Having a quick look at the asterisk build 22.log...

http://build-failures.rhaalovely.net/riscv64/2026-09-27/telephony/asterisk/22.log


Do you know why the pidf_lo_test.o, pidf_to_eprofile.o, eprofile_to_pidf.so
files are are the only ones being linked with the -melf64lriscv parameter?

(They are the ones ending in ld: error.


Looks as if there is some confusion with the RISC-V targets. Sometimes
the makefile thinks it is doing a crosscompile, sometimes it knows it is
a native compile.


Thomas



On 10/7/26 18:37, Jeremie Courreges-Anglas wrote:
> On Wed, Oct 07, 2026 at 10:48:19AM +0100, Stuart Henderson wrote:
>> On 2026/10/07 02:35, jca@wxcvbn.org wrote:
>>> http://build-failures.rhaalovely.net/riscv64/2026-09-27/databases/postgresql-pllua.log
>>> http://build-failures.rhaalovely.net/riscv64/2026-09-27/telephony/asterisk/20.log
>>> http://build-failures.rhaalovely.net/riscv64/2026-09-27/telephony/asterisk/22.log
>> ld: error: src/compat.o: file has conflicting properties: GNU_PROPERTY_RISCV_FEATURE_1_CFI_LP_UNLABELED and GNU_PROPERTY_RISCV_FEATURE_1_CFI_LP_FUNC_SIG
>> ld: warning: src/compat.o: cannot link object files with different floating-point ABI from /usr/lib/crtbeginS.o
>>
>> Is LLD_EMUL set correctly for the arch?
>>
>> /usr/ports/infrastructure/mk/arch-defines.mk                      77:_LLD_EMUL_riscv64 = elf64lriscv
> AFAIK yes.  There are two issues here:
>
> 1. an error that I have downgraded to a warning because the code is
> too pedantic for its own good. Unfinished diff that went nowhere:
> https://reviews.llvm.org/D106378
>
> 2. compat.o has been generated with ld(1) and objcopy(1):
>
> ld -melf64lriscv -r -b binary -o src/compat.o src/compat.luac
> objcopy --rename-section .data=.rodata,contents,alloc,load,readonly src/compat.o
> cc -Wall -Wmissing-prototypes -Wpointer-arith -Wdeclaration-after-statement -Werror=vla -Werror=unguarded-availability-new -Wendif-labels -Wmissing-format-attribute -Wcast-function-type -Wformat-security -Wmissing-variable-declarations -fno-strict-aliasing -fwrapv -fexcess-precision=standard -Wno-unused-command-line-argument -Wno-compound-token-split-by-macro -Wno-format-truncation -Wno-cast-function-type-strict -O2 -pipe -fPIC -DPIC -fvisibility=hidden -shared -o pllua.so src/compile.o src/datum.o src/elog.o src/error.o src/exec.o src/globals.o src/init.o src/jsonb.o src/numeric.o src/objects.o src/paths.o src/pllua.o src/preload.o src/spi.o src/time.o src/trigger.o src/trusted.o -L/usr/local/lib  -L/usr/local/lib  -L/usr/local/lib -L/usr/local/lib -Wl,-Bdynamic -fvisibility=hidden src/compat.o -L/usr/local/lib -llua5.3  -lc
> ld: error: src/compat.o: file has conflicting properties: GNU_PROPERTY_RISCV_FEATURE_1_CFI_LP_UNLABELED and GNU_PROPERTY_RISCV_FEATURE_1_CFI_LP_FUNC_SIG
> ld: warning: src/compat.o: cannot link object files with different floating-point ABI from /usr/lib/crtbeginS.o
> cc: error: linker command failed with exit code 1 (use -v to see invocation)
>
> ld(1) itself is at fault, llvm-readelf -Wa on such an object file gives:
>
>    Properties:    RISC-V feature: ZICFILP-unlabeled, ZICFISS, ZICFILP-func-sig, <unknown flags: 0xfffffff8>
>
> which looks like junk.  I have noticed this since months already but I
> haven't bothered yet to go and find the root cause.
>