textproc/gtk-doc is in BUILD_DEPENDS and force ac_cv_path_GTKDOC_* to null.
Even when --disable-gtk-doc is passed, if the gtk-doc tools are present at
configure time, they will be run (at a time where the gtk-doc package may have
been removed; for e.g. during a bulk).
This is what we've been doing under x11/gnome for a while now.
sthen@ espie@ and jasper@ agree
to ensure that a sane file mode are set.
Also use SUBST_CMD -m for rc.d scripts (it wasn't available at the time the
current code using a separate chmod was written).
Discussed with espie, slight tweaks to previous diff to take _FAKESUDO into
account.
bulk build, and I know at least two people (ajacoutot@ and me) that were
already bitten by this problem.
This somewhat increase check time, unfortunately, but not too much.
Test for this feature to follow after full unlock.
naddy.
Not turned on yet, as it involves a critical new file.
Basically:
- introduce new variations on SUBST cmds to install various files
- use pkg_subst -i in !root mode
- unhardcode install path where possible, to allow for an install-wrapper
that disregards owner/groups.
- do links during fake to disable chown/chgrp/install.
Extra tweak (always-wrap) to actually encode "perl ${PORTSDIR}/infra/bin/install-wrapper" as an install script, required by a few ports that keep track of
the install script for later (eg., ruby, postgres).
to count LIB_DEPENDS and WANTLIB in build-deps, only the actual subpackaged
version count (reminder: even single package ports are actually multi-packaged,
with SUBPACKAGE=-, hence we will count LIB_DEPENDS- and WANTLIB-).
This was actually a discrepancy between manual builds and dpb builds, as the
output of dump-vars won't show plain LIB_DEPENDS. This caused a bit of
confusion wrt multimedia/mlt. Hence the actual fix.
we're not using g?make, like CMake+Ninja. Also helps to simplify some
do-test constructions, including upcoming multimedia/mlt.
Documentation bits to follow.
okay espie@
