Skip to content

Latest commit

 

History

History
2867 lines (2129 loc) · 112 KB

INSTALL

File metadata and controls

2867 lines (2129 loc) · 112 KB

NAME

INSTALL - Build and Installation guide for perl 5.

SYNOPSIS

First, make sure you have an up-to-date version of Perl. If you didn't get your Perl source from CPAN, check the latest version at https://www.cpan.org/src/. Perl uses a version scheme where even-numbered subreleases (like 5.8.x and 5.10.x) are stable maintenance releases and odd-numbered subreleases (like 5.7.x and 5.9.x) are unstable development releases. Development releases should not be used in production environments. Fixes and new features are first carefully tested in development releases and only if they prove themselves to be worthy will they be migrated to the maintenance releases.

The basic steps to build and install perl 5 on a Unix system with all the defaults are to run, from a freshly unpacked source tree:

sh Configure -de
make
make test
make install

Each of these is explained in further detail below.

The above commands will install Perl to /usr/local (or some other platform-specific directory -- see the appropriate file in hints/.) If that's not okay with you, you can run Configure interactively, by just typing "sh Configure" (without the -de args). You can also specify any prefix location by adding "-Dprefix='/some/dir'" to Configure's args. To explicitly name the perl binary, use the command "make install PERLNAME=myperl".

Building perl from source requires an ANSI compliant C compiler. C89 with a minimal subset of C99 features is required. Some other features available in C99 will be probed for and used when found.

These options, and many more, are explained in further detail below.

If you're building perl from a git repository, you should also consult the documentation in pod/perlgit.pod for information on that special circumstance.

If you have problems, corrections, or questions, please see "Reporting Problems" below.

For information on what's new in this release, see the pod/perldelta.pod file. For more information about how to find more specific detail about changes, see the Changes file.

DESCRIPTION

This document is written in pod format as an easy way to indicate its structure. The pod format is described in pod/perlpod.pod, but you can read it as is with any pager or editor. Headings and items are marked by lines beginning with '='. The other mark-up used is

B<text>     embolden text, used for switches, programs or commands
C<code> literal code
L<name>     A link (cross reference) to name
F<file>     A filename

Although most of the defaults are probably fine for most users, you should probably at least skim through this document before proceeding.

In addition to this file, check if there is a README file specific to your operating system, since it may provide additional or different instructions for building Perl. If there is a hint file for your system (in the hints/ directory) you might also want to read it for even more information.

For additional information about porting Perl, see the section on "Porting information" below, and look at the files in the Porting/ directory.

PRELIMINARIES

Changes and Incompatibilities

Please see pod/perldelta.pod for a description of the changes and potential incompatibilities introduced with this release. A few of the most important issues are listed below, but you should refer to pod/perldelta.pod for more detailed information.

Compatibility with earlier versions

WARNING: This version is not binary compatible with earlier versions of Perl. If you have built extensions (i.e. modules that include C code) using an earlier version of Perl, you will need to rebuild and reinstall those extensions.

Pure perl modules without XS or C code should continue to work fine without reinstallation. See the discussion below on "Coexistence with earlier versions of perl 5" for more details.

The standard extensions supplied with Perl will be handled automatically.

On a related issue, old modules may possibly be affected by the changes in the Perl language in the current release. We try hard to make new features "opt-in", such that existing code will work unchanged, and attempt to identify where bug fixes might expose code which was relying on incorrect interpreter behaviour. Please see pod/perldelta.pod for a description of what's changed between this and the previous release. If you are upgrading from an earlier release, please also check the perldeltas describing changes for the intermediate releases, to get a full picture of what changes might affect your installation.

See your installed copy of the perllocal.pod file for a (possibly incomplete) list of locally installed modules. Also see the CPAN module's autobundle function for one way to make a "bundle" of your currently installed modules.

C99

With 5.36 we changed our C compiler baseline requirement from "ANSI C89" to "C89 plus some specific C99 features". We have been using C99 features optionally for some time - we now additionally rely on a few in the core C code and installed headers, which we know work on all supported compilers on all platforms we target. Because earlier versions of Perl still compile with strictly ANSI C89 compilers and there are still a few installations in the wild which use these very old compilers, XS code that targets earlier versions of Perl must not rely on C99 features - that includes XS code in this distribution that is dual life on CPAN. To test that XS code can build on such compilers, some authors configure their perl builds with compiler flags to warn or raise errors on C99 specific features, most often for mixed declarations and code. This obviously will not work if you attempt it for this release - it will not build. However, XS authors should be aware that this means

  • If you change the C compiler flags in your Makefile.PL or similar to add such warnings or errors, you must now only do it for 5.35.4 or earlier.

  • Do not rely on now being able to use C99 features in your XS code, even for platforms with C99 compilers - some installations of earlier versions of perl are deliberately configured to enforce C89 standards so that locally authored extension code conforms to them. If you choose to require C99 for your code, that's fine, but do so knowing that if you distribute it, some installations of perl are configured to fault C99.

Run Configure

Configure will figure out various things about your system. Some things Configure will figure out for itself, other things it will ask you about. To accept the default, just press RETURN. The default is almost always okay. It is normal for some things to be "NOT found", since Configure often searches for many different ways of performing the same function.

At any Configure prompt, you can type &-d and Configure will use the defaults from then on.

After it runs, Configure will perform variable substitution on all the *.SH files and offer to run make depend.

The results of a Configure run are stored in the config.sh and Policy.sh files.

Common Configure options

Configure supports a number of useful options. Run

Configure -h

to get a listing.

Many Configure switches are expressed as key=value pairs, for example:

-Dcc=clang

Sometimes the value to be supplied for a switch is a string which contains spaces. In that case, the value needs to be quoted so as to delimit that "shell word" from any following switch. Example:

sh ./Configure -des \
    -Doptimize="-O2 -pipe -fstack-protector -fno-strict-aliasing" \
    -Dusedevel

Once Configure has run, you will be able to access configuration data via entries in the file config.sh.

config_arg0='./Configure'
config_args='-des -Doptimize=-O2 -pipe -fstack-protector -fno-strict-aliasing -Dusedevel'
config_argc=3
config_arg1='-des'
config_arg2='-Doptimize=-O2 -pipe -fstack-protector -fno-strict-aliasing'
config_arg3='-Dusedevel'

See the Porting/Glossary file for a complete list of Configure variables you can set and their definitions.

C compiler

To compile with gcc, if it's not the default compiler on your system, you should run

sh Configure -Dcc=gcc

This is the preferred way to specify gcc (or any another alternative compiler) so that the hints files can set appropriate defaults.

Installation prefix

By default, for most systems, perl will be installed in /usr/local/{bin, lib, man}. (See "Installation Directories" and "Coexistence with earlier versions of perl 5" below for further details.)

You can specify a different 'prefix' for the default installation directory when Configure prompts you, or by using the Configure command line option -Dprefix='/some/directory', e.g.

sh Configure -Dprefix=/opt/perl

If your prefix contains the string "perl", then the suggested directory structure is simplified. For example, if you use prefix=/opt/perl, then Configure will suggest /opt/perl/lib instead of /opt/perl/lib/perl5/. Again, see "Installation Directories" below for more details. Do not include a trailing slash, (i.e. /opt/perl/) or you may experience odd test failures.

NOTE: You must not specify an installation directory that is the same as or below your perl source directory. If you do, installperl will attempt infinite recursion.

/usr/bin/perl

It may seem obvious, but Perl is useful only when users can easily find it. It's often a good idea to have both /usr/bin/perl and /usr/local/bin/perl be symlinks to the actual binary. Be especially careful, however, not to overwrite a version of perl supplied by your vendor unless you are sure you know what you are doing. If you insist on replacing your vendor's perl, useful information on how it was configured may be found with

perl -V:config_args

(Check the output carefully, however, since this doesn't preserve spaces in arguments to Configure. For that, you have to look carefully at config_arg1, config_arg2, etc.)

By default, Configure will not try to link /usr/bin/perl to the current version of perl. You can turn on that behavior by running

Configure -Dinstallusrbinperl

or by answering 'yes' to the appropriate Configure prompt.

In any case, system administrators are strongly encouraged to put (symlinks to) perl and its accompanying utilities, such as perldoc, into a directory typically found along a user's PATH, or in another obvious and convenient place.

Building a development release

For development releases (odd subreleases, like 5.9.x) if you want to use Configure -d, you will also need to supply -Dusedevel to Configure, because the default answer to the question "do you really want to Configure a development version?" is "no". The -Dusedevel skips that sanity check.

If you are willing to accept all the defaults, and you want terse output, you can run

sh Configure -des

Altering Configure variables for C compiler switches etc.

For most users, most of the Configure defaults are fine, or can easily be set on the Configure command line. However, if Configure doesn't have an option to do what you want, you can change Configure variables after the platform hints have been run by using Configure's -A switch. For example, here's how to add a couple of extra flags to C compiler invocations:

sh Configure -Accflags="-DPERL_EXTERNAL_GLOB -DNO_HASH_SEED"

To clarify, those ccflags values are not Configure options; if passed to Configure directly, they won't do anything useful (they will define a variable in config.sh, but without taking any action based upon it). But when passed to the compiler, those flags will activate #ifdefd code.

For more help on Configure switches, run

sh Configure -h

Major Configure-time Build Options

There are several different ways to Configure and build perl for your system. For most users, the defaults are sensible and will work. Some users, however, may wish to further customize perl. Here are some of the main things you can change.

Threads

On some platforms, perl can be compiled with support for threads. To enable this, run

sh Configure -Dusethreads

The default is to compile without thread support.

Perl used to have two different internal threads implementations. The current model (available internally since 5.6, and as a user-level module since 5.8) is called interpreter-based implementation (ithreads), with one interpreter per thread, and explicit sharing of data. The (deprecated) 5.005 version (5005threads) was removed for release 5.10.

The 'threads' module is for use with the ithreads implementation. The 'Thread' module emulates the old 5005threads interface on top of the current ithreads model.

When using threads, perl uses a dynamically-sized buffer for some of the thread-safe library calls, such as those in the getpw*() family. This buffer starts small, but it will keep growing until the result fits. To get a fixed upper limit, you should compile Perl with PERL_REENTRANT_MAXSIZE defined to be the number of bytes you want. One way to do this is to run Configure with -Accflags=-DPERL_REENTRANT_MAXSIZE=65536.

Large file support

Since Perl 5.6.0, Perl has supported large files (files larger than 2 gigabytes), and in many common platforms like Linux or Solaris this support is on by default.

This is both good and bad. It is good in that you can use large files, seek(), stat(), and -s them. It is bad in that if you are interfacing Perl using some extension, the components you are connecting to must also be large file aware: if Perl thinks files can be large but the other parts of the software puzzle do not understand the concept, bad things will happen.

There's also one known limitation with the current large files implementation: unless you also have 64-bit integers (see the next section), you cannot use the printf/sprintf non-decimal integer formats like %x to print filesizes. You can use %d, though.

If you want to compile perl without large file support, use

sh Configure -Uuselargefiles

64 bit support

If your platform does not run natively at 64 bits, but can simulate them with compiler flags and/or long long or int64_t, you can build a perl that uses 64 bits.

There are actually two modes of 64-bitness: the first one is achieved using Configure -Duse64bitint and the second one using Configure -Duse64bitall. The difference is that the first one is minimal and the second one maximal. The first works in more places than the second.

The use64bitint option does only as much as is required to get 64-bit integers into Perl (this may mean, for example, using "long longs") while your memory may still be limited to 2 gigabytes (because your pointers could still be 32-bit). Note that the name 64bitint does not imply that your C compiler will be using 64-bit ints (it might, but it doesn't have to). The use64bitint simply means that you will be able to have 64 bit-wide scalar values.

The use64bitall option goes all the way by attempting to switch integers (if it can), longs (and pointers) to being 64-bit. This may create an even more binary incompatible Perl than -Duse64bitint: the resulting executable may not run at all in a 32-bit box, or you may have to reboot/reconfigure/rebuild your operating system to be 64-bit aware.

Natively 64-bit systems need neither -Duse64bitint nor -Duse64bitall. On these systems, it might be the default compilation mode, and there is currently no guarantee that passing no use64bitall option to the Configure process will build a 32bit perl. Implementing -Duse32bit* options is planned for a future release of perl.

Long doubles

In some systems you may be able to use long doubles to enhance the range and precision of your double precision floating point numbers (that is, Perl's numbers). Use Configure -Duselongdouble to enable this support (if it is available).

Note that the exact format and range of long doubles varies: the most common is the x86 80-bit (64 bits of mantissa) format, but there are others, with different mantissa and exponent ranges.

"more bits"

You can "Configure -Dusemorebits" to turn on both the 64-bit support and the long double support.

quadmath

One option for more precision is that gcc 4.6 and later have a library called quadmath, which implements the IEEE 754 quadruple precision (128-bit, 113 bits of mantissa) floating point numbers. The library works at least on x86 and ia64 platforms. It may be part of your gcc installation, or you may need to install it separately.

With "Configure -Dusequadmath" you can try enabling its use, but note the compiler dependency, you may need to also add "-Dcc=...". At C level the type is called __float128 (note, not "long double"), but Perl source knows it as NV. (This is not "long doubles".)

Taint Support

Traditional perl has provided a security mechanism based on marking input data as untrusted unless it has been validated by a regex. This mechanism is called tainting and is enabled with the -T or -t options on the command line. This support has a performance cost on all code executed. It is possible to disable this support by providing the setting -Accflags=-DNO_TAINT_SUPPORT or the setting -Accflags=-DSILENT_NO_TAINT_SUPPORT to Configure.

The former option NO_TAINT_SUPPORT is more secure and disables taint support while making the use of the -T or -t options which normally enable taint support into an untrappable exception to ensure that no-one uses them while expecting taint checks to run. With this build mode there can be no confusion if a script supports taint or not.

The latter option, SILENT_NO_TAINT_SUPPORT is less secure but more flexible in that it silently disables taint support while ignoring the -T and -t command line options entirely. In this build mode it is possible to run a script with the -T or -t option and expect taint support but not get it, so this mode is only recommended to people who really know what they are doing.

Both modes improve performance although the amount depends on your workload. It is not unreasonable to expect a 5%-10% improvement in performance by using one of these options. At this time this build mode is supported but not recommended, and if you chose to use such a perl you do so at your own risk. It is possible that there may be interoperability issues with CPAN distributions as many distributions are unaware of this build mode and will expect test scripts to pass or to detect unvalidated data when run under the -T or -t.

Algorithmic Complexity Attacks on Hashes

Perl 5.18 reworked the measures used to secure its hash function from algorithmic complexity attacks. By default it will build with all of these measures enabled along with support for controlling and disabling them via environment variables.

You can override various aspects of this feature by defining various symbols during configure. An example might be:

sh Configure -Accflags=-DPERL_HASH_FUNC_SIPHASH

Unless stated otherwise these options are considered experimental or insecure and are not recommended for production use.

Since Perl 5.18 we have included support for multiple hash functions, although from time to time we change which functions we support, and which function is default (currently SBOX+SIPHASH13 on 64 bit builds and SBOX+ZAPHOD32 for 32 bit builds). You can choose a different algorithm by defining one of the following symbols during configure. Note that there are security implications regarding which hash function you choose to use. The functions are listed roughly by how secure they are believed to be, with the one believed to be most secure at release time being PERL_HASH_FUNC_SIPHASH.

PERL_HASH_FUNC_SIPHASH
PERL_HASH_FUNC_SIPHASH13
PERL_HASH_FUNC_ZAPHOD32

In addition, these, (or custom hash functions), may be "fronted" by the SBOX32 hash function for keys under a chosen size. This hash function is special in that it has proven theoretical security properties, and is very fast to hash, but which by nature is restricted to a maximum key length, and which has rather expensive setup costs (relatively speaking), both in terms of performance and more importantly in terms of memory. SBOX32 requires 1k of storage per character it can hash, and it must populate that storage with 256 32-bit random values as well. In practice the RNG we use for seeding the SBOX32 storage is very efficient, and populating the table required for hashing even fairly long keys is negligible as we only do it during startup. By default we build with SBOX32 enabled, but you can change that by setting the PERL_HASH_USE_SBOX32_ALSO in the Configure process, with something like this

-Accflags='-DPERL_HASH_USE_SBOX32_ALSO=0'

or alternatively you can use the simple define PERL_HASH_NO_SBOX32 like this:

-Accflags='-DPERL_HASH_NO_SBOX32'

By default Perl will use SBOX32 to hash strings 24 bytes or shorter, you can change this length by setting SBOX32_MAX_LEN to the desired length, with the maximum length being 256. For example with this:

-Accflags='-DSBOX_MAX_LEN=128'

As of Perl 5.18 the order returned by keys(), values(), and each() is non-deterministic and distinct per hash, and the insert order for colliding keys is randomized as well, and perl allows for controlling this by the PERL_PERTURB_KEYS environment setting. You can disable this behavior entirely with the define PERL_PERTURB_KEYS_DISABLED with

-Accflags='-DPERL_PERTURB_KEYS_DISABLED'

You can disable the environment variable checks and compile time specify the type of key traversal randomization to be used by defining either PERL_PERTURB_KEYS_RANDOM or PERL_PERTURB_KEYS_DETERMINISTIC with

-Accflags='-DPERL_PERTURB_KEYS_RANDOM'

or

-Accflags='-DPERL_PERTURB_KEYS_DETERMINISTIC'

Since Perl 5.18 the seed used for the hash function is randomly selected at process start, which can be overridden by specifying a seed by setting the PERL_HASH_SEED environment variable. Be aware that PERL_PERTURB_KEYS_DETERMINISTIC and the equivalent environment variable setting will produce the same results if and only if the code does not put non-deterministic data into a hash, and the code is executed in exactly the same context in terms of the environment. If the code populates a hash with random data, or builds a hash using the address of its items, or if the code is run in a different environment context with a different number or selection of environment variables then the result may differ. DETERMINISTIC in this context means "if everything else is kept the same the same results should be observed".

You can change this behavior so that your perl is built with a hard coded seed with the define NO_HASH_SEED by providing to Configure

-Accflags='-DNO_HASH_SEED'

Note that if you do this you should modify the code in hv_func.h to specify your own key. In the future this define may be renamed and replaced with one that requires you to specify the key to use.

NOTE WELL: Perl has never guaranteed any ordering of the hash keys, and the ordering has already changed several times during the lifetime of Perl 5. Also, the ordering of hash keys has always been, and continues to be, affected by the insertion order regardless of whether you build with or without the randomization features. Note that because of this and especially with randomization that the key order of a hash is *undefined* and that things like Data::Dumper, for example, may produce different output between different runs of Perl, since Data::Dumper serializes the key in the native order for the hash. The use of the Data::Dumper Sortkeys option is recommended if you are comparing dumps between different invocations of perl.

See "PERL_HASH_SEED" in perlrun and "PERL_PERTURB_KEYS" in perlrun for details on the environment variables, and "Algorithmic Complexity Attacks" in perlsec for further security details.

The PERL_HASH_SEED and PERL_PERTURB_KEYS environment variables can be disabled by building configuring perl with -Accflags=-DNO_PERL_HASH_ENV.

The PERL_HASH_SEED_DEBUG environment variable can be disabled by configuring perl with -Accflags=-DNO_PERL_HASH_SEED_DEBUG.

MISCELLANEOUS CONFIG

Perl uses various defines to control defaults for its behavior. These values are chosen to represent "sane" config, but users can override these values in their builds if they wish. This is a list of such settings.

PERL_MAX_NESTED_EVAL_BEGIN_BLOCKS_DEFAULT

This define is used to control the default maximum number of nested eval/BEGIN statements, and in this context require should be understood to be a special form of eval so this means require/BEGIN and "use" statements as well.

Currently each BEGIN block inside of an eval EXPR or require operation will use a fairly high number of frames of the perl internal C stack, and this value is used to prevent stack overflows. Normally it is defaulted to 1000 but the default can be configured to another value, for instance 100, like this

-Accflags='-DPERL_MAX_NESTED_EVAL_BEGIN_BLOCKS_DEFAULT=100'

If you don't know what this is then it is safe to ignore it. Do not configure this to 0 or another very low value, it will break a lot of code. If you want to set it to a low value use the run time variable ${^MAX_NESTED_EVAL_BEGIN_BLOCKS} instead.

SOCKS

Perl can be configured to be 'socksified', that is, to use the SOCKS TCP/IP proxy protocol library. SOCKS is used to give applications access to transport layer network proxies. Perl supports only SOCKS Version 5. The corresponding Configure option is -Dusesocks. You can find more about SOCKS from wikipedia at https://en.wikipedia.org/wiki/SOCKS.

Dynamic Loading

By default, Configure will compile perl to use dynamic loading. If you want to force perl to be compiled completely statically, you can either choose this when Configure prompts you or you can use the Configure command line option -Uusedl. With this option, you won't be able to use any new extension (XS) module without recompiling perl itself.

Building a shared Perl library

Currently, for most systems, the main perl executable is built by linking the "perl library" libperl.a with perlmain.o, your static extensions, and various extra libraries, such as -lm.

On systems that support dynamic loading, it may be possible to replace libperl.a with a shared libperl.so. If you anticipate building several different perl binaries (e.g. by embedding libperl into different programs, or by using the optional compiler extension), then you might wish to build a shared libperl.so so that all your binaries can share the same library.

The disadvantages are that there may be a significant performance penalty associated with the shared libperl.so, and that the overall mechanism is still rather fragile with respect to different versions and upgrades.

In terms of performance, on my test system (Solaris 2.5_x86) the perl test suite took roughly 15% longer to run with the shared libperl.so. Your system and typical applications may well give quite different results.

The default name for the shared library is typically something like libperl.so.5.8.8 (for Perl 5.8.8), or libperl.so.588, or simply libperl.so. Configure tries to guess a sensible naming convention based on your C library name. Since the library gets installed in a version-specific architecture-dependent directory, the exact name isn't very important anyway, as long as your linker is happy.

You can elect to build a shared libperl by

sh Configure -Duseshrplib

To build a shared libperl, the environment variable controlling shared library search (LD_LIBRARY_PATH in most systems, DYLD_LIBRARY_PATH for Darwin, LD_LIBRARY_PATH/SHLIB_PATH for HP-UX, LIBPATH for AIX and z/OS, PATH for Cygwin) must be set up to include the Perl build directory because that's where the shared libperl will be created. Configure arranges makefile to have the correct shared library search settings. You can find the name of the environment variable Perl thinks works in your your system by

grep ldlibpthname config.sh

However, there are some special cases where manually setting the shared library path might be required. For example, if you want to run something like the following with the newly-built but not-yet-installed ./perl:

./perl -I. -MTestInit t/misc/failing_test.t

or

./perl -Ilib ~/my_mission_critical_test

then you need to set up the shared library path explicitly. You can do this with

LD_LIBRARY_PATH=`pwd`:$LD_LIBRARY_PATH; export LD_LIBRARY_PATH

for Bourne-style shells, or

setenv LD_LIBRARY_PATH `pwd`

for Csh-style shells. (This procedure may also be needed if for some unexpected reason Configure fails to set up makefile correctly.) (And again, it may be something other than LD_LIBRARY_PATH for you, see above.)

You can often recognize failures to build/use a shared libperl from error messages complaining about a missing libperl.so (or libperl.sl in HP-UX), for example:

18126:./miniperl: /sbin/loader: Fatal Error: cannot map libperl.so

There is also a potential problem with the shared perl library if you want to have more than one "flavor" of the same version of perl (e.g. with and without -DDEBUGGING). For example, suppose you build and install a standard Perl 5.10.0 with a shared library. Then, suppose you try to build Perl 5.10.0 with -DDEBUGGING enabled, but everything else the same, including all the installation directories. How can you ensure that your newly built perl will link with your newly built libperl.so.8 rather with the installed libperl.so.8? The answer is that you might not be able to. The installation directory is encoded in the perl binary with the LD_RUN_PATH environment variable (or equivalent ld command-line option). On Solaris, you can override that with LD_LIBRARY_PATH; on Linux, you can only override at runtime via LD_PRELOAD, specifying the exact filename you wish to be used; and on Digital Unix, you can override LD_LIBRARY_PATH by setting the _RLD_ROOT environment variable to point to the perl build directory.

In other words, it is generally not a good idea to try to build a perl with a shared library if $archlib/CORE/$libperl already exists from a previous build.

A good workaround is to specify a different directory for the architecture-dependent library for your -DDEBUGGING version of perl. You can do this by changing all the *archlib* variables in config.sh to point to your new architecture-dependent library.

External glob

Before File::Glob entered core in 5.6.0 globbing was implemented by shelling out. If the environmental variable PERL_EXTERNAL_GLOB is defined and if the csh shell is available, perl will still do this the old way.

Installation Directories

The installation directories can all be changed by answering the appropriate questions in Configure. For convenience, all the installation questions are near the beginning of Configure. Do not include trailing slashes on directory names. At any point during the Configure process, you can answer a question with &-d and Configure will use the defaults from then on. Alternatively, you can

grep '^install' config.sh

after Configure has run to verify the installation paths.

The defaults are intended to be reasonable and sensible for most people building from sources. Those who build and distribute binary distributions or who export perl to a range of systems will probably need to alter them. If you are content to just accept the defaults, you can safely skip the next section.

The directories set up by Configure fall into three broad categories.

Directories for the perl distribution

By default, Configure will use the following directories for 5.39.0. $version is the full perl version number, including subversion, e.g. 5.12.3, and $archname is a string like sun4-sunos, determined by Configure. The full definitions of all Configure variables are in the file Porting/Glossary.

Configure variable      Default value
$prefixexp              /usr/local
$binexp         $prefixexp/bin
$scriptdirexp   $prefixexp/bin
$privlibexp             $prefixexp/lib/perl5/$version
$archlibexp             $prefixexp/lib/perl5/$version/$archname
$man1direxp             $prefixexp/man/man1
$man3direxp             $prefixexp/man/man3
$html1direxp    (none)
$html3direxp    (none)

$prefixexp is generated from $prefix, with ~ expansion done to convert home directories into absolute paths. Similarly for the other variables listed. As file system calls do not do this, you should always reference the ...exp variables, to support users who build perl in their home directory.

Actually, Configure recognizes the SVR3-style /usr/local/man/l_man/man1 directories, if present, and uses those instead. Also, if $prefix contains the string "perl", the library directories are simplified as described below. For simplicity, only the common style is shown here.

Directories for site-specific add-on files

After perl is installed, you may later wish to add modules (e.g. from CPAN) or scripts. Configure will set up the following directories to be used for installing those add-on modules and scripts.

Configure        Default
variable          value
 $siteprefixexp    $prefixexp
 $sitebinexp       $siteprefixexp/bin
 $sitescriptexp    $siteprefixexp/bin
 $sitelibexp       $siteprefixexp/lib/perl5/site_perl/$version
 $sitearchexp
            $siteprefixexp/lib/perl5/site_perl/$version/$archname
 $siteman1direxp   $siteprefixexp/man/man1
 $siteman3direxp   $siteprefixexp/man/man3
 $sitehtml1direxp  (none)
 $sitehtml3direxp  (none)

By default, ExtUtils::MakeMaker will install architecture-independent modules into $sitelib and architecture-dependent modules into $sitearch.

Directories for vendor-supplied add-on files

Lastly, if you are building a binary distribution of perl for distribution, Configure can optionally set up the following directories for you to use to distribute add-on modules.

Configure          Default
variable            value
 $vendorprefixexp    (none)

 (The next ones are set only if vendorprefix is set.)

 $vendorbinexp       $vendorprefixexp/bin
 $vendorscriptexp    $vendorprefixexp/bin
 $vendorlibexp       $vendorprefixexp/lib/perl5/vendor_perl/$version
 $vendorarchexp
        $vendorprefixexp/lib/perl5/vendor_perl/$version/$archname
 $vendorman1direxp   $vendorprefixexp/man/man1
 $vendorman3direxp   $vendorprefixexp/man/man3
 $vendorhtml1direxp  (none)
 $vendorhtml3direxp  (none)

These are normally empty, but may be set as needed. For example, a vendor might choose the following settings:

$prefix           /usr
$siteprefix       /usr/local
$vendorprefix     /usr

This would have the effect of setting the following:

$binexp           /usr/bin
$scriptdirexp     /usr/bin
$privlibexp       /usr/lib/perl5/$version
$archlibexp       /usr/lib/perl5/$version/$archname
$man1direxp       /usr/man/man1
$man3direxp       /usr/man/man3

$sitebinexp       /usr/local/bin
$sitescriptexp    /usr/local/bin
$sitelibexp       /usr/local/lib/perl5/site_perl/$version
$sitearchexp      /usr/local/lib/perl5/site_perl/$version/$archname
$siteman1direxp   /usr/local/man/man1
$siteman3direxp   /usr/local/man/man3

$vendorbinexp     /usr/bin
$vendorscriptexp  /usr/bin
$vendorlibexp     /usr/lib/perl5/vendor_perl/$version
$vendorarchexp    /usr/lib/perl5/vendor_perl/$version/$archname
$vendorman1direxp /usr/man/man1
$vendorman3direxp /usr/man/man3

Note how in this example, the vendor-supplied directories are in the /usr hierarchy, while the directories reserved for the end user are in the /usr/local hierarchy.

The entire installed library hierarchy is installed in locations with version numbers, keeping the installations of different versions distinct. However, later installations of Perl can still be configured to search the installed libraries corresponding to compatible earlier versions. See "Coexistence with earlier versions of perl 5" below for more details on how Perl can be made to search older version directories.

Of course you may use these directories however you see fit. For example, you may wish to use $siteprefix for site-specific files that are stored locally on your own disk and use $vendorprefix for site-specific files that are stored elsewhere on your organization's network. One way to do that would be something like

sh Configure -Dsiteprefix=/usr/local -Dvendorprefix=/usr/share/perl
otherlibdirs

As a final catch-all, Configure also offers an $otherlibdirs variable. This variable contains a colon-separated list of additional directories to add to @INC. By default, it will be empty. Perl will search these directories (including architecture and version-specific subdirectories) for add-on modules and extensions.

For example, if you have a bundle of perl libraries from a previous installation, perhaps in a strange place:

sh Configure -Dotherlibdirs=/usr/lib/perl5/site_perl/5.8.1
APPLLIB_EXP

There is one other way of adding paths to @INC at perl build time, and that is by setting the APPLLIB_EXP C pre-processor token to a colon- separated list of directories, like this

sh Configure -Accflags='-DAPPLLIB_EXP=\"/usr/libperl\"'

The directories defined by APPLLIB_EXP get added to @INC first, ahead of any others, and so provide a way to override the standard perl modules should you, for example, want to distribute fixes without touching the perl distribution proper. And, like otherlib dirs, version and architecture specific subdirectories are also searched, if present, at run time. Of course, you can still search other @INC directories ahead of those in APPLLIB_EXP by using any of the standard run-time methods: $PERLLIB, $PERL5LIB, -I, use lib, etc.

default_inc_excludes_dot

Since version 5.26.0, default perl builds no longer includes '.' as the last element of @INC. The old behaviour can restored using

sh Configure -Udefault_inc_excludes_dot

Note that this is likely to make programs run under such a perl interpreter less secure.

usesitecustomize

Run-time customization of @INC can be enabled with:

sh Configure -Dusesitecustomize

which will define USE_SITECUSTOMIZE and $Config{usesitecustomize}. When enabled, this makes perl run $sitelibexp/sitecustomize.pl before anything else. This script can then be set up to add additional entries to @INC.

Man Pages

By default, man pages will be installed in $man1dir and $man3dir, which are normally /usr/local/man/man1 and /usr/local/man/man3. If you want to use a .3pm suffix for perl man pages, you can do that with

sh Configure -Dman3ext=3pm

You can disable installation of man pages completely using

sh Configure -Dman1dir=none -Dman3dir=none
HTML pages

Currently, the standard perl installation does not do anything with HTML documentation, but that may change in the future. Further, some add-on modules may wish to install HTML documents. The html Configure variables listed above are provided if you wish to specify where such documents should be placed. The default is "none", but will likely eventually change to something useful based on user feedback.

Some users prefer to append a "/share" to $privlib and $sitelib to emphasize that those directories can be shared among different architectures.

Note that these are just the defaults. You can actually structure the directories any way you like. They don't even have to be on the same filesystem.

Further details about the installation directories, maintenance and development subversions, and about supporting multiple versions are discussed in "Coexistence with earlier versions of perl 5" below.

If you specify a prefix that contains the string "perl", then the library directory structure is slightly simplified. Instead of suggesting $prefix/lib/perl5/, Configure will suggest $prefix/lib.

Thus, for example, if you Configure with -Dprefix=/opt/perl, then the default library directories for 5.9.0 are

Configure variable      Default value
        $privlib        /opt/perl/lib/5.9.0
        $archlib        /opt/perl/lib/5.9.0/$archname
        $sitelib        /opt/perl/lib/site_perl/5.9.0
        $sitearch       /opt/perl/lib/site_perl/5.9.0/$archname

Changing the installation directory

Configure distinguishes between the directory in which perl (and its associated files) should be installed, and the directory in which it will eventually reside. For most sites, these two are the same; for sites that use AFS, this distinction is handled automatically. However, sites that use package management software such as rpm or dpkg, or users building binary packages for distribution may also wish to install perl into a different directory before moving perl to its final destination. There are two ways to do that:

installprefix

To install perl under the /tmp/perl5 directory, use the following command line:

sh Configure -Dinstallprefix=/tmp/perl5

(replace /tmp/perl5 by a directory of your choice).

Beware, though, that if you go to try to install new add-on modules, they too will get installed in under '/tmp/perl5' if you follow this example. That's why it's usually better to use DESTDIR, as shown in the next section.

DESTDIR

If you need to install perl on many identical systems, it is convenient to compile it once and create an archive that can be installed on multiple systems. Suppose, for example, that you want to create an archive that can be installed in /opt/perl. One way to do that is by using the DESTDIR variable during make install. The DESTDIR is automatically prepended to all the installation paths. Thus you simply do:

sh Configure -Dprefix=/opt/perl -des
make
make test
make install DESTDIR=/tmp/perl5
cd /tmp/perl5/opt/perl
tar cvf /tmp/perl5-archive.tar .

Relocatable @INC

To create a relocatable perl tree, use the following command line:

sh Configure -Duserelocatableinc

Then the paths in @INC (and everything else in %Config) can be optionally located via the path of the perl executable.

That means that, if the string ".../" is found at the start of any path, it's substituted with the directory of $^X. So, the relocation can be configured on a per-directory basis, although the default with "-Duserelocatableinc" is that everything is relocated. The initial install is done to the original configured prefix.

This option is not compatible with the building of a shared libperl ("-Duseshrplib"), because in that case perl is linked with an hard-coded rpath that points at the libperl.so, that cannot be relocated.

Site-wide Policy settings

After Configure runs, it stores a number of common site-wide "policy" answers (such as installation directories) in the Policy.sh file. If you want to build perl on another system using the same policy defaults, simply copy the Policy.sh file to the new system's perl build directory, and Configure will use it. This will work even if Policy.sh was generated for another version of Perl, or on a system with a different architecture and/or operating system. However, in such cases, you should review the contents of the file before using it: for example, your new target may not keep its man pages in the same place as the system on which the file was generated.

Alternatively, if you wish to change some or all of those policy answers, you should

rm -f Policy.sh

to ensure that Configure doesn't re-use them.

Further information is in the Policy_sh.SH file itself.

If the generated Policy.sh file is unsuitable, you may freely edit it to contain any valid shell commands. It will be run just after the platform-specific hints files.

Disabling older versions of Perl

Configure will search for binary compatible versions of previously installed perl binaries in the tree that is specified as target tree, and these will be used as locations to search for modules by the perl being built. The list of perl versions found will be put in the Configure variable inc_version_list.

To disable this use of older perl modules, even completely valid pure perl modules, you can specify to not include the paths found:

sh Configure -Dinc_version_list=none ...

If you do want to use modules from some previous perl versions, the variable must contain a space separated list of directories under the site_perl directory, and has to include architecture-dependent directories separately, eg.

sh Configure -Dinc_version_list="5.16.0/x86_64-linux 5.16.0" ...

When using the newer perl, you can add these paths again in the PERL5LIB environment variable or with perl's -I runtime option.

Building Perl outside of the source directory

Sometimes it is desirable to build Perl in a directory different from where the sources are, for example if you want to keep your sources read-only, or if you want to share the sources between different binary architectures. You can do this (if your file system supports symbolic links) by

mkdir /tmp/perl/build/directory
cd /tmp/perl/build/directory
sh /path/to/perl/source/Configure -Dmksymlinks ...

This will create in /tmp/perl/build/directory a tree of symbolic links pointing to files in /path/to/perl/source. The original files are left unaffected. After Configure has finished you can just say

make
make test
make install

as usual, and Perl will be built in /tmp/perl/build/directory.

Building a debugging perl

You can run perl scripts under the perl debugger at any time with perl -d your_script. If, however, you want to debug perl itself, you probably want to have support for perl internal debugging code (activated by adding -DDEBUGGING to ccflags), and/or support for the system debugger by adding -g to the optimisation flags.

A perl compiled with the DEBUGGING C preprocessor macro will support the -D perl command-line switch, have assertions enabled, and have many extra checks compiled into the code; but will execute much more slowly (typically 2-3x) and the binary will be much larger (typically 2-3x).

As a convenience, debugging code (-DDEBUGGING) and debugging symbols (-g) can be enabled jointly or separately using a Configure switch, also (somewhat confusingly) named -DDEBUGGING. For a more eye appealing call, -DEBUGGING is defined to be an alias for -DDEBUGGING. For both, the -U calls are also supported, in order to be able to overrule the hints or Policy.sh settings.

Here are the DEBUGGING modes:

Configure -DDEBUGGING
Configure -DEBUGGING
Configure -DEBUGGING=both

Sets both -DDEBUGGING in the ccflags, and adds -g to optimize.

You can actually specify -g and -DDEBUGGING independently (see below), but usually it's convenient to have both.

Configure -DEBUGGING=-g
Configure -Doptimize=-g

Adds -g to optimize, but does not set -DDEBUGGING.

(Note: Your system may actually require something like cc -g2. Check your man pages for cc(1) and also any hint file for your system.)

Configure -DEBUGGING=none
Configure -UDEBUGGING

Removes -g from optimize, and -DDEBUGGING from ccflags.

If you are using a shared libperl, see the warnings about multiple versions of perl under "Building a shared Perl library".

Note that a perl built with -DDEBUGGING will be much bigger and will run much, much more slowly than a standard perl.

DTrace support

On platforms where DTrace is available, it may be enabled by using the -Dusedtrace option to Configure. DTrace probes are available for subroutine entry (sub-entry) and subroutine exit (sub-exit). Here's a simple D script that uses them:

perl$target:::sub-entry, perl$target:::sub-return {
  printf("%s %s (%s:%d)\n", probename == "sub-entry" ? "->" : "<-",
            copyinstr(arg0), copyinstr(arg1), arg2);
}

Extensions

Perl ships with a number of standard extensions. These are contained in the ext/ subdirectory.

By default, Configure will offer to build every extension which appears to be supported. For example, Configure will offer to build GDBM_File only if it is able to find the gdbm library.

To disable certain extensions so that they are not built, use the -Dnoextensions=... and -Donlyextensions=... options. They both accept a space-separated list of extensions, such as IPC/SysV. The extensions listed in noextensions are removed from the list of extensions to build, while the onlyextensions is rather more severe and builds only the listed extensions. The latter should be used with extreme caution since certain extensions are used by many other extensions and modules: examples of such modules include Fcntl and IO. The order of processing these options is first only (if present), then no (if present).

Of course, you may always run Configure interactively and select only the extensions you want.

If you unpack any additional extensions in the ext/ directory before running Configure, then Configure will offer to build those additional extensions as well. Most users probably shouldn't have to do this -- it is usually easier to build additional extensions later after perl has been installed. However, if you wish to have those additional extensions statically linked into the perl binary, then this offers a convenient way to do that in one step. (It is not necessary, however; you can build and install extensions just fine even if you don't have dynamic loading. See lib/ExtUtils/MakeMaker.pm for more details.) Another way of specifying extra modules is described in "Adding extra modules to the build" below.

If you re-use an old config.sh but change your system (e.g. by adding libgdbm) Configure will still offer your old choices of extensions for the default answer, but it will also point out the discrepancy to you.

Including locally-installed libraries

Perl comes with interfaces to number of libraries, including threads, dbm, ndbm, gdbm, and Berkeley db. For the *db* extension, if Configure can find the appropriate header files and libraries, it will automatically include that extension. The threading extension needs to be specified explicitly (see "Threads").

Those libraries are not distributed with perl. If your header (.h) files for those libraries are not in a directory normally searched by your C compiler, then you will need to include the appropriate -I/your/directory option when prompted by Configure. If your libraries are not in a directory normally searched by your C compiler and linker, then you will need to include the appropriate -L/your/directory option when prompted by Configure. See the examples below.

Examples

gdbm in /usr/local

Suppose you have gdbm and want Configure to find it and build the GDBM_File extension. This example assumes you have gdbm.h installed in /usr/local/include/gdbm.h and libgdbm.a installed in /usr/local/lib/libgdbm.a. Configure should figure all the necessary steps out automatically.

Specifically, when Configure prompts you for flags for your C compiler, you should include -I/usr/local/include, if it's not here yet. Similarly, when Configure prompts you for linker flags, you should include -L/usr/local/lib.

If you are using dynamic loading, then when Configure prompts you for linker flags for dynamic loading, you should again include -L/usr/local/lib.

Again, this should all happen automatically. This should also work if you have gdbm installed in any of (/usr/local, /opt/local, /usr/gnu, /opt/gnu, /usr/GNU, or /opt/GNU).

BerkeleyDB in /usr/local/BerkeleyDB

The version of BerkeleyDB distributed by Oracle installs in a version-specific directory by default, typically something like /usr/local/BerkeleyDB.4.7. To have Configure find that, you need to add -I/usr/local/BerkeleyDB.4.7/include to cc flags, as in the previous example, and you will also have to take extra steps to help Configure find -ldb. Specifically, when Configure prompts you for library directories, add /usr/local/BerkeleyDB.4.7/lib to the list. Also, you will need to add appropriate linker flags to tell the runtime linker where to find the BerkeleyDB shared libraries.

It is possible to specify this from the command line (all on one line):

sh Configure -de \
   -Dlocincpth='/usr/local/BerkeleyDB.4.7/include             \
                                          /usr/local/include' \
   -Dloclibpth='/usr/local/BerkeleyDB.4.7/lib /usr/local/lib' \
   -Aldflags='-R/usr/local/BerkeleyDB.4.7/lib'

locincpth is a space-separated list of include directories to search. Configure will automatically add the appropriate -I directives.

loclibpth is a space-separated list of library directories to search. Configure will automatically add the appropriate -L directives.

The addition to ldflags is so that the dynamic linker knows where to find the BerkeleyDB libraries. For Linux and Solaris, the -R option does that. Other systems may use different flags. Use the appropriate flag for your system.

Specifying a logical root directory

If you are cross-compiling, or are using a compiler which has it's own headers and libraries in a nonstandard location, and your compiler understands the --sysroot option, you can use the -Dsysroot option to specify the logical root directory under which all libraries and headers are searched for. This patch adjusts Configure to search under $sysroot, instead of /.

--sysroot is added to ccflags and friends so that make in ExtUtils::MakeMaker, and other extensions, will use it.

Overriding an old config.sh

If you want to use an old config.sh produced by a previous run of Configure, but override some of the items with command line options, you need to use Configure -O.

GNU-style configure

If you prefer the GNU-style configure command line interface, you can use the supplied configure.gnu command, e.g.

CC=gcc ./configure.gnu

The configure.gnu script emulates a few of the more common configure options. Try

./configure.gnu --help

for a listing.

Unrecognized arguments with a double dash prefix produce an error.

Any other arguments are passed through to Configure, so you could build a threaded perl with:

CC=gcc ./configure.gnu -Dusethreads

(The file is called configure.gnu to avoid problems on systems that would not distinguish the files "Configure" and "configure".)

Malloc Issues

Perl relies heavily on malloc(3) to grow data structures as needed, so perl's performance can be noticeably affected by the performance of the malloc function on your system. The perl source is shipped with a version of malloc that has been optimized for the typical requests from perl, so there's a chance that it may be both faster and use less memory than your system malloc.

However, if your system already has an excellent malloc, or if you are experiencing difficulties with extensions that use third-party libraries that call malloc, then you should probably use your system's malloc. (Or, you might wish to explore the malloc flags discussed below.)

Using the system malloc

To build without perl's malloc, you can use the Configure command

sh Configure -Uusemymalloc

or you can answer 'n' at the appropriate interactive Configure prompt.

Note that Perl's malloc isn't always used by default; that actually depends on your system. For example, on Linux and FreeBSD (and many more systems), Configure chooses to use the system's malloc by default. See the appropriate file in the hints/ directory to see how the default is set.

-DPERL_POLLUTE_MALLOC

NOTE: This flag is enabled automatically on some platforms if you just run Configure to accept all the defaults.

Perl's malloc family of functions are normally called Perl_malloc(), Perl_realloc(), Perl_calloc() and Perl_mfree(). These names do not clash with the system versions of these functions.

If this flag is enabled, however, Perl's malloc family of functions will have the same names as the system versions. This may be required sometimes if you have libraries that like to free() data that may have been allocated by Perl_malloc() and vice versa.

Note that enabling this option may sometimes lead to duplicate symbols from the linker for malloc et al. In such cases, the system probably does not allow its malloc functions to be fully replaced with custom versions.

-DPERL_DEBUGGING_MSTATS

This flag enables debugging mstats, which is required to use the Devel::Peek::mstat() function. You cannot enable this unless you are using Perl's malloc, so a typical Configure command would be

sh Configure -Accflags=-DPERL_DEBUGGING_MSTATS -Dusemymalloc

to enable this option.

What if it doesn't work?

If you run into problems, try some of the following ideas. If none of them help, then see "Reporting Problems" below.

Running Configure Interactively

If Configure runs into trouble, remember that you can always run Configure interactively so that you can check (and correct) its guesses.

All the installation questions have been moved to the top, so you don't have to wait for them. Once you've handled them (and your C compiler and flags) you can type &-d at the next Configure prompt and Configure will use the defaults from then on.

If you find yourself trying obscure command line incantations and config.over tricks, I recommend you run Configure interactively instead. You'll probably save yourself time in the long run.

Hint files

Hint files tell Configure about a number of things:

o

The peculiarities or conventions of particular platforms -- non-standard library locations and names, default installation locations for binaries, and so on.

o

The deficiencies of the platform -- for example, library functions that, although present, are too badly broken to be usable; or limits on resources that are generously available on most platforms.

o

How best to optimize for the platform, both in terms of binary size and/or speed, and for Perl feature support. Because of wide variations in the implementation of shared libraries and of threading, for example, Configure often needs hints in order to be able to use these features.

The perl distribution includes many system-specific hints files in the hints/ directory. If one of them matches your system, Configure will offer to use that hint file. Unless you have a very good reason not to, you should accept its offer.

Several of the hint files contain additional important information. If you have any problems, it is a good idea to read the relevant hint file for further information. See hints/solaris_2.sh for an extensive example. More information about writing good hints is in the hints/README.hints file, which also explains hint files known as callback-units.

Note that any hint file is read before any Policy file, meaning that Policy overrides hints -- see "Site-wide Policy settings".

WHOA THERE!!!

If you are re-using an old config.sh, it's possible that Configure detects different values from the ones specified in this file. You will almost always want to keep the previous value, unless you have changed something on your system.

For example, suppose you have added libgdbm.a to your system and you decide to reconfigure perl to use GDBM_File. When you run Configure again, you will need to add -lgdbm to the list of libraries. Now, Configure will find your gdbm include file and library and will issue a message:

*** WHOA THERE!!! ***
        The previous value for $i_gdbm on this machine was "undef"!
        Keep the previous value? [y]

In this case, you do not want to keep the previous value, so you should answer 'n'. (You'll also have to manually add GDBM_File to the list of dynamic extensions to build.)

Changing Compilers

If you change compilers or make other significant changes, you should probably not re-use your old config.sh. Simply remove it or rename it, then rerun Configure with the options you want to use.

Propagating your changes to config.sh

If you make any changes to config.sh, you should propagate them to all the .SH files by running

sh Configure -S

You will then have to rebuild by running

make depend
make
config.over and config.arch

You can also supply a shell script config.over to override Configure's guesses. It will get loaded up at the very end, just before config.sh is created. You have to be careful with this, however, as Configure does no checking that your changes make sense. This file is usually good for site-specific customizations.

There is also another file that, if it exists, is loaded before the config.over, called config.arch. This file is intended to be per architecture, not per site, and usually it's the architecture-specific hints file that creates the config.arch.

config.h

Many of the system dependencies are contained in config.h. Configure builds config.h by running the config_h.SH script. The values for the variables are taken from config.sh.

If there are any problems, you can edit config.h directly. Beware, though, that the next time you run Configure, your changes will be lost.

cflags

If you have any additional changes to make to the C compiler command line, they can be made in cflags.SH. For instance, to turn off the optimizer on toke.c, find the switch structure marked 'or customize here', and add a line for toke.c ahead of the catch-all *) so that it now reads:

: or customize here

case "$file" in
toke) optimize='-g' ;;
*) ;;

You should not edit the generated file cflags directly, as your changes will be lost the next time you run Configure, or if you edit config.sh.

To explore various ways of changing ccflags from within a hint file, see the file hints/README.hints.

To change the C flags for all the files, edit config.sh and change either $ccflags or $optimize, and then re-run

sh Configure -S
make depend
No sh

If you don't have sh, you'll have to copy the sample file Porting/config.sh to config.sh and edit your config.sh to reflect your system's peculiarities. See Porting/pumpkin.pod for more information. You'll probably also have to extensively modify the extension building mechanism.

Porting information

Specific information for the OS/2, Plan 9, VMS and Win32 ports is in the corresponding README files and subdirectories. Additional information, including a glossary of all those config.sh variables, is in the Porting subdirectory. Porting/Glossary should especially come in handy.

Ports for other systems may also be available. You should check out https://www.cpan.org/ports for current information on ports to various other operating systems.

If you plan to port Perl to a new architecture, study carefully the section titled "Philosophical Issues in Patching and Porting Perl" in the file Porting/pumpkin.pod and the file pod/perlgit.pod. Study also how other non-UNIX ports have solved problems.

Adding extra modules to the build

You can specify extra modules or module bundles to be fetched from the CPAN and installed as part of the Perl build. Either use the -Dextras=... command line parameter to Configure, for example like this:

Configure -Dextras="Bundle::LWP DBI"

or answer first 'y' to the question 'Install any extra modules?' and then answer "Bundle::LWP DBI" to the 'Extras?' question. The module or the bundle names are as for the CPAN module 'install' command. This will only work if those modules are to be built as dynamic extensions. If you wish to include those extra modules as static extensions, see "Extensions" above.

Notice that because the CPAN module will be used to fetch the extra modules, you will need access to the CPAN, either via the Internet, or via a local copy such as a CD-ROM or a local CPAN mirror. If you do not, using the extra modules option will die horribly.

Also notice that you yourself are responsible for satisfying any extra dependencies such as external headers or libraries BEFORE trying the build. For example: you will need to have the Foo database specific headers and libraries installed for the DBD::Foo module. The Configure process or the Perl build process will not help you with these.

suidperl

suidperl was an optional component of earlier releases of perl. It is no longer available. Instead, use a tool specifically designed to handle changes in privileges, such as sudo.

make depend

This will look for all the includes. The output is stored in makefile. The only difference between Makefile and makefile is the dependencies at the bottom of makefile. If you have to make any changes, you should edit makefile, not Makefile, since the Unix make command reads makefile first. (On non-Unix systems, the output may be stored in a different file. Check the value of $firstmakefile in your config.sh if in doubt.)

Configure will offer to do this step for you, so it isn't listed explicitly above.

make

This will attempt to make perl in the current directory.

Expected errors

These error reports are normal, and can be ignored:

...
make: [extra.pods] Error 1 (ignored)
...
make: [extras.make] Error 1 (ignored)

What if it doesn't work?

If you can't compile successfully, try some of the following ideas. If none of them help, and careful reading of the error message and the relevant manual pages on your system doesn't help, then see "Reporting Problems" below.

hints

If you used a hint file, try reading the comments in the hint file for further tips and information.

extensions

If you can successfully build miniperl, but the process crashes during the building of extensions, run

make minitest

to test your version of miniperl.

locale

If you have any locale-related environment variables set, try unsetting them. I have some reports that some versions of IRIX hang while running ./miniperl configpm with locales other than the C locale. See the discussion under "make test" below about locales and the whole "LOCALE PROBLEMS" in perllocale section in the file pod/perllocale.pod. The latter is especially useful if you see something like this

perl: warning: Setting locale failed.
perl: warning: Please check that your locale settings:
        LC_ALL = "En_US",
        LANG = (unset)
    are supported and installed on your system.
perl: warning: Falling back to the standard locale ("C").

at Perl startup.

other environment variables

Configure does not check for environment variables that can sometimes have a major influence on how perl is built or tested. For example, OBJECT_MODE on AIX determines the way the compiler and linker deal with their objects, but this is a variable that only influences build-time behaviour, and should not affect the perl scripts that are eventually executed by the perl binary. Other variables, like PERL_UNICODE, PERL5LIB, and PERL5OPT will influence the behaviour of the test suite. So if you are getting strange test failures, you may want to try retesting with the various PERL variables unset.

LD_LIBRARY_PATH

If you run into dynamic loading problems, check your setting of the LD_LIBRARY_PATH environment variable (or on some systems the equivalent with a different name, see "Building a shared Perl library"). If you're creating a static Perl library (libperl.a rather than libperl.so) it should build fine with LD_LIBRARY_PATH unset, though that may depend on details of your local setup.

nm extraction

If Configure seems to be having trouble finding library functions, try not using nm extraction. You can do this from the command line with

sh Configure -Uusenm

or by answering the nm extraction question interactively. If you have previously run Configure, you should not reuse your old config.sh.

umask not found

If the build processes encounters errors relating to umask(), the problem is probably that Configure couldn't find your umask() system call. Check your config.sh. You should have d_umask='define'. If you don't, this is probably the "nm extraction" problem discussed above. Also, try reading the hints file for your system for further information.

do_aspawn

If you run into problems relating to do_aspawn or do_spawn, the problem is probably that Configure failed to detect your system's fork() function. Follow the procedure in the previous item on "nm extraction".

__inet_* errors

If you receive unresolved symbol errors during Perl build and/or test referring to __inet_* symbols, check to see whether BIND 8.1 is installed. It installs a /usr/local/include/arpa/inet.h that refers to these symbols. Versions of BIND later than 8.1 do not install inet.h in that location and avoid the errors. You should probably update to a newer version of BIND (and remove the files the old one left behind). If you can't, you can either link with the updated resolver library provided with BIND 8.1 or rename /usr/local/bin/arpa/inet.h during the Perl build and test process to avoid the problem.

.*_r() prototype NOT found

On a related note, if you see a bunch of complaints like the above about reentrant functions -- specifically networking-related ones -- being present but without prototypes available, check to see if BIND 8.1 (or possibly other BIND 8 versions) is (or has been) installed. They install header files such as netdb.h into places such as /usr/local/include (or into another directory as specified at build/install time), at least optionally. Remove them or put them in someplace that isn't in the C preprocessor's header file include search path (determined by -I options plus defaults, normally /usr/include).

#error "No DATAMODEL_NATIVE specified"

This is a common error when trying to build perl on Solaris 2.6 with a gcc installation from Solaris 2.5 or 2.5.1. The Solaris header files changed, so you need to update your gcc installation. You can either rerun the fixincludes script from gcc or take the opportunity to update your gcc installation.

Optimizer

If you can't compile successfully, try turning off your compiler's optimizer. Edit config.sh and change the line

optimize='-O'

to

optimize=' '

then propagate your changes with sh Configure -S and rebuild with make depend; make.

Missing functions and Undefined symbols

If the build of miniperl fails with a long list of missing functions or undefined symbols, check the libs variable in the config.sh file. It should look something like

libs='-lsocket -lnsl -ldl -lm -lc'

The exact libraries will vary from system to system, but you typically need to include at least the math library -lm. Normally, Configure will suggest the correct defaults. If the libs variable is empty, you need to start all over again. Run

make distclean

and start from the very beginning. This time, unless you are sure of what you are doing, accept the default list of libraries suggested by Configure.

If the libs variable is missing -lm, there is a chance that libm.so.1 is available, but the required (symbolic) link to libm.so is missing. (same could be the case for other libraries like libcrypt.so). You should check your installation for packages that create that link, and if no package is installed that supplies that link or you cannot install them, make the symbolic link yourself e.g.:

$ rpm -qf /usr/lib64/libm.so
glibc-devel-2.15-22.17.1.x86_64
$ ls -lgo /usr/lib64/libm.so
lrwxrwxrwx 1 16 Jan  7  2013 /usr/lib64/libm.so -> /lib64/libm.so.6

or

$ sudo ln -s /lib64/libm.so.6 /lib64/libm.so

If the libs variable looks correct, you might have the "nm extraction" problem discussed above.

If you still have missing routines or undefined symbols, you probably need to add some library or other, make a symbolic link like described above, or you need to undefine some feature that Configure thought was there but is defective or incomplete. If you used a hint file, see if it has any relevant advice. You can also look through config.h for likely suspects.

toke.c

Some compilers will not compile or optimize the larger files (such as toke.c) without some extra switches to use larger jump offsets or allocate larger internal tables. You can customize the switches for each file in cflags.SH. It's okay to insert rules for specific files into makefile since a default rule only takes effect in the absence of a specific rule.

Missing dbmclose

SCO prior to 3.2.4 may be missing dbmclose(). An upgrade to 3.2.4 that includes libdbm.nfs (which includes dbmclose()) may be available.

error: too few arguments to function 'dbmclose'

Building ODBM_File on some (Open)SUSE distributions might run into this error, as the header file is broken. There are two ways to deal with this

1. Disable the use of ODBM_FILE

   sh Configure ... -Dnoextensions=ODBM_File

2. Fix the header file, somewhat like this:

   --- a/usr/include/dbm.h  2010-03-24 08:54:59.000000000 +0100
   +++ b/usr/include/dbm.h  2010-03-24 08:55:15.000000000 +0100
   @@ -59,4 +59,4 @@ extern datum  firstkey __P((void));

    extern datum   nextkey __P((datum key));

   -extern int     dbmclose __P((DBM *));
   +extern int     dbmclose __P((void));
Warning (mostly harmless): No library found for -lsomething

If you see such a message during the building of an extension, but the extension passes its tests anyway (see "make test" below), then don't worry about the warning message. The extension Makefile.PL goes looking for various libraries needed on various systems; few systems will need all the possible libraries listed. Most users will see warnings for the ones they don't have. The phrase 'mostly harmless' is intended to reassure you that nothing unusual is happening, and the build process is continuing.

On the other hand, if you are building GDBM_File and you get the message

Warning (mostly harmless): No library found for -lgdbm

then it's likely you're going to run into trouble somewhere along the line, since it's hard to see how you can use the GDBM_File extension without the -lgdbm library.

It is true that, in principle, Configure could have figured all of this out, but Configure and the extension building process are not quite that tightly coordinated.

sh: ar: not found

This is a message from your shell telling you that the command 'ar' was not found. You need to check your PATH environment variable to make sure that it includes the directory with the 'ar' command. This is a common problem on Solaris, where 'ar' is in the /usr/ccs/bin directory.

db-recno failure on tests 51, 53 and 55

Old versions of the DB library (including the DB library which comes with FreeBSD 2.1) had broken handling of recno databases with modified bval settings. Upgrade your DB library or OS.

Bad arg length for semctl, is XX, should be ZZZ

If you get this error message from the cpan/IPC-SysV/t/sem.t test, your System V IPC may be broken. The XX typically is 20, and that is what ZZZ also should be. Consider upgrading your OS, or reconfiguring your OS to include the System V semaphores.

cpan/IPC-SysV/t/sem........semget: No space left on device

Either your account or the whole system has run out of semaphores. Or both. Either list the semaphores with "ipcs" and remove the unneeded ones (which ones these are depends on your system and applications) with "ipcrm -s SEMAPHORE_ID_HERE" or configure more semaphores to your system.

GNU binutils

If you mix GNU binutils (nm, ld, ar) with equivalent vendor-supplied tools you may be in for some trouble. For example creating archives with an old GNU 'ar' and then using a new current vendor-supplied 'ld' may lead into linking problems. Either recompile your GNU binutils under your current operating system release, or modify your PATH not to include the GNU utils before running Configure, or specify the vendor-supplied utilities explicitly to Configure, for example by Configure -Dar=/bin/ar.

THIS PACKAGE SEEMS TO BE INCOMPLETE

The Configure program has not been able to find all the files which make up the complete Perl distribution. You may have a damaged source archive file (in which case you may also have seen messages such as gzip: stdin: unexpected end of file and tar: Unexpected EOF on archive file), or you may have obtained a structurally-sound but incomplete archive. In either case, try downloading again from the official site named at the start of this document. If you do find that any site is carrying a corrupted or incomplete source code archive, please report it to the site's maintainer.

invalid token: ##

You are using a non-ANSI-compliant C compiler. To compile Perl, you need to use a compiler that supports ANSI C. If there is a README file for your system, it may have further details on your compiler options.

Miscellaneous

Some additional things that have been reported:

Genix may need to use libc rather than libc_s, or #undef VARARGS.

NCR Tower 32 (OS 2.01.01) may need -W2,-Sl,2000 and #undef MKDIR.

UTS may need one or more of -K or -g, and #undef LSTAT.

FreeBSD can fail the cpan/IPC-SysV/t/sem.t test if SysV IPC has not been configured in the kernel. Perl tries to detect this, though, and you will get a message telling you what to do.

Building Perl on a system that has also BIND (headers and libraries) installed may run into troubles because BIND installs its own netdb.h and socket.h, which may not agree with the operating system's ideas of the same files. Similarly, including -lbind may conflict with libc's view of the world. You may have to tweak -Dlocincpth and -Dloclibpth to avoid the BIND.

Cross-compilation

Perl can be cross-compiled. It is just not trivial, cross-compilation rarely is. Perl is routinely cross-compiled for several platforms: as of June 2019, these include Android, Blackberry 10, ARM Linux, and Solaris. Previous versions of Perl also provided support for Open Zaurus, Symbian, and the IBM OS/400, but it's unknown if those ports are still functional. These platforms are known as the target platforms, while the systems where the compilation takes place are the host platforms.

What makes the situation difficult is that first of all, cross-compilation environments vary significantly in how they are set up and used, and secondly because the primary way of configuring Perl (using the rather large Unix-tool-dependent Configure script) is not awfully well suited for cross-compilation. However, starting from version 5.18.0, the Configure script also knows two ways of supporting cross-compilation, so please keep reading.

See the following files for more information about compiling Perl for the particular platforms:

Android

"Cross-compilation" in README.android or perlandroid

Blackberry

"Cross-compilation" in README.qnx or perlqnx

Solaris

"CROSS-COMPILATION" in README.solaris or perlsolaris

Linux

This document; See below.

Packaging and transferring either the core Perl modules or CPAN modules to the target platform is also left up to the each cross-compilation environment. Often the cross-compilation target platforms are somewhat limited in diskspace: see the section "Minimizing the Perl installation" to learn more of the minimal set of files required for a functional Perl installation.

For some cross-compilation environments the Configure option -Dinstallprefix=... might be handy, see "Changing the installation directory".

About the cross-compilation support of Configure: There's two forms. The more common one requires some way of transferring and running executables in the target system, such as an ssh connection; this is the ./Configure -Dusecrosscompile -Dtargethost=... route. The second method doesn't need access to the target system, but requires you to provide a config.sh, and a canned Makefile; the rest of this section describes the former.

This cross-compilation setup of Configure has successfully been used in a wide variety of setups, such as a 64-bit OS X host for an Android ARM target, or an amd64 Linux host targeting x86 Solaris, or even Windows.

To run Configure in cross-compilation mode the basic switch that has to be used is -Dusecrosscompile:

sh ./Configure -des -Dusecrosscompile -D...

This will make the cpp symbol USE_CROSS_COMPILE and the %Config symbol usecrosscompile available.

During the Configure and build, certain helper scripts will be created into the Cross/ subdirectory. The scripts are used to execute a cross-compiled executable, and to transfer files to and from the target host. The execution scripts are named run-* and the transfer scripts to-* and from-*. The part after the dash is the method to use for remote execution and transfer: by default the methods are ssh and scp, thus making the scripts run-ssh, to-scp, and from-scp.

To configure the scripts for a target host and a directory (in which the execution will happen and which is to and from where the transfer happens), supply Configure with

-Dtargethost=so.me.ho.st -Dtargetdir=/tar/get/dir

The targethost is what e.g. ssh will use as the hostname, the targetdir must exist (the scripts won't create it), the targetdir defaults to /tmp. You can also specify a username to use for ssh/rsh logins

-Dtargetuser=luser

but in case you don't, "root" will be used. Similarly, you can specify a non-standard (i.e. not 22) port for the connection, if applicable, through

-Dtargetport=2222

If the name of cc has the usual GNU C semantics for cross compilers, that is, CPU-OS-gcc, the target architecture (targetarch), plus names of the ar, nm, and ranlib will also be automatically chosen to be CPU-OS-ar and so on. (The ld requires more thought and will be chosen later by Configure as appropriate). This will also aid in guessing the proper operating system name for the target, which has other repercussions, like better defaults and possibly critical fixes for the platform. If Configure isn't guessing the OS name properly, you may need to either add a hint file redirecting Configure's guess, or modify Configure to make the correct choice.

If your compiler doesn't follow that convention, you will also need to specify which target environment to use, as well as ar and friends:

-Dtargetarch=arm-linux
-Dcc=mycrossgcc
-Dar=...

Additionally, a cross-compilation toolchain will usually install it's own logical system root somewhere -- that is, it'll create a directory somewhere which includes subdirectories like 'include' or 'lib'. For example, you may end up with /skiff/local/arm-linux, where /skiff/local/arm-linux/bin holds the binaries for cross-compilation, /skiff/local/arm-linux/include has the headers, and /skiff/local/arm-linux/lib has the library files. If this is the case, and you are using a compiler that understands --sysroot, like gcc or clang, you'll want to specify the -Dsysroot option for Configure:

-Dsysroot=/skiff/local/arm-linux

However, if your don't have a suitable directory to pass to -Dsysroot, you will also need to specify which target environment to use:

-Dusrinc=/skiff/local/arm-linux/include
-Dincpth=/skiff/local/arm-linux/include
-Dlibpth=/skiff/local/arm-linux/lib

In addition to the default execution/transfer methods you can also choose rsh for execution, and rcp or cp for transfer, for example:

-Dtargetrun=rsh -Dtargetto=rcp -Dtargetfrom=cp

Putting it all together:

sh ./Configure -des -Dusecrosscompile \
    -Dtargethost=so.me.ho.st \
    -Dtargetdir=/tar/get/dir \
    -Dtargetuser=root \
    -Dtargetarch=arm-linux \
    -Dcc=arm-linux-gcc \
    -Dsysroot=/skiff/local/arm-linux \
    -D...

or if you are happy with the defaults:

sh ./Configure -des -Dusecrosscompile \
    -Dtargethost=so.me.ho.st \
    -Dcc=arm-linux-gcc \
    -D...

Another example where the cross-compiler has been installed under /usr/local/arm/2.95.5:

sh ./Configure -des -Dusecrosscompile \
    -Dtargethost=so.me.ho.st \
    -Dcc=/usr/local/arm/2.95.5/bin/arm-linux-gcc \
    -Dsysroot=/usr/local/arm/2.95.5

There is also a targetenv option for Configure which can be used to modify the environment of the target just before testing begins during 'make test'. For example, if the target system has a nonstandard /tmp location, you could do this:

-Dtargetenv="export TMPDIR=/other/tmp;"

If you are planning on cross-compiling to several platforms, or some other thing that would involve running Configure several times, there are two options that can be used to speed things up considerably. As a bit of background, when you call Configure with -Dusecrosscompile, it begins by actually partially building a miniperl on the host machine, as well as the generate_uudmap binary, and we end up using that during the build. So instead of building that new perl every single time, you can build it just once in a separate directory, and then pass the resulting binaries to Configure like this:

-Dhostperl=/path/to/second/build/dir/miniperl
-Dhostgenerate=/path/to/second/build/dir/generate_uudmap

Much less commonly, if you are cross-compiling from an ASCII host to an EBCDIC target, or vise versa, you'll have to pass -Uhostgenerate to Configure, to signify that you want to build a generate_uudmap binary that, during make, will be run on the target system.

make test

This will run the regression tests on the perl you just made. If 'make test' doesn't say "All tests successful" then something went wrong.

Note that you can't run the tests in background if this disables opening of /dev/tty. You can use 'make test-notty' in that case but a few tty tests will be skipped.

What if make test doesn't work?

If make test bombs out, just cd to the t directory and run ./TEST by hand to see if it makes any difference.

One way to get more detailed information about failed tests and individual subtests is to run the harness from the t directory:

cd t ; ./perl harness <list of tests>

(this assumes that most basic tests succeed, since harness uses complicated constructs). If no list of tests is provided, harness will run all tests.

If individual tests fail, you can often run them by hand (from the main perl directory), e.g.,

./perl -I. -MTestInit t/op/groups.t

You should also read the individual tests to see if there are any helpful comments that apply to your system. You may also need to setup your shared library path if you get errors like:

/sbin/loader: Fatal Error: cannot map libperl.so

The file t/README in the t subdirectory contains more information about running and modifying tests.

See "Building a shared Perl library" earlier in this document.

locale

Note: One possible reason for errors is that some external programs may be broken due to the combination of your environment and the way 'make test' exercises them. For example, this may happen if you have one or more of these environment variables set: LC_ALL LC_CTYPE LC_COLLATE LANG. In some versions of UNIX, the non-English locales are known to cause programs to exhibit mysterious errors.

If you have any of the above environment variables set, please try

setenv LC_ALL C

(for C shell) or

LC_ALL=C;export LC_ALL

for Bourne or Korn shell) from the command line and then retry make test. If the tests then succeed, you may have a broken program that is confusing the testing. Please run the troublesome test by hand as shown above and see whether you can locate the program. Look for things like: exec, `backquoted command`, system, open("|...") or open("...|"). All these mean that Perl is trying to run some external program.

Timing problems

Several tests in the test suite check timing functions, such as sleep(), and see if they return in a reasonable amount of time. If your system is quite busy and doesn't respond quickly enough, these tests might fail. If possible, try running the tests again with the system under a lighter load. These timing-sensitive and load-sensitive tests include t/op/alarm.t, dist/Time-HiRes/t/alarm.t, dist/Time-HiRes/t/clock.t, dist/Time-HiRes/t/itimer.t, dist/Time-HiRes/t/usleep.t, dist/threads-shared/t/waithires.t, dist/threads-shared/t/stress.t, lib/Benchmark.t, lib/Memoize/t/expmod_t.t, and lib/Memoize/t/speed.t.

You might also experience some failures in t/op/stat.t if you build perl on an NFS filesystem, if the remote clock and the system clock are different.

Out of memory

On some systems, particularly those with smaller amounts of RAM, some of the tests in t/op/pat.t may fail with an "Out of memory" message. For example, on my SparcStation IPC with 12 MB of RAM, in perl5.5.670, test 85 will fail if run under either t/TEST or t/harness.

Try stopping other jobs on the system and then running the test by itself:

./perl -I. -MTestInit t/op/pat.t

to see if you have any better luck. If your perl still fails this test, it does not necessarily mean you have a broken perl. This test tries to exercise the regular expression subsystem quite thoroughly, and may well be far more demanding than your normal usage.

libgcc_s.so.1: cannot open shared object file

This message has been reported on gcc-3.2.3 and earlier installed with a non-standard prefix. Setting the LD_LIBRARY_PATH environment variable (or equivalent) to include gcc's lib/ directory with the libgcc_s.so.1 shared library should fix the problem.

Failures from lib/File/Temp/t/security saying "system possibly insecure"

First, such warnings are not necessarily serious or indicative of a real security threat. That being said, they bear investigating.

Note that each of the tests is run twice. The first time is in the directory returned by File::Spec->tmpdir() (often /tmp on Unix systems), and the second time in the directory from which the test was run (usually the 't' directory, if the test was run as part of 'make test').

The tests may fail for the following reasons:

(1) If the directory the tests are being run in is owned by somebody other than the user running the tests, or by root (uid 0).

This failure can happen if the Perl source code distribution is unpacked in such a way that the user IDs in the distribution package are used as-is. Some tar programs do this.

(2) If the directory the tests are being run in is writable by group or by others, and there is no sticky bit set for the directory. (With UNIX/POSIX semantics, write access to a directory means the right to add or remove files in that directory. The 'sticky bit' is a feature used in some UNIXes to give extra protection to files: if the bit is set for a directory, no one but the owner (or root) can remove that file even if the permissions would otherwise allow file removal by others.)

This failure may or may not be a real problem: it depends on the permissions policy used on this particular system. This failure can also happen if the system either doesn't support the sticky bit (this is the case with many non-UNIX platforms: in principle File::Temp should know about these platforms and skip the tests), or if the system supports the sticky bit but for some reason or reasons it is not being used. This is, for example, the case with HP-UX: as of HP-UX release 11.00, the sticky bit is very much supported, but HP-UX doesn't use it on its /tmp directory as shipped. Also, as with the permissions, some local policy might dictate that the stickiness is not used.

(3) If the system supports the POSIX 'chown giveaway' feature and if any of the parent directories of the temporary file back to the root directory are 'unsafe', using the definitions given above in (1) and (2). For Unix systems, this is usually not an issue if you are building on a local disk. See the documentation for the File::Temp module for more information about 'chown giveaway'.

See the documentation for the File::Temp module for more information about the various security aspects of temporary files.

The core distribution can now run its regression tests in parallel on Unix-like platforms. Instead of running make test, set TEST_JOBS in your environment to the number of tests to run in parallel, and run make test_harness. On a Bourne-like shell, this can be done as

TEST_JOBS=3 make test_harness  # Run 3 tests in parallel

An environment variable is used, rather than parallel make itself, because TAP::Harness needs to be able to schedule individual non-conflicting test scripts itself, and there is no standard interface to make utilities to interact with their job schedulers.

make install

This will put perl into the public directory you specified to Configure; by default this is /usr/local/bin. It will also try to put the man pages in a reasonable place. It will not nroff the man pages, however. You may need to be root to run make install. If you are not root, you must still have permission to install into the directories in question and you should ignore any messages about chown not working.

If "make install" just says "'install' is up to date" or something similar, you may be on a case-insensitive filesystems such as Mac's HFS+, and you should say "make install-all". (This confusion is brought to you by the Perl distribution having a file called INSTALL.)

Installing perl under different names

If you want to install perl under a name other than "perl" (for example, when installing perl with special features enabled, such as debugging), indicate the alternate name on the "make install" line, such as:

make install PERLNAME=myperl

You can separately change the base used for versioned names (like "perl5.8.9") by setting PERLNAME_VERBASE, like

make install PERLNAME=perl5 PERLNAME_VERBASE=perl

This can be useful if you have to install perl as "perl5" (e.g. to avoid conflicts with an ancient version in /usr/bin supplied by your vendor). Without this the versioned binary would be called "perl55.8.8".

Installing perl under a different directory

You can install perl under a different destination directory by using the DESTDIR variable during make install, with a command like

make install DESTDIR=/tmp/perl5

DESTDIR is automatically prepended to all the installation paths. See the example in "DESTDIR" above.

Installed files

If you want to see exactly what will happen without installing anything, you can run

./perl installperl -n
./perl installman -n

make install will install the following:

binaries

        perl,
            perl5.n.n   where 5.n.n is the current release number.  This
                        will be a link to perl.

scripts

        cppstdin        This is used by the deprecated switch perl -P,
                        if your cc -E can't read from stdin.
        corelist        Shows versions of modules that come with
                    different
                        versions of perl.
        cpan            The CPAN shell.
        enc2xs          Encoding module generator.
        h2ph            Extract constants and simple macros from C
                    headers.
        h2xs            Converts C .h header files to Perl extensions.
        instmodsh       A shell to examine installed modules.
        libnetcfg       Configure libnet.
        perlbug         Tool to report bugs in Perl.
        perldoc         Tool to read perl's pod documentation.
        perlivp         Perl Installation Verification Procedure.
        piconv          A Perl implementation of the encoding conversion
                        utility iconv.
        pl2pm           Convert Perl 4 .pl files to Perl 5 .pm modules.
        pod2html,       Converters from perl's pod documentation format
        pod2man,
        pod2text,
        pod2usage
        podchecker      POD syntax checker.
        podselect       Prints sections of POD documentation.
        prove           A command-line tool for running tests.
        psed            A Perl implementation of sed.
        ptar            A Perl implementation of tar.
        ptardiff        A diff for tar archives.
        ptargrep        A grep for tar archives.
        shasum          A tool to print or check SHA checksums.
        splain          Describe Perl warnings and errors.
        xsubpp          Compiler to convert Perl XS code into C code.
        zipdetails      display the internal structure of zip files

library files

                        in $privlib and $archlib specified to
                        Configure, usually under /usr/local/lib/perl5/.

documentation

        man pages       in $man1dir, usually /usr/local/man/man1.
        module man
        pages           in $man3dir, usually /usr/local/man/man3.
        pod/*.pod       in $privlib/pod/.

installperl will also create the directories listed above in "Installation Directories".

Perl's *.h header files and the libperl library are also installed under $archlib so that any user may later build new modules, run the optional Perl compiler, or embed the perl interpreter into another program even if the Perl source is no longer available.

Installing with a version-specific suffix

Sometimes you only want to install the perl distribution with a version-specific suffix. For example, you may wish to install a newer version of perl alongside an already installed production version. To only install the version-specific parts of the perl installation, run

Configure -Dversiononly

or answer 'y' to the appropriate Configure prompt. Alternatively, you can just manually run

./perl installperl -v

and skip installman altogether.

See also "Maintaining completely separate versions" for another approach.

cd /usr/include; h2ph *.h sys/*.h

Some perl scripts need to be able to obtain information from the system header files. This command will convert the most commonly used header files in /usr/include into files that can be easily interpreted by perl. These files will be placed in the architecture-dependent library ($archlib) directory you specified to Configure.

Note: Due to differences in the C and perl languages, the conversion of the header files is not perfect. You will probably have to hand-edit some of the converted files to get them to parse correctly. For example, h2ph breaks spectacularly on type casting and certain structures.

installhtml --help

Some sites may wish to make perl documentation available in HTML format. The installhtml utility can be used to convert pod documentation into linked HTML files and install them.

Currently, the supplied ./installhtml script does not make use of the html Configure variables. This should be fixed in a future release.

The following command-line is an example of one used to convert perl documentation:

./installhtml                   \
    --podroot=.                 \
    --podpath=lib:ext:pod:vms   \
    --recurse                   \
    --htmldir=/perl/nmanual     \
    --htmlroot=/perl/nmanual    \
    --splithead=pod/perlipc     \
    --splititem=pod/perlfunc    \
    --verbose

See the documentation in installhtml for more details. It can take many minutes to execute a large installation and you should expect to see warnings like "no title", "unexpected directive" and "cannot resolve" as the files are processed. We are aware of these problems (and would welcome patches for them).

You may find it helpful to run installhtml twice. That should reduce the number of "cannot resolve" warnings.

cd pod && make tex && (process the latex files)

Some sites may also wish to make the documentation in the pod/ directory available in TeX format. Type

(cd pod && make tex && <process the latex files>)

Starting all over again

If you wish to rebuild perl from the same build directory, you should clean it out with the command

make distclean

or

make realclean

The only difference between the two is that make distclean also removes your old config.sh and Policy.sh files. (A plain 'make clean' is now equivalent to 'make realclean'.)

If you are upgrading from a previous version of perl, or if you change systems or compilers or make other significant changes, or if you are experiencing difficulties building perl, you should not reuse your old config.sh.

If your reason to reuse your old config.sh is to save your particular installation choices, then you can probably achieve the same effect by using the Policy.sh file. See the section on "Site-wide Policy settings" above.

Reporting Problems

Please report problems to the GitHub issue tracker at https://github.com/Perl/perl5/issues, which will ask for the appropriate summary configuration information about your perl, which may help us track down problems far more quickly. But first you should read the advice in this file, carefully re-read the error message and check the relevant manual pages on your system, as these may help you find an immediate solution. Once you've exhausted the documentation, please report bugs to us using the GitHub tracker.

The summary configuration information can be printed with perl -V. If the install fails, or you want to report problems with make test without installing perl, then you can run it by hand from this source directory with ./perl -V.

If the build fails too early to run perl, then please run the ./myconfig shell script, and include its output along with an accurate description of your problem.

If Configure itself fails, and does not generate a config.sh file (needed to run ./myconfig), then please open an issue with the description of how Configure fails along with details of your system -- for example the output from running uname -a.

Please try to make your message brief but clear. Brief, clear bug reports tend to get answered more quickly. Please don't worry if your written English is not great -- what matters is how well you describe the important technical details of the problem you have encountered, not whether your grammar and spelling is flawless.

Trim out unnecessary information. Do not include large files (such as config.sh or a complete Configure or make log) unless absolutely necessary. Do not include a complete transcript of your build session. Just include the failing commands, the relevant error messages, and whatever preceding commands are necessary to give the appropriate context.

If the bug you are reporting has security implications which make it inappropriate to send to a public issue tracker, then see "SECURITY VULNERABILITY CONTACT INFORMATION" in perlsec for details of how to report the issue.

If you are unsure what makes a good bug report please read "How to report Bugs Effectively" by Simon Tatham: https://www.chiark.greenend.org.uk/~sgtatham/bugs.html

Coexistence with earlier versions of perl 5

Perl 5.39.0 is not binary compatible with earlier versions of Perl. In other words, you will have to recompile your XS modules.

In general, you can usually safely upgrade from one stable version of Perl (e.g. 5.30.0) to another similar minor version (e.g. 5.30.1) without re-compiling all of your extensions. You can also safely leave the old version around in case the new version causes you problems for some reason.

Usually, most extensions will probably not need to be recompiled to be used with a newer version of Perl. Here is how it is supposed to work. (These examples assume you accept all the Configure defaults.)

Suppose you already have version 5.8.7 installed. The directories searched by 5.8.7 are typically like:

/usr/local/lib/perl5/5.8.7/$archname
/usr/local/lib/perl5/5.8.7
/usr/local/lib/perl5/site_perl/5.8.7/$archname
/usr/local/lib/perl5/site_perl/5.8.7

Now, suppose you install version 5.8.8. The directories searched by version 5.8.8 will be:

/usr/local/lib/perl5/5.8.8/$archname
/usr/local/lib/perl5/5.8.8
/usr/local/lib/perl5/site_perl/5.8.8/$archname
/usr/local/lib/perl5/site_perl/5.8.8

/usr/local/lib/perl5/site_perl/5.8.7/$archname
/usr/local/lib/perl5/site_perl/5.8.7
/usr/local/lib/perl5/site_perl/

Notice the last three entries -- Perl understands the default structure of the $sitelib directories and will look back in older, compatible directories. This way, modules installed under 5.8.7 will continue to be usable by 5.8.7 but will also accessible to 5.8.8. Further, suppose that you upgrade a module to one which requires features present only in 5.8.8. That new module will get installed into /usr/local/lib/perl5/site_perl/5.8.8 and will be available to 5.8.8, but will not interfere with the 5.8.7 version.

The last entry, /usr/local/lib/perl5/site_perl/, is there so that 5.6.0 and above will look for 5.004-era pure perl modules.

Lastly, suppose you now install 5.10.0, which is not binary compatible with 5.8.x. The directories searched by 5.10.0 (if you don't change the Configure defaults) will be:

/usr/local/lib/perl5/5.10.0/$archname
/usr/local/lib/perl5/5.10.0
/usr/local/lib/perl5/site_perl/5.10.0/$archname
/usr/local/lib/perl5/site_perl/5.10.0

/usr/local/lib/perl5/site_perl/5.8.8

/usr/local/lib/perl5/site_perl/5.8.7

/usr/local/lib/perl5/site_perl/

Note that the earlier $archname entries are now gone, but pure perl modules from earlier versions will still be found.

This way, you can choose to share compatible extensions, but also upgrade to a newer version of an extension that may be incompatible with earlier versions, without breaking the earlier versions' installations.

Maintaining completely separate versions

Many users prefer to keep all versions of perl in completely separate directories. This guarantees that an update to one version won't interfere with another version. (The defaults guarantee this for libraries after 5.6.0, but not for executables. TODO?) One convenient way to do this is by using a separate prefix for each version, such as

sh Configure -Dprefix=/opt/perl5.39.0

and adding /opt/perl5.39.0/bin to the shell PATH variable. Such users may also wish to add a symbolic link /usr/local/bin/perl so that scripts can still start with #!/usr/local/bin/perl.

Others might share a common directory for maintenance sub-versions (e.g. 5.10 for all 5.10.x versions), but change directory with each major version.

If you are installing a development subversion, you probably ought to seriously consider using a separate directory, since development subversions may not have all the compatibility wrinkles ironed out yet.

Upgrading from 5.38.0 or earlier

Perl 5.39.0 may not be binary compatible with Perl 5.38.0 or earlier Perl releases. Perl modules having binary parts (meaning that a C compiler is used) will have to be recompiled to be used with 5.39.0. If you find you do need to rebuild an extension with 5.39.0, you may safely do so without disturbing the older installations. (See "Coexistence with earlier versions of perl 5" above.)

See your installed copy of the perllocal.pod file for a (possibly incomplete) list of locally installed modules. Note that you want perllocal.pod, not perllocale.pod, for installed module information.

Minimizing the Perl installation

The following section is meant for people worrying about squeezing the Perl installation into minimal systems (for example when installing operating systems, or in really small filesystems).

Leaving out as many extensions as possible is an obvious way: Encode, with its big conversion tables, consumes a lot of space. On the other hand, you cannot throw away everything. The Fcntl module is pretty essential. If you need to do network programming, you'll appreciate the Socket module, and so forth: it all depends on what do you need to do.

In the following we offer two different slimmed down installation recipes. They are informative, not normative: the choice of files depends on what you need.

Firstly, the bare minimum to run this script

use strict;
use warnings;
foreach my $f (</*>) {
   print("$f\n");
}

in Linux with perl-5.39.0 is as follows (under $Config{prefix}):

./bin/perl
./lib/perl5/5.39.0/strict.pm
./lib/perl5/5.39.0/warnings.pm
./lib/perl5/5.39.0/i686-linux/File/Glob.pm
./lib/perl5/5.39.0/feature.pm
./lib/perl5/5.39.0/XSLoader.pm
./lib/perl5/5.39.0/i686-linux/auto/File/Glob/Glob.so

Secondly, for perl-5.10.1, the Debian perl-base package contains 591 files, (of which 510 are for lib/unicore) totaling about 3.5MB in its i386 version. Omitting the lib/unicore/* files for brevity, the remaining files are:

/usr/bin/perl
/usr/bin/perl5.10.1
/usr/lib/perl/5.10.1/Config.pm
/usr/lib/perl/5.10.1/Config_git.pl
/usr/lib/perl/5.10.1/Config_heavy.pl
/usr/lib/perl/5.10.1/Cwd.pm
/usr/lib/perl/5.10.1/DynaLoader.pm
/usr/lib/perl/5.10.1/Errno.pm
/usr/lib/perl/5.10.1/Fcntl.pm
/usr/lib/perl/5.10.1/File/Glob.pm
/usr/lib/perl/5.10.1/Hash/Util.pm
/usr/lib/perl/5.10.1/IO.pm
/usr/lib/perl/5.10.1/IO/File.pm
/usr/lib/perl/5.10.1/IO/Handle.pm
/usr/lib/perl/5.10.1/IO/Pipe.pm
/usr/lib/perl/5.10.1/IO/Seekable.pm
/usr/lib/perl/5.10.1/IO/Select.pm
/usr/lib/perl/5.10.1/IO/Socket.pm
/usr/lib/perl/5.10.1/IO/Socket/INET.pm
/usr/lib/perl/5.10.1/IO/Socket/UNIX.pm
/usr/lib/perl/5.10.1/List/Util.pm
/usr/lib/perl/5.10.1/POSIX.pm
/usr/lib/perl/5.10.1/Scalar/Util.pm
/usr/lib/perl/5.10.1/Socket.pm
/usr/lib/perl/5.10.1/XSLoader.pm
/usr/lib/perl/5.10.1/auto/Cwd/Cwd.so
/usr/lib/perl/5.10.1/auto/DynaLoader/autosplit.ix
/usr/lib/perl/5.10.1/auto/DynaLoader/dl_expandspec.al
/usr/lib/perl/5.10.1/auto/DynaLoader/dl_find_symbol_anywhere.al
/usr/lib/perl/5.10.1/auto/DynaLoader/dl_findfile.al
/usr/lib/perl/5.10.1/auto/Fcntl/Fcntl.so
/usr/lib/perl/5.10.1/auto/File/Glob/Glob.so
/usr/lib/perl/5.10.1/auto/Hash/Util/Util.so
/usr/lib/perl/5.10.1/auto/IO/IO.so
/usr/lib/perl/5.10.1/auto/List/Util/Util.so
/usr/lib/perl/5.10.1/auto/POSIX/POSIX.so
/usr/lib/perl/5.10.1/auto/POSIX/autosplit.ix
/usr/lib/perl/5.10.1/auto/POSIX/load_imports.al
/usr/lib/perl/5.10.1/auto/Socket/Socket.so
/usr/lib/perl/5.10.1/lib.pm
/usr/lib/perl/5.10.1/re.pm
/usr/share/doc/perl/AUTHORS.gz
/usr/share/doc/perl/Documentation
/usr/share/doc/perl/README.Debian
/usr/share/doc/perl/changelog.Debian.gz
/usr/share/doc/perl/copyright
/usr/share/lintian/overrides/perl-base
/usr/share/man/man1/perl.1.gz
/usr/share/man/man1/perl5.10.1.1.gz
/usr/share/perl/5.10.1/AutoLoader.pm
/usr/share/perl/5.10.1/Carp.pm
/usr/share/perl/5.10.1/Carp/Heavy.pm
/usr/share/perl/5.10.1/Exporter.pm
/usr/share/perl/5.10.1/Exporter/Heavy.pm
/usr/share/perl/5.10.1/File/Spec.pm
/usr/share/perl/5.10.1/File/Spec/Unix.pm
/usr/share/perl/5.10.1/FileHandle.pm
/usr/share/perl/5.10.1/Getopt/Long.pm
/usr/share/perl/5.10.1/IPC/Open2.pm
/usr/share/perl/5.10.1/IPC/Open3.pm
/usr/share/perl/5.10.1/SelectSaver.pm
/usr/share/perl/5.10.1/Symbol.pm
/usr/share/perl/5.10.1/Text/ParseWords.pm
/usr/share/perl/5.10.1/Text/Tabs.pm
/usr/share/perl/5.10.1/Text/Wrap.pm
/usr/share/perl/5.10.1/Tie/Hash.pm
/usr/share/perl/5.10.1/attributes.pm
/usr/share/perl/5.10.1/base.pm
/usr/share/perl/5.10.1/bytes.pm
/usr/share/perl/5.10.1/bytes_heavy.pl
/usr/share/perl/5.10.1/constant.pm
/usr/share/perl/5.10.1/fields.pm
/usr/share/perl/5.10.1/integer.pm
/usr/share/perl/5.10.1/locale.pm
/usr/share/perl/5.10.1/overload.pm
/usr/share/perl/5.10.1/strict.pm
/usr/share/perl/5.10.1/unicore/*
/usr/share/perl/5.10.1/utf8.pm
/usr/share/perl/5.10.1/utf8_heavy.pl
/usr/share/perl/5.10.1/vars.pm
/usr/share/perl/5.10.1/warnings.pm
/usr/share/perl/5.10.1/warnings/register.pm

A nice trick to find out the minimal set of Perl library files you will need to run a Perl program is

perl -e 'do "prog.pl"; END { print "$_\n" for sort keys %INC }'

(this will not find libraries required in runtime, unfortunately, but it's a minimal set) and if you want to find out all the files you can use something like the below

strace perl -le 'do "x.pl"' 2>&1 \
                            | perl -nle '/^open\(\"(.+?)"/ && print $1'

(The 'strace' is Linux-specific, other similar utilities include 'truss' and 'ktrace'.)

-DNO_MATHOMS

If you configure perl with -Accflags=-DNO_MATHOMS, the functions from mathoms.c will not be compiled in. Those functions are no longer used by perl itself; for source compatibility reasons, though, they weren't completely removed.

-DNO_PERL_INTERNAL_RAND_SEED

If you configure perl with -Accflags=-DNO_PERL_INTERNAL_RAND_SEED, perl will ignore the PERL_INTERNAL_RAND_SEED environment variable.

-DNO_PERL_RAND_SEED

If you configure perl with -Accflags=-DNO_PERL_RAND_SEED, perl will ignore the PERL_RAND_SEED environment variable.

DOCUMENTATION

Read the manual entries before running perl. The main documentation is in the pod/ subdirectory and should have been installed during the build process. Type man perl to get started. Alternatively, you can type perldoc perl to use the supplied perldoc script. This is sometimes useful for finding things in the library modules.

AUTHOR

Original author: Andy Dougherty [email protected] , borrowing very heavily from the original README by Larry Wall, with lots of helpful feedback and additions from the [email protected] folks.

If you have problems, corrections, or questions, please see "Reporting Problems" above.

REDISTRIBUTION

This document is part of the Perl package and may be distributed under the same terms as perl itself, with the following additional request: If you are distributing a modified version of perl (perhaps as part of a larger package) please do modify these installation instructions and the contact information to match your distribution. Additional information for packagers is in PACKAGING.