Segfault running tests on Linux

[simonster]

On Linux, when I try to run the tests on PyCall master, I get a segfault:

signal (11): Segmentation fault
PyObject_IsInstance at /usr/lib/x86_64-linux-gnu/libpython2.7.so (unknown line)
pyisinstance at /home/simon/.julia/PyCall/src/PyCall.jl:100
jlcall_pyisinstance_21189 at  (unknown line)
jl_apply_generic at /usr/local/julia/usr/bin/../lib/libjulia.so (unknown line)
pyptr_query at /home/simon/.julia/PyCall/src/conversions.jl:147
pytype_query at /home/simon/.julia/PyCall/src/conversions.jl:783
convert at /home/simon/.julia/PyCall/src/conversions.jl:832
roundtripeq at /home/simon/.julia/PyCall/test/runtests.jl:6
jlcall_roundtripeq_21432 at  (unknown line)
jl_apply_generic at /usr/local/julia/usr/bin/../lib/libjulia.so (unknown line)
anonymous at test.jl:92
do_test at test.jl:50
jl_apply_generic at /usr/local/julia/usr/bin/../lib/libjulia.so (unknown line)
unknown function (ip: 0x7f5083fe4ed3)
unknown function (ip: 0x7f5083fe426e)
unknown function (ip: 0x7f5083ff9a0b)
unknown function (ip: 0x7f5083ffa72c)
jl_load at /usr/local/julia/usr/bin/../lib/libjulia.so (unknown line)
include at ./boot.jl:260
jl_apply_generic at /usr/local/julia/usr/bin/../lib/libjulia.so (unknown line)
include_from_node1 at ./loading.jl:271
jl_apply_generic at /usr/local/julia/usr/bin/../lib/libjulia.so (unknown line)
process_options at ./client.jl:316
_start at ./client.jl:411
unknown function (ip: 0x7f50810f0d89)
jl_apply_generic at /usr/local/julia/usr/bin/../lib/libjulia.so (unknown line)
unknown function (ip: 0x401ae7)
unknown function (ip: 0x4016af)
__libc_start_main at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
unknown function (ip: 0x4016f9)
unknown function (ip: (nil))
fish: Job 1, “julia test/runtests.jl” terminated by signal SIGSEGV (Address boundary error)

Adding the line Libdl.dlopen(libpython, Libdl.RTLD_LAZY|Libdl.RTLD_DEEPBIND|Libdl.RTLD_GLOBAL) to the start of __init__ appears to fix this. I'm not really sure what's going on, but it appears the segfault is somehow related to the libpython symbols not being available to other libraries.

I can't reproduce this on OS X, but dlopen is global there by default.

My versioninfo() is:

julia> versioninfo()
Julia Version 0.5.0-dev+3
Commit 7cd05db* (2015-09-09 19:05 UTC)
Platform Info:
  System: Linux (x86_64-linux-gnu)
  CPU: Intel(R) Core(TM) i7-3930K CPU @ 3.20GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

[simonster]

In case it's useful, without RTLD_GLOBAL, importing ctypes here throws:

ImportError('/usr/lib/python2.7/lib-dynload/_ctypes.x86_64-linux-gnu.so: undefined symbol: PyFloat_Type',)

so the catch clause gets taken. With the dlopen line above importing ctypes works, but if I force the code in the catch clause to be run, the tests still segfault.

[tbreloff]

I'm wondering if this crash is related in any way. This was on OS X.

[stevengj]

Have you tried the PyCall master?

[stevengj]

I can reproduce. Will incorporate the dlopen workaround shortly.

[jmxpearson]

Thank you for this. I have been running into this issue as of this afternoon.