which does much of the same work but prints a result instead of
setting a global variable. Adjust the rebuildmake function to suit.
Improve the comments describing how we try to figure out whether
the existing tooldir make needs to be rebuilt.
Use getmakevar to expand MAKEOBJDIRPREFIX or MAKEOBJDIR, or (if
neither of those variables is set) use a non-recursive "make obj" in
the top level source directory.
* Do not let TOP_objdir default to ${TOP}. It's now set only by the
-M or -O command line options, or after using getmakevar to expand
MAKEOBJDIRPREFIX or MAKEOBJDIR.
* Make try_set_TOOLDIR handle MAKEOBJDIRPERFIX passed from the
environment. It runs too early to use getmakevar, so it will not work
if the value contains embedded '${...}' variable references.
* Use '$TOOLDIR' (with a literal '$') instead of 'nonexistent' in
a message when rebuilding make when TOOLDIR is not yet known.
is set to "yes" -- defaults to "no" except for build.sh builds. This
results in a deterministic .a file rather than one that reflects
timestamps and permissions on the source files.
Also, clean up the ar flags we're using, and remove a redundant use of
ranlib that on a modern POSIX ar can be done with the "s" flag.
Discussed on tech-toolchain
prints something that does not look like an identifier, then use uname
-m instead. (Cygwin prints "unknown", and OpenBSD prints a long string
containing several spaces; this code should handle both.)
0 Minimal output ("quiet")
1 Describe what is occurring
2 Describe what is occurring and echo the actual command
3 Ignore the effect of the "@" prefix in make commands
4 Trace shell commands using the shell's -x flag
The default remains MAKEVERBOSE=2.
and store in the (non-exported) $TOP_objdir.
(We can't do the same for the environment $MAKEOBJDIR since it may contain
make(1) expressions that aren't easily parseable by sh(1))
Remember appropriate variations of '-M modp' and '-O mod' in $TOP_objdir
(replacing $makeobjdir misuse).
Use $TOP_objdir when trying to guess the TOOLDIR,
and when detecting if the top-level objdir needs to be created.
Fixes problems observed when testing fixes for PR 39360 and 39361.
Fix the TOOLDIR path used when guessing the TOOLDIR.
PR 39360 from Andrew Cagney <cagney@gnu.org>
Use stricter result checking from make(1) when guessing the TOOLDIR.
PR 39361 from Andrew Cagney <cagney@gnu.org>
Apparently there are things you can do in the environment or mk.conf
to set your object directory in a way that the previous code couldn't
detect. (This is an interim measure until we can properly detect and
create the top level object directory in all cases.)
or bomb in non-expert mode.
* If a previous build.sh run with the -U (unprivileged) flag created a
METALOG, then subsequent build.sh runs must also specify the -U flag.
In expert mode, this is just a warning.
* While I was here, changed spaces to tabs in one existing line.
to the CD-ROM image.
Make CDEXTRA and CDBUILDEXTRA be able to contain lists of files and directories
which are added in turn.
Add support for CDEXTRA_SKIP which contains a list of fnmatch(3) items to
skip during the copy. This is _very_ useful for skipping CVS dirs for example.
CDBUILDEXTRA and CDEXTRA_SKIP are to be used in Makefiles creating the images,
CDEXTRA remains to be used from build.sh.
Update documentation and build.sh help text to show that -C can be used to
specify a list of files and directories rather than just a single dir.
includes source sets as well. The infrastructure for this was already
there, although it needed a bug fix. Will look at adding arbitrary
directories next.
* Add a try_set_TOOLDIR function which tries to be clever about setting
TOOLDIR if possible;
* Call try_set_TOOLDIR from rebuildmake;
* Add nobomb_getmakevar, which works like raw_getmakevar except
never bombs;
* Reimplement raw_getmakevar in terms of nobomb_getmakevar;
* Add a second arg to find_in_PATH to control the result when the
program cannot be found.
MACHINE_ARCH only if it's not already set. This prevents
build.sh -m evbarm-eb
from ending up with MACHINE_ARCH=arm instead of =armeb (since MACHINE is
stripped of the -eb suffix just above).
flag. This ensures that -j<N> isn't accidentally inherited from the
environment, because the "--- foo ---" headers printed with -j<N> would
interfere with parsing the output.
when I make my wrapper script source build.sh. I can also think of other
times when it would be a PITA.
Abort the build if make fails in raw_getmakevar() - as tends to happen when
make itself is badly broken.
where ${MAKEWRAPPERMACHINE} is the suffix XXX on the target's make
wrapper, nbmake-XXX.
Fixes toolchain/30673, "single arch host disklabel isn't enough".
* Try to guess a suitable value for HOST_SH, if it was not set in
the environment. First try host-specific heuristics (Solaris has
/usr/xpg4/bin/sh); then try to find the name of the shell that was used
to run build.sh itself (by parsing the output from ps -p $$ -o comm);
then fall back to "sh".
* Having found a value for HOST_SH, copy it to BSHELL and CONFIG_SHELL.
* Use ${HOST_SH} instead of /bin/sh when creating the make wrapper.
* Use ${HOST_SH} instead of unqualified sh when running shell scripts.
- allow "-m macppc/evbppc" to also take "-a powerpc64" to build a
powerpc64 toolchain.
XXX this may change, as an evbppc64 port may be required.
XXX deal with that when we have some code...
which invokes make iso-image in the etc directory.
* Add an "iso-image" action in build.sh, which
invokes make iso-image.
* Document the above in doc/BUILDING.mdoc.
* Re-generate BUILDING.
Approved by christos
tools so that a parallel make doesn't start running dependall before
cleandir has finished.
Unfortunately make doesn't support .WAIT on the command line :-(
The Iyonix is a desktop machine from Castle Technology, based on a 600MHz
XScale[tm] 80321 processor.
* Uses the bootloader from NetBSD/acorn32, which is now 32-bit compatible.
* Currently boots multiuser with a serial console.
* Device support is not yet complete.
With help from abs.
test with "${MKxxx}" = "no" instead of -z "${MKxxx}"
Ignore errors when running pwd -P. (GNU coreutils incorrectly complains)
Both problems pointed out in private email from Christian Limpach.
rather than adding '-s' to make's command line in the makewrapper,
use the .SILENT target based on MAKEVERBOSE's value.
This means that you can do:
./build.sh -N 1 makewrapper
$TOOLDIR/bin/nbmake-$MACHINE
# runs at MAKEVERBOSE==1, with command lines suppressed
$TOOLDIR/bin/nbmake-$MACHINE MAKEVERBOSE=2
# runs at MAKEVERBOSE==2 for this invocation, with command lines shown
trying to use them internally to build.sh.
Fixes issue where build.sh would run "make cleandir" if MKUPDATE=yes was
set in mk.conf and -u wasn't given to build.sh.
to be at any time that build.sh is run without -o, because we need
the objdir of the top-level to exist if build.sh is going to provide
a default DESTDIR/RELEASEDIR, and with the previous code, if
MAKEOBJDIRPREFIX was used and the objdir of the source tree didn't
exist, build.sh would incorrectly determine the DESTDIR.
(mmm, obscure bugs).
* add "releasekernel=conf", to install a gzipped copy of the kernels
built by "kernel=conf" to RELEASEDIR/MACHINE/binary/kernel
build.sh:
* be more consistent in various status messages
* move the {release,}kernel=conf conf parsing code into getkernelconf()
and use in both buildkernel() and releasekernel()
* only warn once when building multiple kernels without updating the tools
the .OBJDIR of the top of the source tree, but expressing that in 1
line or less starts to get complicated). Requested by Hubert.
* When using a -el or -eb MACHINE "shortcut", use that original value
as the MACHINE for the default makewrapper filename. Requested by Simon.
which sets the appropriate MACHINE_ARCH and rewrites MACHINE to lose
the -e[bl] suffix.
* Don't default to a MACHINE_ARCH for evbmips or sbmips.
Per discussion wth Simon Burge.
* If not expert mode, provide defaults for:
DESTDIR /top/of/obj/destdir.${MACHINE}
RELEASEDIR /top/of/obj/releasedir
* Collate the various status messages output during the run
and display them at the end in a "summary report".
* Cross-check the validity of MACHINE against MACHINE_ARCH.
Fixes PR [toolchain/20193] from David Maxwell.
* Highlight that the tools will be rebuilt if UPDATE isn't set.
* Add stronger language recommending against -E unless you
* Improve whitespace use in usage()
Code stuff:
* Be consistent about using "${var}"
* Be more consistent how [ tests ] are run
* Improve some comments
* Rename getmakevar() to raw_getmakevar()
* Rename safe_getmakevar() to getmakevar()
* cd back to ${TOP} after a kernel build.
* Always keep the tmpdir around for the build (it's where the
status messages are collated, for one).
Add "makewrapper" operation to build.sh.
[The above changes were suggested by Alan Barrett on current-users@]
Fail if building as non-root unless: UNPRIVED is set, -U is given, or -E
(expert mode) is given.
[Suggested by James Wetterau in private email]
Improve the examples in BUILDING.
parameters, and a list of "operations" defines what to do.
The full usage is show below. Notes:
`-b' has been deprecated (it always occurs now)
`-d' is replaced by "distribution"
`-R rel' onlys sets RELEASEDIR; use "release" to build a release
`-k kern' has been replaced by "kernel=kern"
`-i idir' has been replaced by "install=idir"
-r now occurs before nbmake is rebuilt
Add a copyright (long due!). Rework the code to (a KNF inspired) ShellNF.
Use functions appropriately. Allow `nb' prefix to be easily changed to
something else. Solve world peace (just kidding)
This is part of the (never achievable) goal of attempting to make
NetBSD easier to build...
--8<-- new usage follows --8<--
Usage: build.sh [-EnorUu] [-a arch] [-B buildid] [-D dest] [-j njob] [-M obj]
[-m mach] [-O obj] [-R release] [-T tools] [-V var=[value]]
[-w wrapper] [operation [...] ]
System build operations (all imply "obj" and "tools"):
build Run "make build"
distribution Run "make distribution" (includes etc/ files)
release Run "make release" (includes kernels & distrib media)
Other operations:
obj Run "make obj" (default unless -o)
tools Build and install tools
kernel=conf Build kernel with config file `conf'
install=idir Run "make installworld" to `idir'
(useful after 'distribution' or 'release')
Options:
-a arch Set MACHINE_ARCH to arch (otherwise deduced from MACHINE)
-B buildId Set BUILDID to buildId
-D dest Set DESTDIR to dest
-E Set "expert" mode; disables some DESTDIR checks
-j njob Run up to njob jobs in parallel; see make(1)
-M obj Set obj root directory to obj (sets MAKEOBJDIRPREFIX)
-m mach Set MACHINE to mach (not required if NetBSD native)
-n Show commands that would be executed, but do not execute them
-O obj Set obj root directory to obj (sets a MAKEOBJDIR pattern)
-o Set MKOBJDIRS=no (do not create objdirs at start of build)
-R release Set RELEASEDIR to release
-r Remove contents of TOOLDIR and DESTDIR before building
-T tools Set TOOLDIR to tools. If unset, and TOOLDIR is not set in
the environment, nbmake will be (re)built unconditionally.
-U Set UNPRIVED
-u Set UPDATE
-V v=[val] Set variable `v' to `val'
-w wrapper Create nbmake script as wrapper
(default: ${TOOLDIR}/bin/nbmake-${MACHINE})
(Thanks Alan!)
* Before attempting to use KERNOBJDIR, we now need to "make obj" in
${KERNSRCDIR}/${KERNARCHDIR}/compile, not in ${TOP}/etc as used to be
the case.
* Fix one place where getmakevar was invoked unconditionally. It needs
to be conditional on $runcmd != "echo", so that we can rely on the
existence of the new $make executable.
* Add a sanity check to getmakevar, to bomb if a similar error is
introduced in the future.
* Changed the bomb function to print to stderr instead of to stdout, and
to kill the top level shell process. Without this, an attempted bomb
from inside getmakevar did not work properly.
* Moved some duplicated code into a new safe_getmakevar function, which
calls getmakevar if that is safe, or else emits a literal '$' followed
by the variable name.
Improvements from me:
* Always 'trap "exit 1" 1 2 3 15', so the kill in bomb() doesn't cause ugly
output.
make INSTALLWORLDDIR=installworlddir installworld
after every other operation has completed.
If you want to run this target without rebuilding, use -b as well...
- Cache the output of `uname -s` and `uname -m` for future use.
* When building a kernel with the -k option, make obj in $TOP/etc to
ensure that the correct KERNOBJDIR is created and used. Previously,
KERNOBJDIR could sometimes be an objdir outside the source tree
and sometimes $TOP/sys/arch/$ARCH/compile inside the source tree,
depending on whether or not a complete build had beed done before
the kernel build.
* Renamed KERNCONFNAME to kernconfname, for consistency with what
seems to be a convention that variables local to build.sh (as
opposed to make or environment variables) have lowercase names.
* Renamed the do_buildonlytools and do_buildonlykernel variables to
do_buildtools and do_buildkernel, because the word "only" is no
longer appropriate now that more than one of them can be active
simultaneously.
* Changed a few '|| exit 1' tests to '|| bomb "message"', and added a
few more similar tests.
* Changed all 'echo "===> ..."' to use $runcmd, and added a few more
similar echos. Previously, some used $runcmd and some did not.
kernel using the $TOOLDIR toolchain. If kern doesn't contain `/', it will
be searched for in ${KERNCONFDIR} (typically, sys/arch/${MACHINE}/conf).
- Improve ./build.sh usage.
- Add some examples to the documentation.
This is heavily based on [toolchain/18739] from Alan Barrett <apb@cequrux.com>,
with improvements by me.
apb