$Id: README-CVS,v 1.21 2007/03/12 04:54:24 jhs Exp $
HOW TO BUILD PIKE FROM CVS
If you like to live at the bleeding edge you can download Pike from
CVS with all the latest additions from the developers. There are two
major branches in the archive, the latest stable branch and latest
development branch. Stable versions have an even minor version
number, i.e. 7.0.x, 7.2.x, 7.4.x, whereas the development branches
have an odd minor version.
Keep in mind that the CVS versions are under heavy development and
have not been tested nearly as well as the offcial releases. You use
the code AT YOUR OWN RISK.
Building from CVS requires the same tools as building from a source
distribution (gnu m4, bison and a C compiler, suggested also GNU make
and libz), and then some. In addition to thosee requirements, you
need a working Pike, autoconf and gcc (to generate the dependency
files; another compiler can be used to do the actual compilation).
Not all autoconf versions meet Pike's requirements. Autoconf version
2.13 and 2.52 are known to work. Versions 2.53 through at least 2.57
are known to not work.
CHECKING OUT PIKE FROM CVS
1. Get a recent version of cvs.
2. Login to the CVS server:
cvs -d :pserver:email@example.com:/cvs login
Just hit enter on the password prompt.
3. Check out the source:
cvs -z3 -d :pserver:firstname.lastname@example.org:/cvs co Pike/7.7
substitute 7.7 for whatever version you want to get.
The top-level makefile (in this directory, not the src directory) has
all the magic you need to build Pike directly from CVS. Just type
'make'. It is preferable to build from the toplevel since it avoids
contaminating the source tree with object files and other generated
Other interesting make targets are:
install Compile and install in default location.
install_interactive Interactive install.
tinstall Test install, i.e. install in build directory.
verify Do a test install and run the testsuite with the
just_verify Run the testsuite directly with the Pike binary in
the build tree.
run_hilfe Run hilfe without installing Pike.
pike Build only the Pike core, do not recurse into the
documentation Build the reference documentation from the
source. See the refdoc subdirectory.
depend Build the files that tracks dependencies between
the source files. This is necessary to ensure
correct rebuilding if some of the source files
change, but not if you only intend to use the
build tree once. It's not run by default.
source Prepare the source tree for compilation without
the need for a preexisting installed Pike.
force_autoconfig Force a build of the configure scripts. This is
useful e.g. if a new module directory is added in
force_configure Force configure to be run (recursively). If
you've installed a new library and want Pike to
detect it, then the simplest way is to remove
config.cache in the build directory (or perhaps
just delete the relevant variables in it) and then
use this target.
dump_modules Dump the Pike modules directly in the build tree.
That makes Pike load faster if it's run directly
from there, e.g. through the bin/pike script (see
below). These dumped modules are not used for
anything else. After this has been run once, any
changed Pike modules will be redumped
automatically by the main build targets.
undump_modules Remove any modules dumped by dump_modules, and
remove the redump step described above.
force_dump_modules Force all Pike modules to be redumped, not just
those whose source files have changed.
snapshot Create a snapshot export tarball.
export Create a source dist and bump up the build number
(if you have cvs write access). Please do not
check in the generated files.
clean Remove all the built binary files.
cvsclean Remove all files that are generated automatically,
i.e. bring the tree back to the state as if it
was checked out from the CVS.
CONFIGURE OPTIONS AND BUILD VARIABLES
If you want to pass arguments to the configure script (see below), the
simplest way is to use the CONFIGUREARGS variable, like this:
make CONFIGUREARGS="--prefix=/usr/local/my-pike --with-security"
The arguments passed through CONFIGUREARGS are remembered in the build
tree and reused if CONFIGUREARGS is undefined or the empty string.
You therefore don't need to repeat them every time, but you can still
change them later if you like. There's a special case for the --help
argument: If CONFIGUREARGS is set to '--help' then the help text from
the configure script is shown and nothing else is done, and the stored
CONFIGUREARGS setting isn't affected.
The build targets also creates a script 'pike' in the bin subdirectory
which runs the built Pike directly without installing it first. If
you want to use Pike this way (which is mainly useful if you update
from CVS often), you should consider doing 'make dump_modules' to make
it start faster.
Some options for the configure script are:
--prefix=/foo/bar if you want to install Pike in /foo/bar,
default is /usr/local.
--without-gdbm compile without gdbm support
--without-bignums disable support for large integers
--without-gmp compile without gmp support (implies
--with-rtldebug compile with runtime debug checks
--without-cdebug compile without debug symbols (-g)
--with-debug same as --with-rtldebug --with-cdebug
--without-debug same as --without-rtldebug --without-cdebug
--without-copt compile without -O2
--without-threads compile without threads support (see
also the section 'If It Doesn't Work' below)
--without-zlib compile without gzip compression libary
--without-dynamic-modules compile statically, no dynamic loading
used (makes the binary larger)
--without-mysql compile without mysql support
--with-profiling enables profiling Pike code but slows
down interpreter a little
--with-poll use poll instead of select
--with-dmalloc compile with memory tracking, makes Pike
very slow, use for debugging only.
--with-security enable internal object security system
You might also want to set the following environment variables:
CFLAGS Put extra flags for your C compiler here.
CPPFLAGS Put extra flags for your C preprocessor here
(such as -I/usr/gnu/include)
LDFLAGS Put extra flags to your linker here, such as
-L/usr/gnu/lib and -R/usr/gnu/lib
Instructions if you want to do the build more manually:
1. cd src ; ./run_autoconfig
This creates configure files and Makefile.in files.
2. Create a build directory an cd to it. Do NOT build in the source
dir, doing so will make it impossible to do 'make export' later.
3. Run the newly created configure file located in the src dir from
the build dir. Make sure to use an absolute path! This creates the
Makefiles you need, e.g. Makefile from Makefile.in and machine.h
from machine.h.in. If you don't use an absolute path the debug
information will be all warped...
4. If needed, edit config.h and Makefile to suit your purposes. We
have tried to make it so that you don't have to change config.h or
Makefile at all. If you need to do what you consider 'unnecessary
changes' then mail us and we'll try to fit it into configure. If
possible, use gnu make, gcc, gnu sed and bison.
5. Run 'make'
This builds Pike.
6. Optionally, run 'make verify' to check that the compiled driver
works as it should (might be a good idea). This will take a little
time and use quite a lot of memory, because the test program is
quite large. If everything works out fine no extra messages are
7) If you want to install Pike, write 'make install'. This will put
your Pike in <prefix>/pike/<version>/. This way, you can install
many Pike versions in parallell on the system if you want to. To
put it below <prefix> directly, as other packages usually do, run
'make INSTALLARGS="--traditional" install' instead.
After doing this, DO NOT commit the generated files. They are placed
in .cvsignore files so you shouldn't have to bother with them.
IF IT DOESN'T WORK
o Try again.
o Try running 'make depend'.
o Your sh might be too buggy to run ./configure (this is the case on
A/UX). Try using bash, zsh or possibly ksh. To use bash, first
run /bin/sh and type:
$ export CONFIG_SHELL
$ $CONFIG_SHELL ./configure
o If you are not using GNU make, compile in the source dir rather
than using a separate build dir.
o ./configure relies heavily on sed, if you have several sed in your
path try another sed (preferably gnu sed).
o configure might have done something wrong, check machine.h and
report any errors back to us.
o Your gmp/gdbm libraries might not be working or incorrectly
installed; start over by running configure with the appropriate
--without-xxx arguments. Also note that threads might give
problems with I/O and signals. If so you need to run configure
o Try a different compiler, malloc, compiler-compiler and/or make
(if you have any other).
If you find a bug in the interpreter, typically if Pike dumps core,
the first thing to do is to make sure it is compiled with the
--with-rtldebug configure flag. If not, reconfigure and recompile
with that and see if you get another error. When you've done this,
please report the bug to us at http://community.roxen.com/crunch/ and
include as much as you can muster of the following:
o The Pike version. (Try pike --version or look in src/version.h)
o What kind of system hardware/software you use (OS, compiler, etc.)
o The piece of code that crashes or bugs, preferably in a very
small Pike-script with the bug isolated. Please send a complete
running example of something that causes the bug.
o A description of what it is that bugs and when.
o If you know how, then also give us a backtrace and dump of vital
variables at the point of crash.
o Or, if you found the error and corrected it, just send us the
bugfix along with a description of what you did and why.