addprocs works in terminal, fails in Emacs

[danielmatz]

I have a script that runs fine in the terminal, but fails when I try to run it from Emacs.

I can reproduce this with a file that contains only addprocs(1).

Here's the error output:

signal (4): Illegal instruction: 4
while loading /Users/dmatz/Desktop/test.jl, in expression starting on line 1
abort at /usr/local/opt/llvm37-julia/lib/llvm-3.7/lib/libLLVM-3.7.dylib (unknown line)
uv_thread_create at /private/tmp/julia-20160920-99724-ulo33f/deps/srccache/libuv-8d5131b6c1595920dd30644cd1435b4f344b46c8/src/unix/thread.c:88
init_once at /private/tmp/julia-20160920-99724-ulo33f/deps/srccache/libuv-8d5131b6c1595920dd30644cd1435b4f344b46c8/src/threadpool.c:173
__pthread_once_handler at /usr/lib/system/libsystem_pthread.dylib (unknown line)
_os_once at /usr/lib/system/libsystem_platform.dylib (unknown line)
pthread_once at /usr/lib/system/libsystem_pthread.dylib (unknown line)
uv_once at /private/tmp/julia-20160920-99724-ulo33f/deps/srccache/libuv-8d5131b6c1595920dd30644cd1435b4f344b46c8/src/unix/thread.c:239
uv__work_submit at /private/tmp/julia-20160920-99724-ulo33f/deps/srccache/libuv-8d5131b6c1595920dd30644cd1435b4f344b46c8/src/threadpool.c:184
uv_getaddrinfo at /private/tmp/julia-20160920-99724-ulo33f/deps/srccache/libuv-8d5131b6c1595920dd30644cd1435b4f344b46c8/src/unix/getaddrinfo.c:186
jl_getaddrinfo at /private/tmp/julia-20160920-99724-ulo33f/src/jl_uv.c:725
getaddrinfo at ./socket.jl:574
jlcall_getaddrinfo_33144 at /usr/local/Cellar/julia/0.5.0/lib/julia/sys.dylib (unknown line)
getaddrinfo at ./socket.jl:589
connect! at ./socket.jl:686
connect at ./stream.jl:959 [inlined]
connect_to_worker at ./managers.jl:483
unknown function (ip: 0x317697039)
jl_call_method_internal at /private/tmp/julia-20160920-99724-ulo33f/src/./julia_internal.h:189 [inlined]
jl_apply_generic at /private/tmp/julia-20160920-99724-ulo33f/src/gf.c:1942
connect at ./managers.jl:425
jlcall_connect_23326 at /usr/local/Cellar/julia/0.5.0/lib/julia/sys.dylib (unknown line)
jl_call_method_internal at /private/tmp/julia-20160920-99724-ulo33f/src/./julia_internal.h:189 [inlined]
jl_apply_generic at /private/tmp/julia-20160920-99724-ulo33f/src/gf.c:1942
create_worker at ./multi.jl:1786
setup_launched_worker at ./multi.jl:1733
unknown function (ip: 0x317692b1d)
jl_call_method_internal at /private/tmp/julia-20160920-99724-ulo33f/src/./julia_internal.h:189 [inlined]
jl_apply_generic at /private/tmp/julia-20160920-99724-ulo33f/src/gf.c:1942
#649 at ./task.jl:360
unknown function (ip: 0x31769273f)
jl_call_method_internal at /private/tmp/julia-20160920-99724-ulo33f/src/./julia_internal.h:189 [inlined]
jl_apply_generic at /private/tmp/julia-20160920-99724-ulo33f/src/gf.c:1942
jl_apply at /private/tmp/julia-20160920-99724-ulo33f/src/./julia.h:1392 [inlined]
start_task at /private/tmp/julia-20160920-99724-ulo33f/src/task.c:253
Allocations: 1273026 (Pool: 1272116; Big: 910); GC: 0

signal (6): Abort trap: 6
while loading /Users/dmatz/Desktop/test.jl, in expression starting on line 1
sigprocmask at /usr/lib/system/libsystem_kernel.dylib (unknown line)
Allocations: 1273026 (Pool: 1272116; Big: 910); GC: 0

After it errors out, there is an orphaned julia process that has 100% CPU usage.

I'm on Mac OS X. I used homebrew to install Julia.

julia> versioninfo()
Julia Version 0.5.0
Commit 3c9d753* (2016-09-19 18:14 UTC)
Platform Info:
  System: Darwin (x86_64-apple-darwin14.5.0)
  CPU: Intel(R) Core(TM) i7-3615QM CPU @ 2.30GHz
  WORD_SIZE: 64
  BLAS: libopenblas (DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas
  LIBM: libopenlibm
  LLVM: libLLVM-3.7.1 (ORCJIT, ivybridge)

[vtjnash]

From looking at the codepath there, it looks like this should only happen if the parent process (in this case emacs), set RLIMIT_STACK to an invalid value? So this is probably an emacs bug, although we can probably be more robust against it.

[vtjnash]

min repro:

ulimit -S -s 7001
./julia -e 'getaddrinfo("localhost")'

[andreasnoack]

@vtjnash This still segfaults. Was the fix ever applied to our libuv fork?

[vtjnash]

I don't think it was, no.

[jgorham]

Just stumbled over this thread as well- I'm guessing the precompiled dmg was compiled with this older version of libuv. Would be nice to get an updated version up if possible! I'm going to pull down the source now.

[swadey]

also seems to happen when JULIA_NUM_THREADS is set to > 1

[hawkish]

This looks like my issue. #24169