gcc-11.1.0 - template-id not allowed for destructor about mold

[rui314]

Thank you for reporting! I don't have gcc-11, but I believe the above change fixes the issue.

[lancethepants]

Thanks! Well since the issue is still open, I think these are all the remaining issues with GCC11. I did make -j50 and this is what remains. I do really have interest in compilers and linkers, but limited c/c++ knowledge. Looks like more semantic trivial gcc-isms from the looks of it.

g++  -g -IoneTBB/include -IxxHash -pthread -std=c++20 -Wno-deprecated-volatile -Wno-switch -DGIT_HASH=\"99e836208a5f45cd41b572e51a29fdddd48f8b98\" -O2  -c -o subprocess.o subprocess.cc
subprocess.cc: In function ‘std::function<void()> fork_child()’:
subprocess.cc:33:34: error: taking address of temporary array
   33 |     if (read(pipefd[0], (char[1]){}, 1) == 1)
      |                                  ^~
subprocess.cc: In lambda function:
subprocess.cc:48:44: error: taking address of temporary array
   48 |   return [=]() { write(pipefd[1], (char []){1}, 1); };
      |                                            ^~~
subprocess.cc: In instantiation of ‘void try_resume_daemon(Context<E>&, char**) [with E = X86_64]’:
subprocess.cc:267:1:   required from here
subprocess.cc:156:15: error: taking address of temporary array
  156 |   i64 r = read(conn, (char[1]){}, 1);
      |           ~~~~^~~~~~~~~~~~~~~~~~~~~~
subprocess.cc: In instantiation of ‘void daemonize(Context<E>&, char**, std::function<void()>*, std::function<void()>*) [with E = X86_64]’:
subprocess.cc:267:1:   required from here
subprocess.cc:222:31: error: taking address of temporary array
  222 |   *on_complete = [=]() { write(conn, (char []){1}, 1); };
      |                          ~~~~~^~~~~~~~~~~~~~~~~~~~~~~
subprocess.cc: In instantiation of ‘void try_resume_daemon(Context<E>&, char**) [with E = I386]’:
subprocess.cc:268:1:   required from here
subprocess.cc:156:15: error: taking address of temporary array
  156 |   i64 r = read(conn, (char[1]){}, 1);
      |           ~~~~^~~~~~~~~~~~~~~~~~~~~~
subprocess.cc: In instantiation of ‘void daemonize(Context<E>&, char**, std::function<void()>*, std::function<void()>*) [with E = I386]’:
subprocess.cc:268:1:   required from here
subprocess.cc:222:31: error: taking address of temporary array
  222 |   *on_complete = [=]() { write(conn, (char []){1}, 1); };
      |                          ~~~~~^~~~~~~~~~~~~~~~~~~~~~~
subprocess.cc: In function ‘void process_run_subcommand(Context<E>&, int, char**) [with E = X86_64]’:
subprocess.cc:258:1: warning: ‘noreturn’ function does return
  258 | }
      | ^
subprocess.cc: In function ‘void process_run_subcommand(Context<E>&, int, char**) [with E = I386]’:
subprocess.cc:258:1: warning: ‘noreturn’ function does return
At global scope:
cc1plus: note: unrecognized command-line option ‘-Wno-deprecated-volatile’ may have been intended to silence earlier diagnostics
make: *** [<builtin>: subprocess.o] Error 1

[rui314]

In an attempt to use GCC 11, I tried Ubuntu 21, but it looks like GCC 11 is not available via apt. Did you install GCC 11 from source?

[lancethepants]

Yes, I am using a toolchain compiled from source, so I do have to compile zlib and openssl mold's other dependencies as well. I can go to gcc10 if it's known to be better supported. I had compiled llvm12 but I then realized that the system libstdc++ that it was trying to link against is too old (Debian 10), and I did not compile libc++. When you use clang, is it just linking against the system's libstdc++ library? Now that I have gcc11 I can tell llvm12 to use its libraries, I'll try that. Just kind of nice to use 1 toolchain you know. Have you every compiled using libc++?

[rui314]

My mold binary is linked against libstdc++ which I believe installed as a part of libstdc++6 package. I didn't build libstdc++ nor clang++ myself.

$ ldd mold
        linux-vdso.so.1 (0x00007fffedbb6000)
        libcrypto.so.1.1 => /lib/x86_64-linux-gnu/libcrypto.so.1.1 (0x00007fe124a7c000)
        libtbb.so.2 => /home/ruiu/mold/oneTBB/build/linux_intel64_gcc_cc9.3.0_libc2.31_kernel5.10.0_release/libtbb.so.2 (0x00007fe124a34000)
        libmimalloc.so.1.7 => /home/ruiu/mold/mimalloc/out/release/libmimalloc.so.1.7 (0x00007fe1249fc000)
        libz.so.1 => /lib/x86_64-linux-gnu/libz.so.1 (0x00007fe1249e0000)
        libxxhash.so.0 => /home/ruiu/mold/xxHash/libxxhash.so.0 (0x00007fe1249d0000)
        libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fe1249c8000)
        libstdc++.so.6 => /lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007fe1247e7000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fe124698000)
        libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fe12467d000)
        libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007fe12465a000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fe124468000)
        /lib64/ld-linux-x86-64.so.2 (0x00007fe124d73000)

If you just want to use mold on your machine, you can build mold as a statically-linked executable. All you need to do is to build mold as make STATIC=1 on a Ubuntu 20 machine (in a docker environment or something).

[rui314]

I made a few changes to make it compile with GCC 11.