several orders of magnitude and 'make index' now takes 30 minutes or so
instead of several days on my test machine. The approach now is to take
one pass through every package and extract some key information including
the explicitly listed dependencies. After the data is extracted, the
dependencies are flattened in one step which avoids the extremely
inefficient recursive make that was previously used.
shared object "provides" and "requires" information. This is only
turned on just now if ${CHECK_SHLIBS} is set to "YES".
Example output:
% pkg_info -B libutf | grep '^PROVIDES'
PROVIDES=/usr/pkg/lib/libutf.so.2
% pkg_info -B ssam | grep '^REQUIRES'
REQUIRES=/usr/lib/libc.so.12
REQUIRES=/usr/pkg/lib/libutf.so.2
%
- add ONLY_FOR_ARCHS and NOT_FOR_OPSYS to the list of variables that goes
in the print-summary-data output. This is the first step in speeding up
the 'make index' process.
to link pthreaded code, ala PTHREAD_CFLAGS.
this adds support for systems which have no libpthread, but rather,
have pthreads functions in libc_r, such as FreeBSD 5.x.
export PTHREAD_CFLAGS and PTHREAD_LDFLAGS via MAKE_ENV.
so compiler-dependent settings can be set later.
USE_GCC2 - GNU Compiler Collection 2.x (default)
USE_GCC3 - GNU Compiler Collection 3.x (used already)
USE_MIPSPRO - Silicon Graphics, Inc. MIPSpro Compiler
USE_SUNPRO - Sun Microsystems, Inc. WorkShop/Forte/Sun ONE Studio
Compiler Collection
(Intel compilers on Linux, anyone? :) )
do-configure target. To do this I've introduced a new target,
pre-configure-overrides, that depends on ${_CONFIGURE_PREREQ}. It's run
after pre-configure but before do-configure.
This fixes pkg/22077, and allows benchmarks/benchfft to build correctly on
a -current cats box.
Reviewed by wiz@
in `makeinfo --version'.
This is needed for older makeinfo which output thing like:
makeinfo (GNU texinfo 3.12) 1.68
Copyright (C) 1998 Free Software Foundation, Inc.
....
This should fix the problem encountered by Thorsten Frueauf on 1.5.4_ALPHA
(I reproduced it on stock 1.5.3) and reported in private email via
jschauma@.
arguments to makeinfo(1). This was made necessary by the fact that under
IRIX64 makeinfo(1) does not use a default split-size of 50000, which breaks
our PLISTs. After feedback from seb@, initialize MAKEINFO_ARGS per default
to _OPSYS_MAKEINFO_ARGS, which is set in defs.<OPSYS>.mk (currently only
for IRIX using 64-bit ABI).
when a package use the buildlink2 framework but does not define
USE_MAKEINFO. Well it was not after all.
This caused annoying messages because missing's commands emit annoying error
messages when the script is invoked with only 'makeinfo' as argument
(typically run this way by configure scripts). And more it does not
handle makeinfo's '--output=...' argument.
I first thought that it could be used as a nice way to get ride of the need
for makeinfo when it was only dubiously run during build or installation
of a package. But it also has the annoying behavior of creating empty files
because of the typical automake generated Makefile target for info file
build.
Making the buildlink2's makeinfo hiding script only logging an error
and doing 'exit 1' is actually a better tool to spot the need for makeinfo.
directory file remove it if it is empty i.e. if it contains only one
menu/entry line i.e. only one line starting with '*'.
This allow INFO_DIR to be set to a package specific directory while
not registering in the PLIST the Info directory file (${INFOR_DIR}/dir).
Registering such an Info directory file in the PLIST is not really
possible as a md5 signature of it holding the registered info entries
would be stored in PKG_DBDIR. At deinstall time this signature would not be
matched because the DEINSTALL script would have modified it by
un-registering the info files from it. Hence the package removal would
be reported as incomplete...
You are probably right if this looks like a hack to you...
makeinfo commands.
The goal of the new framework is twofold:
- reduce the number of '@exec' and '@unexec' in PLIST by
using INSTALL/DEINSTALL scripts to handle entries addition/removal
Info directory file.
- achieve lighter dependencies by avoiding unnecessary run-time
dependency on the gtexinfo package and if needed with the help of the
standalone install-info command provided by the recently imported package
pkgtools/pkg_install-info.
A package must be sightly updated to use this new framework and
must define the variable USE_NEW_TEXINFO. This variable will
be removed from the pkgsrc tree when all package would have been
updated.
For details see section 10.24 of Packages.txt, comments in
mk/{texinfo.mk,buildlink2/bsd.buildlink2.mk} and upcoming mail to
<tech-pkg at netbsd dot org>.
followup to PR pkg/20529:
Some libraries (in this example, a given Linux's native libpthread) don't
come with a version number at all and/or are in fact symlinks. Test for
this and act accordingly in fake-la.
This allows packages that use pthread to buildlink under various Linuxen.
Should close PR pkg/20529 and PR pkg/21854. Thanks, Joachim.
if it's up to date. For dependencies (including the current package)
that are not up to date (or are not installed), a one line note is
printed thusly:
% cd audio/xmradio/
% make show-needs-update
audio/mad => mad-0.14.2b => needs update to mad-0.14.2bnb1
audio/xmradio => (none) => needs install of xmradio-1.2
that
% make
===> do-fetch [digest-20021220] ===> Checking for vulnerabilities in digest-20021220
=> Lock acquired on behalf of process 356
=> Lock released on behalf of process 356
becomes
% make
===> do-fetch [digest-20021220] ===> Checking for vulnerabilities in digest-20021220
=> Lock acquired on behalf of process 814
=> Lock released on behalf of process 814
We introduce two new variables: ABI and IMAKEOPTS
ABI determines the correct ABI to use and is set during bootstrapping
from /etc/compiler.defaults.
IMAKEOPTS are the options passed to imake to allow it to set the right
definitions and find the right paths etc. Use IMAKEOPTS with SunOS, too.
IRIX will always require xpkgwedge, since it's got a bogus X11BASE.
When linking against libXaw, make sure we get the right ABI.
care not to blow away our bootstrap-pkgsrc stuff in the initial phase.
Also mark devel/bmake and devel/mk-files as broken on non-NetBSD so as not
to blow away our precious files from the bootstrap process in the middle
of a bulk-build. Now let's see if bulk-building works on Linux...
is "-Wl,--whole-archive -lgcc -Wl,--no-whole-archive" on
NetBSD-1.5* and empty otherwise.
This is needed on build of programs which dlopen() extension
libraries.
(Some pkgs do the libgcc linking already, but independantly
of the OS version.)
defs.*.mk files) when running the configure script. Most of them look for
an awk parser, and this way they will always pick the right one without
having to check all the possibilities.
