This means instead of setting MODFONT_TYPES to an extension like "otf",
font ports now specify filenames to install (wildcards allowed) like
MODFONT_FONTFILES=*.otf.
ok sthen@
This drops the mktemp template adjustment that we carried as a local
change. If people really think that six X is not enough talk with upstream.
Bulk test done by tb@
OK sthen@
The openbsd/mips64 port is effectively unmaintainable, largely due to not
being able to upstream the direct syscall to libc conversion:
https://github.com/golang/go/issues/61546
Remove mips64 from GO_ARCHS, along with armv7 (arm is already included).
Remove mips64 from the lang/go Makefile and sort the MACHINE_ARCH handling
list. Do the same in go.port.mk, adding riscv64 in the process (neither
mips64 or riscv64 were included previously).
ok jca@ sthen@
The macppc snap with the Mar 4 kernel has a fixed base-clang. The
bump affects only powerpc packages, and will cause pkg_add -u to
update to fixed packages, where the CFI might unbreak a C++ exception.
previous attempt last fall led to breakage, this is now fixed and has
survived a bulk kindly tested by tb@. It is based on how GitHub
packages its tarballs. If needed, heuristics for other sources like
gitlab, srht can be added later through the same mechanism.
This makes lines in DIST_TUPLE (github) ports like the following
superfluous:
WRKDIST = ${WRKDIR}/project-${COMMIT_HASH}
update all packages (that don't have PKG_ARCH=*). This will ensure that after
the next set of packages built, people running pkg_add -u will get binaries
built with the recent retpoline-related changes. Discussed with tb@, jca@
the project name) and using the subdir component. This is done by
stripping the directory part from the project component for the
post-extract mv(1) to subdir.
Use :T instead of :C,^.*/,, as suggested by kn@.
Problem discovered and raised by rsadowski@.
ok kn@
it results in ignoring the port unless ld.lld from ports clang is present
_when the Makefile is parsed_, so doesn't have chance to actually install
the relevant clang package.
This resulted in incomplete i386 bulks since mid November because it
only resulted in IGNORE rather than any kind of error, so didn't trigger
any reports of build failures. (Failure paths included gfortran -> graphviz
-> doxygen, gfortran -> numpy -> boost, etc).
(USE_LLD=ports is currently *only* used for gcc, and only on i386, so
other archs unaffected).
ok tb@
lang/rust module is intented to be only minimal.
makes devel/cargo module to use lang/rust module.
- makes MODCARGO_WANTLIB re-exports MODRUST_WANTLIB
- it changes the WANTLIB to the right value for sparc64, but as SYSTEM_VERSION-rust is bumped, it should be fine
adds SYSTEM_VERSION-rust ?= 0 to arch-defines.mk
- this way, when lang/rust isn't used, -V 0 is passed (it is a no-op),
and else, it is the value defined in lang/rust module.
while here, correct few ports to use the right WANTLIB value
ok tb@
"I'm happy with it" espie@ (regarding the usage of _SYSTEM_VERSION-rust)
kn