perlwin32 - Perl under Windows |
perlwin32 - Perl under Windows
These are instructions for building Perl under Windows 9x/NT/2000/XP on the Intel x86 and Itanium architectures.
Before you start, you should glance through the README file found in the top-level directory to which the Perl distribution was extracted. Make sure you read and understand the terms under which this software is being distributed.
Also make sure you read BUGS AND CAVEATS below for the known limitations of this port.
The INSTALL file in the perl top-level has much information that is only relevant to people building Perl on Unix-like systems. In particular, you can safely ignore any information that talks about ``Configure''.
You may also want to look at two other options for building a perl that will work on Windows NT: the README.cygwin and README.os2 files, each of which give a different set of rules to build a Perl that will work on Win32 platforms. Those two methods will probably enable you to build a more Unix-compatible perl, but you will also need to download and use various other build-time and run-time support software described in those files.
This set of instructions is meant to describe a so-called ``native'' port of Perl to Win32 platforms. This includes both 32-bit and 64-bit Windows operating systems. The resulting Perl requires no additional software to run (other than what came with your operating system). Currently, this port is capable of using one of the following compilers on the Intel x86 architecture:
Borland C++ version 5.02 or later Microsoft Visual C++ version 4.2 or later Mingw32 with GCC version 2.95.2 or better
The last of these is a high quality freeware compiler. Support for it is still experimental. (Older versions of GCC are known not to work.)
This port can also be built on the Intel IA64 using:
Microsoft Platform SDK Nov 2001 (64-bit compiler and tools)
The MS Platform SDK can be downloaded from http://www.microsoft.com/.
This port fully supports MakeMaker (the set of modules that is used to build extensions to perl). Therefore, you should be able to build and install most extensions found in the CPAN sites. See Usage Hints for Perl on Win32 below for general hints about this.
dmake is a freely available make that has very nice macro features and parallelability.
A port of dmake for Windows is available from:
http://www.cpan.org/authors/id/GSAR/dmake-4.1pl1-win32.zip
(This is a fixed version of the original dmake sources obtained from http://www.wticorp.com/ As of version 4.1PL1, the original sources did not build as shipped and had various other problems. A patch is included in the above fixed version.)
Fetch and install dmake somewhere on your path (follow the instructions in the README.NOW file).
There exists a minor coexistence problem with dmake and Borland C++ compilers. Namely, if a distribution has C files named with mixed case letters, they will be compiled into appropriate .obj-files named with all lowercase letters, and every time dmake is invoked to bring files up to date, it will try to recompile such files again. For example, Tk distribution has a lot of such files, resulting in needless recompiles every time dmake is invoked. To avoid this, you may use the script ``sync_ext.pl'' after a successful build. It is available in the win32 subdirectory of the Perl source distribution.
The nmake Makefile also has known incompatibilities with the ``command.com'' shell that comes with Windows 9x. You will need to use dmake and makefile.mk to build under Windows 9x.
The surest way to build it is on Windows NT/2000/XP, using the cmd shell.
Make sure the path to the build directory does not contain spaces. The build usually works in this circumstance, but some tests will fail.
See Make above.
You can also use dmake to build using Visual C++; provided, however, you set OSRELEASE to ``microsft'' (or whatever the directory name under which the Visual C dmake configuration lives) in your environment and edit win32/config.vc to change ``make=nmake'' into ``make=dmake''. The latter step is only essential if you want to use dmake as your default make for building extensions using MakeMaker.
http://sourceforge.net/projects/mingw
Perl compiles with earlier releases of gcc (2.95 and up) that can be downloaded from the same place. If you use gcc-3.2, comment out the line:
USE_GCC_V3_2 *= define
in win32\makefile.mk
You also need dmake. See Make above on how to get it.
Make sure you install the binaries that work with MSVCRT.DLL as indicated in the README for the GCC bundle. You may need to set up a few environment variables (usually ran from a batch file).
There are a couple of problems with the version of gcc-2.95.2-msvcrt.exe released 7 November 1999:
A potentially simpler to install (but probably soon-to-be-outdated) bundle of the above package with the mentioned fixes already applied is available here:
http://downloads.ActiveState.com/pub/staff/gsar/gcc-2.95.2-msvcrt.zip ftp://ftp.ActiveState.com/pub/staff/gsar/gcc-2.95.2-msvcrt.zip
You will have to make sure that CCTYPE is set correctly and that CCHOME points to wherever you installed your compiler.
The default value for CCHOME in the makefiles for Visual C++ may not be correct for some versions. Make sure the default exists and is valid.
If you have either the source or a library that contains des_fcrypt(),
enable the appropriate option in the makefile. des_fcrypt()
is not
bundled with the distribution due to US Government restrictions
on the export of cryptographic software. Nevertheless, this routine
is part of the ``libdes'' library (written by Eric Young) which is widely
available worldwide, usually along with SSLeay ( for example,
ftp://ftp.funet.fi/pub/crypt/mirrors/dsi/libdes/ ). Set CRYPT_SRC to the
name of the file that implements des_fcrypt(). Alternatively, if
you have built a library that contains des_fcrypt(), you can set
CRYPT_LIB to point to the library name. The location above contains
many versions of the ``libdes'' library, all with slightly different
implementations of des_fcrypt(). Older versions have a single,
self-contained file (fcrypt.c) that implements crypt(), so they may be
easier to use. A patch against the fcrypt.c found in libdes-3.06 is
in des_fcrypt.patch.
An easier alternative may be to get the pre-patched and ready-to-use fcrypt.c that can be found here:
http://downloads.ActiveState.com/pub/staff/gsar/fcrypt.c ftp://ftp.ActiveState.com/pub/staff/gsar/fcrypt.c
Perl will also build without des_fcrypt(), but the crypt()
builtin will
fail at run time.
Be sure to read the instructions near the top of the makefiles carefully.
Type ``dmake'' (or ``nmake'' if you are using that make).This should build everything. Specifically, it will create perl.exe, perl58.dll at the perl toplevel, and various other extension dll's under the lib\auto directory. If the build fails for any reason, make sure you have done the previous steps correctly.
Type ``dmake test'' (or ``nmake test''). This will run most of the tests from the testsuite (many tests will be skipped).
There should be no test failures when running under Windows NT/2000/XP. Many tests will fail under Windows 9x due to the inferior command shell.
Some test failures may occur if you use a command shell other than the native ``cmd.exe'', or if you are building from a path that contains spaces. So don't do that.
If you are running the tests from a emacs shell window, you may see failures in op/stat.t. Run ``dmake test-notty'' in that case.
If you're using the Borland compiler, you may see a failure in op/taint.t arising from the inability to find the Borland Runtime DLLs on the system default path. You will need to copy the DLLs reported by the messages from where Borland chose to install it, into the Windows system directory (usually somewhere like C:\WINNT\SYSTEM32) and rerun the test.
If you're using Borland compiler versions 5.2 and below, you may run into problems finding the correct header files when building extensions. For example, building the ``Tk'' extension may fail because both perl and Tk contain a header file called ``patchlevel.h''. The latest Borland compiler (v5.5) is free of this misbehaviour, and it even supports an option -VI- for backward (bugward) compatibility for using the old Borland search algorithm to locate header files.
If you run the tests on a FAT partition, you may see some failures for
link()
related tests:
Failed Test Stat Wstat Total Fail Failed List
../ext/IO/lib/IO/t/io_dup.t 6 4 66.67% 2-5 ../lib/File/Temp/t/mktemp.t 9 1 11.11% 2 ../lib/File/Temp/t/posix.t 7 1 14.29% 3 ../lib/File/Temp/t/security.t 13 1 7.69% 2 ../lib/File/Temp/t/tempfile.t 20 2 10.00% 2 4 comp/multiline.t 6 2 33.33% 5-6 io/dup.t 8 6 75.00% 2-7 op/write.t 47 7 14.89% 1-3 6 9-11
Testing on NTFS avoids these errors.
Furthermore, you should make sure that during make test
you do not
have any GNU tool packages in your path: some toolkits like Unixutils
include some tools (type
for instance) which override the Windows
ones and makes tests fail. Remove them from your path while testing to
avoid these errors.
Please report any other failures as described under BUGS AND CAVEATS.
Type ``dmake install'' (or ``nmake install''). This will put the newly
built perl and the libraries under whatever INST_TOP
points to in the
Makefile. It will also install the pod documentation under
$INST_TOP\$VERSION\lib\pod
and HTML versions of the same under
$INST_TOP\$VERSION\lib\pod\html
. To use the Perl you just installed,
you will need to add two components to your PATH environment variable,
$INST_TOP\$VERSION\bin
and $INST_TOP\$VERSION\bin\$ARCHNAME
.
For example:
set PATH c:\perl\5.6.0\bin;c:\perl\5.6.0\bin\MSWin32-x86;%PATH%
If you opt to comment out INST_VER and INST_ARCH in the makefiles, the installation structure is much simpler. In that case, it will be sufficient to add a single entry to the path, for instance:
set PATH c:\perl\bin;%PATH%
If you put extensions in unusual places, you can set PERL5LIB to a list of paths separated by semicolons where you want perl to look for libraries. Look for descriptions of other environment variables you can set in the perlrun manpage.
You can also control the shell that perl uses to run system()
and
backtick commands via PERL5SHELL. See the perlrun manpage.
Perl does not depend on the registry, but it can look up certain default
values if you choose to put them there. Perl attempts to read entries from
HKEY_CURRENT_USER\Software\Perl
and HKEY_LOCAL_MACHINE\Software\Perl
.
Entries in the former override entries in the latter. One or more of the
following entries (of type REG_SZ or REG_EXPAND_SZ) may be set:
lib-$] version-specific standard library path to add to @INC lib standard library path to add to @INC sitelib-$] version-specific site library path to add to @INC sitelib site library path to add to @INC vendorlib-$] version-specific vendor library path to add to @INC vendorlib vendor library path to add to @INC PERL* fallback for all %ENV lookups that begin with "PERL"
Note the $]
in the above is not literal. Substitute whatever version
of perl you want to honor that entry, e.g. 5.6.0
. Paths must be
separated with semicolons, as usual on win32.
If you want perl to use globbing that emulates the quirks of DOS
filename conventions, you might want to consider using File::DosGlob
to override the internal glob()
implementation. See the File::DosGlob manpage for
details.
The crucial thing to understand about the Windows environment is that the command line you type in is processed twice before Perl sees it. First, your command shell (usually CMD.EXE on Windows NT, and COMMAND.COM on Windows 9x) preprocesses the command line, to handle redirection, environment variable expansion, and location of the executable to run. Then, the perl executable splits the remaining command line into individual arguments, using the C runtime library upon which Perl was built.
It is particularly important to note that neither the shell nor the C runtime do any wildcard expansions of command-line arguments (so wildcards need not be quoted). Also, the quoting behaviours of the shell and the C runtime are rudimentary at best (and may, if you are using a non-standard shell, be inconsistent). The only (useful) quote character is the double quote (``). It can be used to protect spaces and other special characters in arguments.
The Windows NT documentation has almost no description of how the quoting rules are implemented, but here are some general observations based on experiments: The C runtime breaks arguments at spaces and passes them to programs in argc/argv. Double quotes can be used to prevent arguments with spaces in them from being split up. You can put a double quote in an argument by escaping it with a backslash and enclosing the whole argument within double quotes. The backslash and the pair of double quotes surrounding the argument will be stripped by the C runtime.
The file redirection characters ``<'', ``>'', and ``|'' can be quoted by double quotes (although there are suggestions that this may not always be true). Single quotes are not treated as quotes by the shell or the C runtime, they don't get stripped by the shell (just to make this type of quoting completely useless). The caret ``^'' has also been observed to behave as a quoting character, but this appears to be a shell feature, and the caret is not stripped from the command line, so Perl still sees it (and the C runtime phase does not treat the caret as a quote character).
Here are some examples of usage of the ``cmd'' shell:
This prints two doublequotes:
perl -e "print '\"\"' "
This does the same:
perl -e "print \"\\\"\\\"\" "
This prints ``bar'' and writes ``foo'' to the file ``blurch'':
perl -e "print 'foo'; print STDERR 'bar'" > blurch
This prints ``foo'' (``bar'' disappears into nowhereland):
perl -e "print 'foo'; print STDERR 'bar'" 2> nul
This prints ``bar'' and writes ``foo'' into the file ``blurch'':
perl -e "print 'foo'; print STDERR 'bar'" 1> blurch
This pipes ``foo'' to the ``less'' pager and prints ``bar'' on the console:
perl -e "print 'foo'; print STDERR 'bar'" | less
This pipes ``foo\nbar\n'' to the less pager:
perl -le "print 'foo'; print STDERR 'bar'" 2>&1 | less
This pipes ``foo'' to the pager and writes ``bar'' in the file ``blurch'':
perl -e "print 'foo'; print STDERR 'bar'" 2> blurch | less
Discovering the usefulness of the ``command.com'' shell on Windows 9x is left as an exercise to the reader :)
One particularly pernicious problem with the 4NT command shell for Windows NT is that it (nearly) always treats a % character as indicating that environment variable expansion is needed. Under this shell, it is therefore important to always double any % characters which you want Perl to see (for example, for hash variables), even when they are quoted.
Note that not all of the extensions available from CPAN may work in the Win32 environment; you should check the information at http://testers.cpan.org/ before investing too much effort into porting modules that don't readily build.
Most extensions (whether they require a C compiler or not) can be built, tested and installed with the standard mantra:
perl Makefile.PL $MAKE $MAKE test $MAKE install
where $MAKE is whatever 'make' program you have configured perl to use. Use ``perl -V:make'' to find out what this is. Some extensions may not provide a testsuite (so ``$MAKE test'' may not do anything or fail), but most serious ones do.
It is important that you use a supported 'make' program, and ensure Config.pm knows about it. If you don't have nmake, you can either get dmake from the location mentioned earlier or get an old version of nmake reportedly available from:
ftp://ftp.microsoft.com/Softlib/MSLFILES/nmake15.exe
Another option is to use the make written in Perl, available from CPAN.
http://www.cpan.org/modules/by-module/Make/
You may also use dmake. See Make above on how to get it.
Note that MakeMaker actually emits makefiles with different syntax depending on what 'make' it thinks you are using. Therefore, it is important that one of the following values appears in Config.pm:
make='nmake' # MakeMaker emits nmake syntax make='dmake' # MakeMaker emits dmake syntax any other value # MakeMaker emits generic make syntax (e.g GNU make, or Perl make)
If the value doesn't match the 'make' program you want to use, edit Config.pm to fix it.
If a module implements XSUBs, you will need one of the supported C compilers. You must make sure you have set up the environment for the compiler for command-line compilation.
If a module does not build for some reason, look carefully for why it failed, and report problems to the module author. If it looks like the extension building support is at fault, report that with full details of how the build failed using the perlbug utility.
Instead, the following solution works rather well. The nice things about it are 1) you can start using it right away; 2) it is more powerful, because it will do the right thing with a pattern like */*/*.c; 3) you can decide whether you do/don't want to use it; and 4) you can extend the method to add any customizations (or even entirely different kinds of wildcard expansion).
C:\> copy con c:\perl\lib\Wild.pm # Wild.pm - emulate shell @ARGV expansion on shells that don't use File::DosGlob; @ARGV = map { my @g = File::DosGlob::glob($_) if /[*?]/; @g ? @g : $_; } @ARGV; 1; ^Z C:\> set PERL5OPT=-MWild C:\> perl -le "for (@ARGV) { print }" */*/perl*.c p4view/perl/perl.c p4view/perl/perlio.c p4view/perl/perly.c perl5.005/win32/perlglob.c perl5.005/win32/perllib.c perl5.005/win32/perlglob.c perl5.005/win32/perllib.c perl5.005/win32/perlglob.c perl5.005/win32/perllib.c
Note there are two distinct steps there: 1) You'll have to create Wild.pm and put it in your perl lib directory. 2) You'll need to set the PERL5OPT environment variable. If you want argv expansion to be the default, just set PERL5OPT in your default startup environment.
If you are using the Visual C compiler, you can get the C runtime's command line wildcard expansion built into perl binary. The resulting binary will always expand unquoted command lines, which may not be what you want if you use a shell that does that for you. The expansion done is also somewhat less powerful than the approach suggested above.
To ensure smooth transitioning of existing code that uses the ActiveState port, there is a bundle of Win32 extensions that contains all of the ActiveState extensions and most other Win32 extensions from CPAN in source form, along with many added bugfixes, and with MakeMaker support. This bundle is available at:
http://www.cpan.org/authors/id/GSAR/libwin32-0.18.zip
See the README in that distribution for building and installation instructions. Look for later versions that may be available at the same location.
The LLP64 data model is different from the LP64 data model that is the
norm on 64-bit Unix platforms. In the former, int
and long
are
both 32-bit data types, while pointers are 64 bits wide. In addition,
there is a separate 64-bit wide integral type, __int64
. In contrast,
the LP64 data model that is pervasive on Unix platforms provides int
as the 32-bit type, while both the long
type and pointers are of
64-bit precision. Note that both models provide for 64-bits of
addressability.
64-bit Windows running on Itanium is capable of running 32-bit x86 binaries transparently. This means that you could use a 32-bit build of Perl on a 64-bit system. Given this, why would one want to build a 64-bit build of Perl? Here are some reasons why you would bother:
Perl scripts on UNIX use the ``#!'' (a.k.a ``shebang'') line to indicate to the OS that it should execute the file using perl. Win32 has no comparable means to indicate arbitrary files are executables.
Instead, all available methods to execute plain text files on Win32 rely on the file ``extension''. There are three methods to use this to execute perl scripts:
pl2bat foo.pl
will create the file ``FOO.BAT''. Note ``pl2bat'' strips any .pl suffix and adds a .bat suffix to the generated file.
If you use the 4DOS/NT or similar command shell, note that ``pl2bat'' uses the ``%*'' variable in the generated batch file to refer to all the command line arguments, so you may need to make sure that construct works in batch files. As of this writing, 4DOS/NT users will need a ``ParameterChar = *'' statement in their 4NT.INI file or will need to execute ``setdos /p*'' in the 4DOS/NT startup file to enable this to work.
Using ``pl2bat'' has a few problems: the file name gets changed, so scripts that rely on$0
to find what they must do may not
run properly; running ``pl2bat'' replicates the contents of the
original script, and so this process can be maintenance intensive
if the originals get updated often. A different approach that
avoids both problems is possible.
A script called ``runperl.bat'' is available that can be copied to any filename (along with the .bat suffix). For example, if you call it ``foo.bat'', it will run the file ``foo'' when it is executed. Since you can run batch files on Win32 platforms simply by typing the name (without the extension), this effectively runs the file ``foo'', when you type either ``foo'' or ``foo.bat''. With this method, ``foo.bat'' can even be in a different location than the file ``foo'', as long as ``foo'' is available somewhere on the PATH. If your scripts are on a filesystem that allows symbolic links, you can even avoid copying ``runperl.bat''.
Here's a diversion: copy ``runperl.bat'' to ``runperl'', and type ``runperl''. Explain the observed behavior, or lack thereof. :) Hint: .gnidnats llits er'uoy fi ,``lrepnur'' eteled :tniH
perldoc
is also a useful tool for browsing information contained
in the documentation, especially in conjunction with a pager
like less
(recent versions of which have Win32 support). You may
have to set the PAGER environment variable to use a specific pager.
``perldoc -f foo'' will print information about the perl operator
``foo''.
One common mistake when using this port with a GUI library like Tk
is assuming that Perl's normal behavior of opening a command-line
window will go away. This isn't the case. If you want to start a copy
of perl
without opening a command-line window, use the wperl
executable built during the installation process. Usage is exactly
the same as normal perl
on Win32, except that options like -h
don't work (since they need a command-line window to print to).
If you find bugs in perl, you can run perlbug
to create a
bug report (you may have to send it manually if perlbug
cannot
find a mailer on your system).
Norton AntiVirus interferes with the build process, particularly if set to ``AutoProtect, All Files, when Opened''. Unlike large applications the perl build process opens and modifies a lot of files. Having the the AntiVirus scan each and every one slows build the process significantly. Worse, with PERLIO=stdio the build process fails with peculiar messages as the virus checker interacts badly with miniperl.exe writing configure files (it seems to either catch file part written and treat it as suspicious, or virus checker may have it ``locked'' in a way which inhibits miniperl updating it). The build does complete with
set PERLIO=perlio
but that may be just luck. Other AntiVirus software may have similar issues.
Some of the built-in functions do not act exactly as documented in the perlfunc manpage, and a few are not implemented at all. To avoid surprises, particularly if you have had prior exposure to Perl in other operating environments or if you intend to write code that will be portable to other environments. See the perlport manpage for a reasonably definitive list of these differences.
Not all extensions available from CPAN may build or work properly in the Win32 environment. See Building Extensions.
Most socket()
related calls are supported, but they may not
behave as on Unix platforms. See the perlport manpage for the full list.
Signal handling may not behave as on Unix platforms (where it
doesn't exactly ``behave'', either :). For instance, calling die()
or exit()
from signal handlers will cause an exception, since most
implementations of signal()
on Win32 are severely crippled.
Thus, signals may work only for simple things like setting a flag
variable in the handler. Using signals under this port should
currently be considered unsupported.
Please send detailed descriptions of any problems and solutions that
you may find to <perlbug@perl.com>, along with the output produced
by perl -V
.
This document is maintained by Gurusamy Sarathy.
This port was originally contributed by Gary Ng around 5.003_24, and borrowed from the Hip Communications port that was available at the time. Various people have made numerous and sundry hacks since then.
Borland support was added in 5.004_01 (Gurusamy Sarathy).
GCC/mingw32 support was added in 5.005 (Nick Ing-Simmons).
Support for PERL_OBJECT was added in 5.005 (ActiveState Tool Corp).
Support for fork()
emulation was added in 5.6 (ActiveState Tool Corp).
Win9x support was added in 5.6 (Benjamin Stuhl).
Support for 64-bit Windows added in 5.8 (ActiveState Corp).
Last updated: 20 April 2002
perlwin32 - Perl under Windows |