Julia: Segfault on clean build, FreeBSD 12.1

[ararslan]

I updated my FreeBSD system from 12.0 to 12.1, cleaned out my Julia clone (git clean -fdx) and rebuilt from scratch with the default options and no ccache. It seems there was a segfault while building the sysimage:

    LINK usr/lib/libjulia.so.1.5
ld: warning: found local symbol '__bss_start' in global part of symbol table in file /usr/home/alex/Projects/julia/usr/lib/libLLVM.so
ld: warning: found local symbol '_end' in global part of symbol table in file /usr/home/alex/Projects/julia/usr/lib/libLLVM.so
ld: warning: found local symbol '_edata' in global part of symbol table in file /usr/home/alex/Projects/julia/usr/lib/libLLVM.so
    CC ui/repl.o
    LINK usr/bin/julia
    JULIA usr/lib/julia/corecompiler.ji
gmake[1]: *** [sysimage.mk:61: /usr/home/alex/Projects/julia/usr/lib/julia/corecompiler.ji] Segmentation fault (core dumped)
gmake: *** [Makefile:81: julia-sysimg-ji] Error 2

The full build output can be found here. The system compiler is Clang 8.0.1 and the local GCC version is 9.2.0.

EDIT: Also note the LLVM deprecation messages during the build; I don't recall getting those on Linux or macOS.

[ararslan]

Seems to build and work just fine if I rebuild with USE_BINARYBUILDER=0. A problem with the cross-compiled LLVM, maybe? @staticfloat?

[ararslan]

Also, the nightly binaries as of fbc2c0aec1, version 1.5.0-DEV.221, seem to be operable AFAICT...

[giordano]

Any chance to find out which is the offending binary, maybe by using binary search? :grin:

[ararslan]

which is the offending binary

Not sure what you mean. Like which dependency is the problem?

[giordano]

Yes

[staticfloat]

Also, the nightly binaries as of fbc2c0a, version 1.5.0-DEV.221, seem to be operable AFAICT...

So you're saying the nightly binaries work, but you can't compile from source using BB deps?

[ararslan]

Yes, exactly.

[staticfloat]

Can you run the bootstrap command in gdb or lldb to see where the segfault is coming from?

[ararslan]

I'm not sure how exactly to do that. Any tips?

[JeffBezanson]

This should work:

cd base
gdb ../julia
(gdb) r --output-ji x compiler/compiler.jl

[ararslan]

$ gdb ../julia
GNU gdb (GDB) 9.1 [GDB v9.1 for FreeBSD]
<...>
Reading symbols from ../julia...
(gdb) r --output-ji x compiler/compiler.jl
Starting program: /usr/home/alex/Projects/julia/julia --output-ji x compiler/compiler.jl
[New LWP 100768 of process 10478]

Thread 1 received signal SIGSEGV, Segmentation fault.
0x000000080048c0c8 in uv__to_stat (src=0x7fffffffde10, dst=0x19) at src/unix/fs.c:1104
1104    src/unix/fs.c: No such file or directory.
(gdb) bt
#0  0x000000080048c0c8 in uv__to_stat (src=0x7fffffffde10, dst=0x19) at src/unix/fs.c:1104
#1  0x000000080048bdbf in uv__fs_stat (path=0x80000006400 <error: Cannot access memory at address 0x80000006400>, buf=0x19)
    at src/unix/fs.c:1260
#2  0x0000000000000000 in ?? ()

[ararslan]

Looks like it could be an issue with the libuv provided by BinaryBuilder. Any suggestions for how to confirm?

[ararslan]

Quoting Elliot from a discussion elsewhere,

I think this is something having to do with attempting to link a static library that was built on 11 against system libraries from 12

One potential fix here may be to dynamically link libuv.

[ararslan]

Something has changed, and now it's a sigabort rather than a segfault

$ gdb ../julia
GNU gdb (GDB) 9.1 [GDB v9.1 for FreeBSD]
<...>
Reading symbols from ../julia...
(gdb) r --output-ji x compiler/compiler.jl
Starting program: /usr/home/alex/Projects/julia/julia --output-ji x compiler/compiler.jl
[New LWP 100873 of process 31424]

Thread 1 received signal SIGABRT, Aborted.
0x0000000800d7a1ba in thr_kill () from /lib/libc.so.7
(gdb) bt
#0  0x0000000800d7a1ba in thr_kill () from /lib/libc.so.7
#1  0x0000000800d785e4 in raise () from /lib/libc.so.7
#2  0x0000000800cec7e9 in abort () from /lib/libc.so.7
#3  0x000000080049fbb0 in uv__io_poll (loop=0x800810380 <default_loop_struct>, timeout=0) at src/unix/kqueue.c:239
#4  0x000000080048802d in uv_run (loop=0x800810380 <default_loop_struct>, mode=UV_RUN_DEFAULT) at src/unix/core.c:359
#5  0x00000008003b63fa in jl_atexit_hook (exitcode=<optimized out>) at init.c:284
#6  0x0000000000202395 in main (argc=<optimized out>, argv=<optimized out>) at repl.c:228

Could be the same underlying issue though