- Moved most of the guts of Makefile to Makefile.inc1 to become the
backend for the build system.
- The new Makefile doesn't suffer from problems including the wrong
sys.mk because it doesn't use anything in there or bsd.own.mk. So,
from now on, the proper build command is just `make world' (or
buildworld).
- The intermediate makefiles called Makefile.inc0 and Makefile.upgrade
fiddle with the OBJFORMAT and MAKEOBJDIRPREFIX variables so that
both aout and elf object trees can coexist. Makefile.upgrade contains
the aout->elf transition build.
- A cross build environment is now very close to reality. Specifying
TOOLDIR, setting OBJFORMAT and MAKEOBJDIRPREFIX allow that.
See the comments in Makefile for more info.
for building and installing a local sendmail.cf..
I'm a little nervous about the implications of having an obj dir built
under etc (to get to the obj dir for sendmail), but the make rules appear
to DTRT.
recently added definitions from sys.mk to bsd.own.mk. Include the
src-relative bsd.own.mk in src/Makefile to pick up all new definitions.
Don't check that MACHINE_ARCH is defined in src/Makefile, since it is
(and should have been) guaranteed to be defined.
working when the target system is not binary compatible. Use various
hacks to work around minor problems in the source and binary tree
layouts:
- caesar and strfile are built normally (the source layout is good),
then installed by copying them to ${WORLDTMP}/usr/bin (they are
installed in ${WORLDTMP}/usr/games, but I don't want to put that
in $PATH).
- colldef and mklocale are built and installed normally. Messy and
incomplete relative path searches for them and caesar and strfile
can now go away.
- internal tools that aren't installed are now built and left lying
around for the `make all' pass to use. If the target system is
not binary compatible, it is critical that these tools don't get
rebuilt. Cleaning of the obj tree before building the internal
tools should ensure this.
- most internal tools are built using internal build-tools targets,
but tn3270 is simpler for a change - it has all the tools in a
separate tree, so they can be built using `make all'.
build at least compile_et and lex, and although almost any version
of yacc could work, the version in -stable doesn't actually work
with -current makefiles because it doesn't support -o.
Submitted by: Ian Holland <ianh@tortuga.com.au>
- don't announce `mtree' as `mtools'.
- don't install to ${DESTDIR}/usr/sbin (which often doesn't exist if
DESTDIR is set and may be read-only if DESTDIR is not set).
- install to (${WORLDTMP}/usr/sbin so that the new mtree is actually
in $PATH if DESTDIR is set.
- don't use the host make or the host sys.mk. This is probably
unimportant.
- use a temporary obj dir like the one for `make'. This was mainly
necessary because I forgot to remove the MAKEOBJDIRPREFIX unsetting
which was just a bug for mtree. A non-quick fix would handle mtree
more like a bootstrap tool (the only additional complications are to
create ${BINDIR} and avoid excessive cleaning). Except a non-quick
fix would change much more.
(in particular, if DESTDIR is empty or "/"), then the host's ldconfig will
be the target's ldconfig by the time it is run.
Fixed disordering of env.
Don't know too much about libcrypt. Use exactly the same definition of
_libcrypt as lib/Makefile.
Don't build strip twice.
commit - don't wander off to bootstrap mtree and include in the middle
of bootstrapping lex, and don't forget what we were doing and build
some lex obj dirs twice.
crypt library should be used when building bootstrap-libraries. This
make it work on machines that don't have the secure directory.
Thanks to Paul Allenby <pallenby@mikom.csir.co.za> for bringing it to
my attention.
for the rest of the build. I'm not certain, but I think this determines
which crypt() goes into /sbin/init. This change shouldn't hurt anyway. :-)
Based on a suggestion by: bde
real source tree, the bootstrap target would attempt have cpio copy the
files over themselves, unlinking them first. I think this only happened
with make -DNOCLEAN world at the transition between a symlinked
objdir/tmp/usr/include/{sys,net,..} and real files.
with real copies. I'm sick of !@#&!^!@#*& mtree chowning directories in
my src/sys/* tree after it follows the symlinks. I still believe that
mtree is broken for doing this (introduced in mtree.c rev 1.5).
helpful to stop it running /usr/libexec/aout/as for real while bootstrapping.
Only force a strict path when we really have built all the tools in
$OBJDIR/tmp/usr/libexec/*.
Move a.out libraries to /usr/lib/aout to make space for ELF libs.
Make rtld usr /usr/lib/aout as default library path.
Make ldconfig reject /usr/lib as an a.out library path.
Fix various Makefiles for LIBDIR!=/usr/lib breakage.
This will after a make world & reboot give a system that no
longer uses /usr/lib/*, infact one could remove all the old
libraries there, they are not used anymore.
We are getting close to an ELF make world, but I'll let this
all settle for a week or two...
Move our old a.out utils to /usr/libexec/aout.
Enable binutils and put the utils in /usr/libexec/elf
Enable objformat, a little helper program that calls the right
utils based on /etc/objformat and $OBJFORMAT.
This will enable the ELF generating tools.
Remember that this is only step one, the system is still compiled
and run in a.out format ONLY.
Problem left to solve: The BSD manpages wins over the GNU equivalents
as the are installed last. We need to distinguish between the manpages
somehow...
execution is usually unnecessary in BSD Makefiles because BSD make
invokes shells with -e. Using it to give conditional execution is
often wrong in BSD makefiles because BSD make joins shell commands
when invoked in certain ways (in particular, as `make -jN'). Example
makefile:
---
clean:
cd /
false && true
rm -rf * # a dangerous command
---
This should terminate after the `false && true' command fails, but
it doesn't when the commands are joined (`false && true' is a non-
simple command, so -e doesn't cause termination). The b-maked version:
---
clean:
cd /
false; true
rm -rf * # a dangerous command
---
terminates after the `false' command fails (`false' is a simple
command, so -e causes termination). However, for versions of
make like gnu make that don't invoke shells with -e, this change
completely breaks the makefile.
This is one of the fixes for the bug suite that caused `make world'
to sometimes put raw cpp output in .depend files. Building of cc
sometimes failed, but the failure did not terminate the build
immediately, and various wrong versions of the cc components were
used until one was wrong enough to cause a fatal error.
The headers that are installed in WORLDTMP are part of the interface
that includes libraries like libc, so they must be installed together.
This means that lib-tools and build-tools should be merged. The FreeBSD
build only works in hosted form where it is assumed that the installed
version contains adequate tools to build the latest release.
`make world' by about 14% here (down to 4490 seconds real on a
K6/233). Temporarily skip this optimization when building with
-j, since there are still many broken makefiles.
Fixed NOCLEANDIR option. Cleaning of `.depend' was broken.
Put -nostdinc in CFLAGS, not in CC, and don't override the default
CC. This fixes enforcing use of ${WORLDTMP}/usr/include.
Don't install library man pages in ${WORLDTMP}.
`make world' times by about 10%. Use a new `bootstrap-libraries'
target to build just enough (static) libraries to bootstrap the
tools, and delay building of shared libraries until all the tools
have been built.
Fixed `/usr/bin/make world'. The version of `make' that we buildis
nand use is named "make", not ${MAKE}.
cleaning doesn't affect the normal copy. Save a little time by not doing
unnecessary clean, depend nd cleandepend steps.
Reviewed by: mckay@freebsd.org
Added xargs to build-tools. It may be used for kdump and truss.
(slightly stale) shared libraries get installed before non-tools
are linked. The 2-stage build of the libraries has never really
worked, since we link to the libraries built in the first stage.
so that dirtier-than-previously tools aren't left lying around to be
installed.
Filter out NOPIC from ${MK_FLAGS} for `make ${OBJDIR}' for ld (and
some other things), so that the obj dir for ld/rtld always gets
re-made.
Fixed a literal "make".
Build libraries (for linking the tools to) in a more correct order
(for linking freshly created shared libraries to each other). This
is probably a no-op now that shared libraries for tools aren't
built, but I didn't test any intermediate versions. Security-related
directories that are not built by default may now be misordered
for the shared case.
Don't build libcompat specially. It isn't used for tools, and
shouldn't be used in /usr/src (it is only used for IPXrouted and
crufty games).
Added missing ${.CURDIR} to existence tests for library directories.
Existence tests for top-level directories are still broken.
Test for library directories actually being built, not for directories
above them.
earlier. This is probably unnecessary.
Added now-necessary -B for installing headers for libss.
Removed now-unnecessary -B for building cleandepend for rpcgen.
Add the path to the source .mk files to the beginning of .MAKEFLAGS
instead of to the end, so that there is more chance of it having
priority. Additions in /etc/make.conf still have priority for
sub-makes if they are also at the beginning, although this is
probably not wanted for building /usr/src.
Use ${MK_FLAGS} for building dependencies for tools. This saves
time building dependencies for tools that won't be created (mainly
profiled libraries) and will be necessary to give correct dependencies
when tools are built static.
Spell the object directory as ${OBJDIR} consistently (even when we know
that ${OBJDIR} == "obj").
Print ${DESTDIR} in messages where appropriate.
Fixed some misformattings.
- After the install has been done, but before makedb is run, run
'ldconfig -R' to pick up new shlib dependencies.
Suggested by: Amancio and a whole lot of other folks
of doing the normal build. When the alpha bootstrap is complete (there
is more work to do!), the alpha will build like the i386 does now.
I changed i386 references to ${MACHINE} now that we're multi-architecture.
Did I ever spam this file good with that last commit. Despite 3
reviewers, we still managed to revoke the eBones fixes, TCL 8.0 support,
libvgl and a host of other new things from this file in the process of
parallelizing the Makefile. DOH! I think we need more pointy hats - this
particular incident is worthy of a small children's birthday party's worth of
pointy hats. ;-)
I certainly intend to take more care with the processing of aged diffs
in the future, even if it does mean reading through 20K's worth of them.
I might also be a bit more anal about asking for more up-to-date changes
before looking at them. ;)
way - for buildworld only.
Rev.1.136 broke the (deprecated) external includes target. Rev.1.144 was
a half-baked backout of rev.1.136. It broke SHARED=symlinks for the
buildworld target and installation of subdirs of src/include for the
includes target.
wrong switches were being passed to 'make' that caused sys.mk to come from
the /usr/share/mk location but the bsd.*.mk files to come from the
source tree in src/share/mk. This was bound to break sooner or later.
Submitted by: Dmitrij Tejblum <dima@tejblum.dnttm.rssi.ru>
everything that a full install would (modulo bugs in rpcsvc/Makefile).
The missing things weren't used (except for mp.h from libmp), but
there are relatively few of them and it's easier to be complete.
call the 'symlinks' target to build the /usr/obj/tmp/install/sys tree since
In the includes target, call the header install target explicitly and then
call the 'symlinks' target to build the /usr/obj/tmp/install/sys tree since
we set this up manually earlier, and if we do a 'make all install', we
replace the symlink tree in the obj dirs with new copies of the files
which breaks any chance of -DNOCLEAN working.
a couple *.mk files to enable -current world building on really old
machines (e.g., 2.1.5).
Reviewed by: too many many people to list here, special thanks to bde
tree) instead of the system default ones in /usr/share/mk. This
only works for makes run from the top level, and a few not too out
of date system makefiles must be present to start up. Bootstraps
from can still be done from scratch by giving the path to an up to
date share/mk on the command line.
Removed bootstrap `mk' target and use of it. The bootstrap didn't
actually work for the non-null DESTDIR case.
instead of in the middle of it. Soon, "c++rt0.o" will be linked
into all shared libraries. An up-to-date version of it must be in
place before any shared libraries are built.
This is one piece of the solution for PR gnu/3035 (gcc -shared).
This will make a number of things easier in the future, as well as (finally!)
avoiding the Id-smashing problem which has plagued developers for so long.
Boy, I'm glad we're not using sup anymore. This update would have been
insane otherwise.
MK_FLAGS contains -DNOMAN and -DNOPROFILE for speed, so I assume
NOINFO is okay as well.
Add gnu/usr.bin/texinfo to build-tools so everything does the right
thing.
directory. Also, add gnu/usr.bin/as to the lib-tools target since libgcc
generates weak symbols, and libstdc++/libg++ use them. ie: coming from a
2.1.x system will fail without 'as'.
worked for me, but no guarantees yet.
Basically, move cc earlier to lib-tools target, before libraries are
built. libgcc is built at the same time, so it's special entry is
removed from the libraries target. Also, add bison/gperf immediately
before cc, since they are needed to build cc and are not present on an
earler-than-2.2 system. (libgcc is partly built with the freshly compiled
gcc-2.7.2.1, since it uses __attribute__'s that were not present until now)
This _should_ be enough to get running from older releases, perhaps as far
back as 2.0.5.
for bootstrap" tweak to the lex Makefile to stop it building the library
too early.
This untangles things a bit more, it stops new bootstraps failing because
libl/libfl uses 'ld -O' before ld is updated.
because lex has a library (libln.a) and bsd.lib.mk has been modified to use
'tsort -q'. Tsort has not been build yet...
tsort should be before lex. [Whoops! Thanks! -jkh]
Submitted-By: roberto@keltia.freenix.fr (Ollivier Robert)
readable, as well as fix up some of the warts along the way. Among the
things fixed is the infamous "cd foo && make depend all install" bogon,
which is particularly important for users of the -DNOCLEAN option.
Also, a few typo and DESTDIR fixes and a -DNOGAMES hack from Gary Jennejohn.
cleaning during a make. This may give you more rope to hang yourself if
you are caught with some subtle dependency on installed binaries in your
build, but if you are doing daily 'make -DNOCLEAN world' it's not too bad
at all. It could take as little as 30 minutes to do an entire sync-up of
your binaries if everything's up to date, especially if you are using
'INSTALL=install -C' in /etc/make.conf (highly recomended!).
Also, add a "reinstall" target. You can do a 'make DESTDIR=/mnt reinstall'
where /mnt is the nfs root of a machine and you get the install parts of
the make world run on it.
I saw this on -hackers quite some time ago and included it in my Makefile
and have been using it on and off for a while. Alas, I cannot find the
actual message with the author's name...
