acts as a Linux API emulation layer providing

     substantial Linux API functionality

   . A collection of tools which provide Linux look and feel.

   The Cygwin DLL currently works with all recent, commercially released x86 32

   bit and 64 bit versions of Windows, with the exception of Windows CE.

 On both MinGW and Cygwin, PCRE should build correctly using:

   ./configure && make && make install

 This should create two libraries called libpcre and libpcreposix, and, if you

 have enabled building the C++ wrapper, a third one called libpcrecpp. These are

 independent libraries: when you like with libpcreposix or libpcrecpp you must

 also link with libpcre, which contains the basic functions. (Some earlier

 releases of PCRE included the basic libpcre functions in libpcreposix. This no

 longer happens.)

 A user submitted a special-purpose patch that makes it easy to create

 "pcre.dll" under mingw32 using the "msys" environment. It provides "pcre.dll"

 as a special target. If you use this target, no other files are built, and in

 particular, the pcretest and pcregrep programs are not built. An example of how

 this might be used is:

   ./configure --enable-utf --disable-cpp CFLAGS="-03 -s"; make pcre.dll

 Using Cygwin's compiler generates libraries and executables that depend on

 cygwin1.dll. If a library that is generated this way is distributed,

 cygwin1.dll has to be distributed as well. Since cygwin1.dll is under the GPL

 licence, this forces not only PCRE to be under the GPL, but also the entire

 application. A distributor who wants to keep their own code proprietary must

 purchase an appropriate Cygwin licence.

 MinGW has no such restrictions. The MinGW compiler generates a library or

 executable that can run standalone on Windows without any third party dll or

 licensing issues.

 But there is more complication:

 If a Cygwin user uses the -mno-cygwin Cygwin gcc flag, what that really does is

 to tell Cygwin's gcc to use the MinGW gcc. Cygwin's gcc is only acting as a

 front end to MinGW's gcc (if you install Cygwin's gcc, you get both Cygwin's

 gcc and MinGW's gcc). So, a user can:

 . Build native binaries by using MinGW or by getting Cygwin and using

   -mno-cygwin.

 . Build binaries that depend on cygwin1.dll by using Cygwin with the normal

   compiler flags.

 The test files that are supplied with PCRE are in Unix format, with LF

 characters as line terminators. It may be necessary to change the line

 terminators in order to get some of the tests to work. We hope to improve

 things in this area in future.

 BUILDING PCRE ON WINDOWS WITH CMAKE

 CMake is an alternative build facility that can be used instead of the

 traditional Unix "configure". CMake version 2.4.7 supports Borland makefiles,

 MinGW makefiles, MSYS makefiles, NMake makefiles, UNIX makefiles, Visual Studio

 6, Visual Studio 7, Visual Studio 8, and Watcom W8. The following instructions

 were contributed by a PCRE user.

 1.  Download CMake 2.4.7 or above from http://www.cmake.org/, install and ensure

     that cmake\bin is on your path.

 2.  Unzip (retaining folder structure) the PCRE source tree into a source

     directory such as C:\pcre.

 3.  Create a new, empty build directory: C:\pcre\build\

 4.  Run CMakeSetup from the Shell envirornment of your build tool, e.g., Msys

     for Msys/MinGW or Visual Studio Command Prompt for VC/VC++

 5.  Enter C:\pcre\pcre-xx and C:\pcre\build for the source and build

     directories, respectively

 6.  Hit the "Configure" button.

 7.  Select the particular IDE / build tool that you are using (Visual Studio,

     MSYS makefiles, MinGW makefiles, etc.)

 8.  The GUI will then list several configuration options. This is where you can

     enable UTF-8 support, etc.

 9.  Hit "Configure" again. The adjacent "OK" button should now be active.

 10. Hit "OK".

 11. The build directory should now contain a usable build system, be it a

     solution file for Visual Studio, makefiles for MinGW, etc.

 USE OF RELATIVE PATHS WITH CMAKE ON WINDOWS

 A PCRE user comments as follows:

 I thought that others may want to know the current state of

 CMAKE_USE_RELATIVE_PATHS support on Windows.

 Here it is:

 -- AdditionalIncludeDirectories is only partially modified (only the

 first path - see below)

 -- Only some of the contained file paths are modified - shown below for

 pcre.vcproj

 -- It properly modifies

 I am sure CMake people can fix that if they want to. Until then one will

 need to replace existing absolute paths in project files with relative

 paths manually (e.g. from VS) - relative to project file location. I did

 just that before being told to try CMAKE_USE_RELATIVE_PATHS. Not a big

 deal.

 AdditionalIncludeDirectories="E:\builds\pcre\build;E:\builds\pcre\pcre-7.5;"

 AdditionalIncludeDirectories=".;E:\builds\pcre\pcre-7.5;"

 RelativePath="pcre.h">

 RelativePath="pcre_chartables.c">

 RelativePath="pcre_chartables.c.rule">

 TESTING WITH RUNTEST.BAT

 1. Copy RunTest.bat into the directory where pcretest.exe has been created.

 2. Edit RunTest.bat and insert a line that indentifies the relative location of

    the pcre source, e.g.:

    set srcdir=..\pcre-7.4-RC3

 3. Run RunTest.bat from a command shell environment. Test outputs will

    automatically be compared to expected results, and discrepancies will

    identified in the console output.

 4. To test pcrecpp, run pcrecpp_unittest.exe, pcre_stringpiece_unittest.exe and

    pcre_scanner_unittest.exe.

 BUILDING UNDER WINDOWS WITH BCC5.5

 Michael Roy sent these comments about building PCRE under Windows with BCC5.5:

   Some of the core BCC libraries have a version of PCRE from 1998 built in,

   which can lead to pcre_exec() giving an erroneous PCRE_ERROR_NULL from a

   version mismatch. I'm including an easy workaround below, if you'd like to

   include it in the non-unix instructions:

   When linking a project with BCC5.5, pcre.lib must be included before any of

   the libraries cw32.lib, cw32i.lib, cw32mt.lib, and cw32mti.lib on the command

   line.

 BUILDING UNDER WINDOWS CE WITH VISUAL STUDIO 200x

 Vincent Richomme sent a zip archive of files to help with this process. They

 can be found in the file "pcre-vsbuild.zip" in the Contrib directory of the FTP

 site.

 BUILDING PCRE ON OPENVMS

 Dan Mooney sent the following comments about building PCRE on OpenVMS. They

 relate to an older version of PCRE that used fewer source files, so the exact

 commands will need changing. See the current list of source files above.

 "It was quite easy to compile and link the library. I don't have a formal

 make file but the attached file [reproduced below] contains the OpenVMS DCL

 commands I used to build the library. I had to add #define

 POSIX_MALLOC_THRESHOLD 10 to pcre.h since it was not defined anywhere.

 The library was built on:

 O/S: HP OpenVMS v7.3-1

 Compiler: Compaq C v6.5-001-48BCD

 Linker: vA13-01

 The test results did not match 100% due to the issues you mention in your

 documentation regarding isprint(), iscntrl(), isgraph() and ispunct(). I

 modified some of the character tables temporarily and was able to get the

 results to match. Tests using the fr locale did not match since I don't have

 that locale loaded. The study size was always reported to be 3 less than the

 value in the standard test output files."

 =========================

 $! This DCL procedure builds PCRE on OpenVMS

 $!

 $! I followed the instructions in the non-unix-use file in the distribution.

 $!

 $ COMPILE == "CC/LIST/NOMEMBER_ALIGNMENT/PREFIX_LIBRARY_ENTRIES=ALL_ENTRIES

 $ COMPILE DFTABLES.C

 $ LINK/EXE=DFTABLES.EXE DFTABLES.OBJ

 $ RUN DFTABLES.EXE/OUTPUT=CHARTABLES.C

 $ COMPILE MAKETABLES.C

 $ COMPILE GET.C

 $ COMPILE STUDY.C

 $! I had to set POSIX_MALLOC_THRESHOLD to 10 in PCRE.H since the symbol

 $! did not seem to be defined anywhere.

 $! I edited pcre.h and added #DEFINE SUPPORT_UTF8 to enable UTF8 support.

 $ COMPILE PCRE.C

 $ LIB/CREATE PCRE MAKETABLES.OBJ, GET.OBJ, STUDY.OBJ, PCRE.OBJ

 $! I had to set POSIX_MALLOC_THRESHOLD to 10 in PCRE.H since the symbol

 $! did not seem to be defined anywhere.

 $ COMPILE PCREPOSIX.C

 $ LIB/CREATE PCREPOSIX PCREPOSIX.OBJ

 $ COMPILE PCRETEST.C

 $ LINK/EXE=PCRETEST.EXE PCRETEST.OBJ, PCRE/LIB, PCREPOSIX/LIB

 $! C programs that want access to command line arguments must be

 $! defined as a symbol

 $ PCRETEST :== "$ SYS$ROADSUSERS:[DMOONEY.REGEXP]PCRETEST.EXE"

 $! Arguments must be enclosed in quotes.

 $ PCRETEST "-C"

 $! Test results:

 $!

 $!   The test results did not match 100%. The functions isprint(), iscntrl(),

 $!   isgraph() and ispunct() on OpenVMS must not produce the same results

 $!   as the system that built the test output files provided with the

 $!   distribution.

 $!

 $!   The study size did not match and was always 3 less on OpenVMS.

 $!

 $!   Locale could not be set to fr

 $!

 =========================

 Last Updated: 17 March 2009

 ****