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...
without a packagename bump, or because they have too long comments.
steven@ and I cleaned up most of them.
As a result, there's no longer any WRKPKG directory with their temporary
files.
We also use the `sort -u' feature of pkg_create for dependencies, so that
the command lines to pkg_create get simpler, which will allow for easier
error-checking later on.
SUPDISTFILES, so we get them as well.
This misbehavior noticed by Mikolaj Kucharski.
(the intention is obviously to regrab everything to verify whether
anything changed, and that includes SUPDISTFILES)
Only functional change:
Add sha256 at the end of the list of supported ciphers, so that new
makesums will add the new cipher.
The rest is business as usual.
the multi-packages case, as should be.
introduce _ALLPKGPATHS, and use it in show-required-by, so that
show-required-by works for multi-packages.
problem noticed by bernd@
Adapt make-plist to use the same code as pkg_create to deduce fragment
names, remove some old special cases that should no longer matter.
Zap directories from dependencies in a way depending on the dependency.
Far from perfect yet, as directories should be registered multiple times
in packages that do not depend on each other, and could also be used to
figure out where to put new files...
of FORCE_UPDATE. Also break out in case update fails (otherwise the old
installed package would be enough to continue building).
Sprinkle REPORT_PROBLEM to know more precisely what stuff is breaking.
replace some
if eval cmd; then
with
if (eval exec cmd); then
because if eval... seems to break early, even with the test guard, thus
giving a chance to REPORT_PROBLEM to do stuff instead of having to wait.
(Note: a trap would probably be simpler ?)
bsd.port.subdir.mk
Use absolute paths to build readmes files, based on READMES_TOP, which
can be (or not) PORTSDIR.
Generate category readmes in tmp directory as well.
minor clean-ups
so that the `normal' cases is MULTI_PACKAGES, (with possibly one
special '-' subpackage).
Adjust a few tests accordingly so that people don't notice the '-'
SUBPACKAGE: mostly do not print some directory changes, adjust FULLPKGPATH,
and set up PLIST- templates correctly.
careful: they do not get defined to a default value unless the `main'
variable is defined, because they get set later otherwise...
remove the .if defined(MULTI_PACKAGES) guards as they don't serve any purpose.
Simplify the PLIST tests to make them more uniform.
Move SED_PLIST to the list of obsolete variables instead of giving it special
treatment.
and WANTLIB.
Get the rules for the LIBSPECS_COOKIES out of the loop that dictates their
usage.
Introduce *DEP3 variables, to keep *-depends-list compatible with old stuff.
Use _BUILDLIB_DEPENDS and _BUILDWANTLIB everywhere this makes sense.
allows us to get rid of some NO_DEPENDS tests later, also allows for
RUN_DEPENDS to become indexed on SUBPACKAGE.
LIB_DEPENDS and WANTLIB are going to need more surgery, since they need
to be checked during build and packaging...
consider it to be the main package, and do some equal treatment for each
package.
(to do: add subpackage where needed to WANTLIB, RUN_DEPENDS, LIB_DEPENDS
and act on them accordingly).
Also define _DONE_FILES for _fetch-makefile to avoid blocking if it's
not invoked from a higher level target.
can be used to run targets that may fail without impacting further stuff,
or to collect stats on anything.
For instance
BULK_DO= mkdir -p ${PORTSDIR}/config/${FULLPKGPATH}; \
cp -f ${WRKBUILD}/config.{log,status} ${PORTSDIR}/config/${FULLPKGPATH} || true; \
cd ${PACKAGE_REPOSITORY}/${MACHINE_ARCH}/all && perl ${PORTSDIR}/infrastructure/package/check-modes ${PKGNAMES} || true
- make sure locks happen when they should, including for lib-depends-check.
- make sure various targets set PACKAGING as they should.
- let BULK=Yes be invoked only from make package.
Fix for the BULK_COOKIE and UPDATE_COOKIE issue. They can no longer depend
on PACKAGE_COOKIES, since this needs to get built with PACKAGING set. So
instead, simply remove the cookies when we create a new package succesfully.
By any kind of reasoning, both bulk and update are `reset' when a new
package appears, so this seems to be the right semantics in most cases.
Thanks for wilfried and nikolay for error spotting. Hoping this will be
what's needed, finally.
building a subpackage, and then we definitely want the `current' stuff,
not the stuff that's already installed.
Allows shared libs updates in MULTI_PACKAGES to proceed gracefully without
having to uninstall stuff.
(we do not do this stuff in general, because in other cases, builds are
`staggered', e.g., a port is built against the existing base, not the stuff
in other WRKDIRS)
Add a message to that effect so that users don't get confused.
(specifically, make itself does not change dirs, so if you do ^Z,
you're not where you think you are...)
end up there with PACKAGING not set.
We do not try to `fix it' ourselves, because PACKAGING not set may have
some non obvious issues, like PKG_ARCH or PERMIT_PACKAGE_* not set correctly
at all.
Move BULK_COOKIE for package to the redirector, so that we do not
clean in subpackage cases, but instead wait for all the packages to be built.
Also, introduce FORCE_UPDATE=hard: with signatures, -F installed is
most often not necessary...
package cookie, so that an update will get all subpackages built.
And a fix for an old issue: let make update proceed to update ALL subpackages,
as seems the most natural. Create a subupdate target if one specifically
wants to update a single package (and use it as dependency in the FORCE_UPDATE
case).
cookies, and also insist on running make package, which is necessary to
make sure all MULTI_PACKAGES get built.
Repair make clean=packages: make sure all packages get named using the same
scheme as other multi-packages targets (that way, arch-indep subpackages
get cleaned properly)
into real targets.
- now make package will create missing ftp/cdrom links when PERMIT says so.
- PKG_ARCH=* packages get built into packages/no-arch, and linked from
elsewhere.
Everything gets based off PACKAGE_REPOSITORY, PKGREPOSITORYBASE and friends
get ditched.
All `package' targets go through the same routine: iterate through every
subpackage with PACKAGING set.
This includes describe, lib-depends-check, package, dump-vars, install-all,
which will iterate through subdescribe, sublib-depends-check, subpackage,
subdump-vars, and install.
(names are subject to change).
Much simpler logic, plus hey, you get an install-all target !
Might be some minor breakage, most stuff appear to work just fine...
_MASTER_LOCK, we keep a list of _LOCKS_HELD by one port builder in
an env variable, and we don't relock stuff that's already locked.
This allows us to, e.g., have fake depend on regress without issue
(even though, internally, it's more efficient to use the _internal-*
targets to avoid testing locks).
and leaves you at the ftp prompt in the directory where stuff is fetched
from.
(I find myself using this quite often to figure out which version of stuff
exists, especially for CPAN).
SHARED_LIBS enumerates the libraries like so:
SHARED_LIBS= foo 1.0 \
zop 2.5 \
tag 3.0
This just defines LIBfoo_VERSION=1.0, LIBzop_VERSION=2.5, LIBtag_VERSION=3.0
and adds these to SUBST_VAR, further tweaks must be done, usually at the
MAKE_FLAGS/MAKE_ENV level.
For gnu ports that use automake, the supplementary MODGNU_SHARED_LIBS
variable can be used like this:
MODGNU_SHARED_LIBS= foo '-no-undefined'
and it will define libfoo_la_LD_FLAGS=--version-info 1:0:0 -no-undefined
Some gnu ports have weird variable names for libraries. For instance,
the libORBit-2 is set using libORBit_2.
LIBORbit-2_ALIAS=ORBit_2
will make sure the correct name is used.
much feedback from bernd@
the main directory... since pseudo-flavors are not entered as pkgpath,
this is necessary to ensure the top-level directory gets cleaned up
correctly...
Problem reported by Moritz Grimm.
it makes more sense to have a subdir variable as input, split it
into toset and dir as output, and to unsetenv FLAVOR SUBPACKAGE along
the lines.
End result should be equivalent, but slightly more readable.
no objection from my neighbors either...
for errors (and get rid of _NODEPS since it's not needed any more)
* set check to Failed in *libresolve_fragment, so that we can test
for it and report properly.
guys here have no objection...
over 256 characters, so we create an extra file for each wantlib, and
depend on it. That way the full _DEPlibs_COOKIE is regenerated (and retested)
each time WANTLIB changes.
Now, resolve-lib can take a big list of libraries with full paths,
and it can solve a big list of spec at once.
Basically, we move most of the parsing of spec paths into resolve-lib.
Since print-package-signature does build a full list of libs, let's solve
it all at once, instead of invoking a costly perl script repeatedly.
Add some caching possibilities for out-of-date. Specifically:
- store libraries for each package under the directory _PORT_LIBS_CACHE
- use the dependency cache _DEPENDS_FILE to avoid recreating dependency
chains, add a new file _DEPENDS_CACHE that will accumulate all dependencies,
and extract these with a simple script extract-dependencies.
Use echo to build libraries lists instead of ls, that's a bit simpler...
Some more clean-up will happen: it's probably simpler to parse libspecs
at once, extract the libraries needed and go fetch the corresponding libraries
just once.
