| Summary: | Ports using gfortran broken due to bsd.sys.mk (-fuse-ld=/...) | ||
|---|---|---|---|
| Product: | Base System | Reporter: | Stefan Eßer <se> |
| Component: | misc | Assignee: | freebsd-bugs (Nobody) <bugs> |
| Status: | New --- | ||
| Severity: | Affects Some People | CC: | arichardson, fortran, portmgr |
| Priority: | --- | ||
| Version: | Unspecified | ||
| Hardware: | Any | ||
| OS: | Any | ||
--- This is not a ports bug, AFAICT, but affects building of some ports --- Compiling ports that use gfortran is broken on systems that use clang as default compiler. The cause is the following fragment of bsd.sys.mk, which is also included when building ports: # Please keep this if in sync with kern.mk .if ${LD} != "ld" && (${CC:[1]:H} != ${LD:[1]:H} || ${LD:[1]:T} != "ld") # Add -fuse-ld=${LD} if $LD is in a different directory or not called "ld". # Note: Clang 12+ will prefer --ld-path= over -fuse-ld=. .if ${COMPILER_TYPE} == "clang" # Note: Clang does not like relative paths in -fuse-ld so we map ld.lld -> lld. LDFLAGS+= -fuse-ld=${LD:[1]:S/^ld.//1W} .else # GCC does not support an absolute path for -fuse-ld so we just print this # warning instead and let the user add the required symlinks. .warning LD (${LD}) is not the default linker for ${CC} but -fuse-ld= is not supported .endif .endif This causes a build error in lang/ratfor, for example: gfortran9 -O -Wl,-rpath=/usr/local/lib/gcc9 -Wl,-rpath=/usr/local/lib/gcc9 -L/usr/local/lib/gcc9 -B/usr/local/bin -fstack-protector-strong -fuse-ld=/usr/local/bin/ld test.f -o test gfortran9: error: unrecognized command line option '-fuse-ld=/usr/local/bin/ld' (Full build log: http://beefy18.nyi.freebsd.org/data/head-amd64-default/p549547_s365984/logs/errors/ratfor-1985.06_13.log) This particular -fuse-ld option should only be passed to compilers that actually accept it. For GCC or GFORTRAN only -fuse-ld=bfd is acceptable to use the GNU linker. I have not found any method to suppress the addition of that -fuse-ld option to the gfortran command line (and this should affect all ports that use gfortran as part of the build process) other than patching a ".f:" rule into the internal Makefile of the port. That should not be required and it caused me quite some effort to track down the cause of the build failure in this system make file. Inclusion of bsd.sys.mk into the ports system provides useful defaults for a number of parameters, but it should be robust if used in constellations that do not occur in the base system (i.e. use of gfortran in this particular case).