From now on, building ports requires X11BASE.
* lots of ports missed this variable (porters always have X11 installed)
jasper made a first pass on fixing these, but no doubt it will drift
away soon
* users should use packages
* people building ports should know what they are doing and having
X11BASE as a requirement is perfectly reasonnable
* we are merging config.x11.site into config.site because of a stupid
autofoo bug ; currently when USE_X11 is set, some ports may loose the
ability to even load config.site
tested on a bulk by landry@
ok robert@ jasper@ landry@ sthen@ naddy@ "I won't cry if it dies" espie@
by default for ports that set USE_LIBTOOL=Yes. Faster, better,
stronger, and cuts bulk builds time by ~15% in my tests on amd64.
Volunteers welcomed to fix the few USE_LIBTOOL=gnu users in the tree.
Report any oddities to ports@...
'Go ahead' espie@ 'Please proceed' steven@
ok ajacoutot@ jasper@ phessler@ sthen@
helps getting past shell quoting errors, so that the infrastructure
will report actual trouble (is also simpler than :S/<...>/ for pkgspecs).
REQUIRES CURRENT MAKE TO WORK!!!
PACKAGE_COOKIE, so that print-plist-contents keeps working.
Only the packaging part (e.g., PACKAGE_COOKIES) should be affected.
Alter PKGNAMES while we're there, as it's only used for debugging, and
it's better that way.
(fixes pkgmklocatedb)
More precisely:
- for "older" libspec in LIB_DEPENDS, only compare them against
the libs from the LIB_DEPENDS.
- check LIB_DEPENDS against all libspecs for the package (both from WANTLIB
and LIB_DEPENDS) to turn them into LIB_DEPENDS.
- create a full list of installed libs and of 'inter-package' libraries to
solve all WANTLIB in one go.
This is actually faster than the old code, but a bit more complicated.
In particular, create a tmp package until we're certain it is okay
(thanks to register-plist). -> remove race condition in dpb3, which
assumes a package existing is enough for it to proceed.
- simplify IGNORE handling, always define it, even when NO_IGNORE is set,
and only use it for the targets it's meant to influence.
- make it possible to multiply-IGNORE a port, so stuff that's BROKEN and
NOT_FOR_ARCH will show up as both.
- DESCRIBE_TARGETs can die, since we no longer need special IGNORE dance
for them.
- add an IGNORE_IS_FATAL tweak, so that ignored ports can actually error out.
- reorg dump-vars to choose what to show, including new IGNORE, and COMES_WITH.
- name an explicit prepare step, which does check build dependencies and
install them (useful for build timing purposes)
- move the do-fetch test out, so that we can add pre-extract and do-extract
there eventually.
but they can (sometimes) be useful: add DANGEROUS knob to prevent running
them accidentally, but allow people to run makesum/update-plist/update-patches
recursively.
- there's no need for :: on pure dependencies, : works just fine for
accumulating (and :: is not 100% standard on all systems)
- append directly to _FETCH_MAKEFILE_NAMES. This avoids an untidy blank space
happen often (/usr/ports NFS mounted & /usr/ports/obj a symlink to a
local dir, /usr/ports as a symlink to /home/wherever..)
Use readlink -fn only if USE_SYSTRACE is set so less things are run when
bsd.port.mk is included. Should make everyone happy.
Issue reported by and original fix ajacoutot@, sounds reasonable to espie@
dirs forgotten all around. One can still unset it in /etc/mk.conf to get
back to the older behaviour, or still override it with /usr/obj/ports or
/usr/wobj or whatever. At least, now it can be a separate partition.
Most of us were already overriding this value, OpenBSD is all about sane
& simple default settings.
${PORTSDIR}/obj suggestion from jakemsr@.
Discussed at length with several, 'Works for me' krw@ ok todd@ wcmaier@.
but select some variables instead.
checked by naddy@ on a bulk-build.
(apart from making stuff less easy to break, it also means Makefiles stop
inheriting a WIDE set of .PHONY targets, which is a good thing)
redundant makefile parts found at many places in the tree
DISPLAY and XAUTHORITY variables/checks to be nuked soon from many ports
ok ajacoutot@, sthen@, jasper@
this fixes an old bug where package creation would fail if one has some
particular file under PKGPATH because it would give pkg_create the wrong
args
from Wim Lewis on ports@, thank you!
"looks right" sthen@, "sounds reasonable" landry@, ok jasper@ espie@
*solely* as a facility for people working on the ports tree.
not to be used for anything at all in the ports tree itself !!!
since it's not protected against any kind of race.
last one in the pipe to make sure failed regression tests don't bake a
cookie when REGRESS_LOG is set (default).
Also redirect stderr to regress log and introduce REGRESS_STATUS_IGNORE
variable so post-regress targets like found in cpan.port.mk can do the
actual failure detection.
Problem noticed by bluhm@, latching of exit code suggested by steven@
Testing, feedback and ok on earlier version by bluhm@, 'go for it' espie@
well.
Incidentally, this is also an issue with update: in some cases, the update
target wouldn't work (specifically, when dependencies had changed).
Fix it as well.
There might be a need to revisit the DEPENDS_TARGET stuff later, this is
a bit of a kludge...
every word. Also use :T to select the final word of the PKGPATH, so
that hooks work seamlessly for `non-core' modules, thus removing the
need for most modules to be core.
SIMPLE_LOCK for a while, this means something else is fetching
the same distfile, so first check whether our target actually exists
before we try to fetch it all over again.
the variable subst means the subst shouldn't occur anywhere, but only at
beginning of name... useful for backsubst of SYSCONFDIR, since etc is so
frequent. Extended to other variables such as PREFIX.
tweak make-plist to use OpenBSD::Subst.
*this means this only works with current, but you've had a week to update...*
- pass an argument for the file into which we want to save the result, to
avoid getting it polluted with error messages (defaults to /dev/stdout for
debug).
- create the makefile fragment as a temp file and only copy it when complete.
- copy it in one chunk, so that one can read a partial mirror-maker file
and have it be usable.
This does allow for people to start a make mirror-maker in one shell, and
start fetching stuff right away, before mirror-maker is finished.
This also produces usable mirror-maker Makefiles even if the ports tree
contains bogus entries.
dependency, but only the ones actually required.
If we end up needing to rebuild them from source, then change back
to _internal-package-only, so that we benefit from the whole build
and get all packages again.
we can compute it first, and only start pkg_create if it didn't error
out.
This gets rid of the very verbose and confusing error messages
pkg_create meets when the ports tree cannot solve some dependencies.
Do port-lib-depends-check in two steps: first generate a list of
binary: lib1,lib2,lib3
in WRKINST/.saved_libs
then run check-lib-depends on it.
Speeds up tweaks to WANTLIB quite a lot, as we do not rescan every
binary all the time...
