Xiyue Deng <manphiz@gmail.com> writes:
> Hi ports maintainers,
>
> Some time ago I reported that www/ruby-passenger failed to build on
> mips64el/loongson due to a linker issue[1]. Last month the port has
> been updated and I tried to give it another shot. Unfortunately it
> failed again but due to another compilation error. I have attached the
> build log and please take a look and suggest what to test next.
>
> Thanks!
>
> [1] https://marc.info/?l=openbsd-ports&m=152359916007397&w=2
I tried to dig a little more despite my lack of understand of the rake
system. It looks like rake doesn't take into account about
environmental settings from ports build system, so any setting regarding
the compiler/linker doesn't take effect at all - I verified it by
manually changing the compiler settings to add "-v" and saw that it was
using the default compiler and default linker. (By the way the base-gcc
will miscompile src/cxx_supportlib/WebSocketCommandReverseServer.h and
either ports-clang or ports-gcc is required.)
So next I manually changed all "#{cxx}" settings to
"/usr/local/bin/clang++ -fuse-ld=lld" (similarly with "#{cc}") hoping it
will use clang and LLD (including vendored libraries like libev and
libuv). And then it hit another failure message saying can't create
dynamic relocation and advise me to recompile using "-fPIC" or pass
"-Wl,-z,notext" (sorry I lost the build log about this error, but it
looks similar to [2] but with mips symbols). Adding only "-fPIC"
resulted in the same error, and adding "-Wl,-z,notext" resulted in
another fatal error included in the attached build log. I have also
attached my modification on the ruby-passenger ports hope someone can
give a suggestion on how to properly change compilers using rake.
Any advice on what to try next is appreciated.
[2] https://github.com/pytorch/pytorch/issues/21700
No comments:
Post a Comment