devel/iscd-commons: Count CPUs with sysconf instead of libpmc
The port's lplib3.c patch counts CPUs with pmc_init()/pmc_ncpu() but
libCommons.so is not linked with libpmc, so loading it fails with
"Undefined symbol pmc_init" (e.g. cad/py-gmsh). pmc_init() also
needs hwpmc loaded. Use sysconf(_SC_NPROCESSORS_ONLN).
While here, fix portlint warnings and document the patch.
PR: 300280
devel/subversion: remove the BDB option
Berkeley DB support has been deprecated upstream for over a decade and
is slated for removal in a future Subversion release. Pass
--without-berkeley-db --disable-bdb6 explicitly instead of gating it
behind an opt-in option: configure otherwise autodetects and links
against whatever BDB happens to be present on the build host, with no
corresponding LIB_DEPENDS/pkg-plist entries to track it.
Drop the pre-configure apu-config BDB-version probe (moot once BDB is
never enabled), the %%BDB%% pkg-plist entries for libsvn_fs_base and
its pkgconfig file, and the DATADIR/bdb/*.py BDB-only scripts (excluded
at extraction time via EXTRACT_AFTER_ARGS, same as the tools/ tree's
other dead-weight subtrees).
Remove files/pkg-message.in: unreferenced by any SUB_FILES/PKGMESSAGE,
and its content described the now-moot BDB deprecation warning.
devel/subversion-lts is intentionally untouched and keeps BDB support
as-is.
math/cminpack: fix plist on powerpc64le 16.0
Since 16.0 long double is IEEE binary128 on powerpc64le, so
libcminpackld gets built and has to be in the plist.
graphics/photocraft: update to 0.18.0
Make HEIF an option (still on, like upstream's own builds) and add an
AVIF option for export through rav1e, off like upstream since the app
can't read AVIF back.
Changelog: https://github.com/storytold/photocraft/releases/tag/v0.18.0
games/hedgewars: enable on powerpc64le
Free Pascal enables the invalid-operation FPU trap on PowerPC and Lua's
parser trips it on every constant outside the int range, so mask the
FPU exceptions in the engine like a C program has them.
java/openjdk*: fix kqueue selectors on big-endian
KQueue.getDescriptor() reads the kevent ident, a uintptr_t, as a 32-bit
int. On big-endian that is the zero high half, so every ready event is
dropped and Selector.select() returns at once with nothing selected.
Netty servers like OpenSearch then spin forever rebuilding the selector.
Read the field at its full width instead.
