Wes McKinney created ARROW-2247:
-----------------------------------

             Summary: [Python] Statically-linking boost_regex in both libarrow 
and libparquet results in segfault
                 Key: ARROW-2247
                 URL: https://issues.apache.org/jira/browse/ARROW-2247
             Project: Apache Arrow
          Issue Type: Bug
            Reporter: Wes McKinney


This is a backtrace loading {{libparquet.so}} on Ubuntu 14.04 using boost 
1.66.1 from conda-forge. Both libarrow and libparquet contain {{boost_regex}} 
statically linked. 

{code}
In [1]: import ctypes

In [2]: ctypes.CDLL('libparquet.so')

Program received signal SIGSEGV, Segmentation fault.
0x00007fffed4ad3fb in std::basic_string<char, std::char_traits<char>, 
std::allocator<char> >::basic_string(std::string const&) () from 
/usr/lib/x86_64-linux-gnu/libstdc++.so.6
(gdb) bt
#0  0x00007fffed4ad3fb in std::basic_string<char, std::char_traits<char>, 
std::allocator<char> >::basic_string(std::string const&) () from 
/usr/lib/x86_64-linux-gnu/libstdc++.so.6
#1  0x00007fffed74c1fc in 
boost::re_detail_106600::cpp_regex_traits_char_layer<char>::init() ()
   from /home/wesm/cpp-toolchain/lib/libboost_regex.so.1.66.0
#2  0x00007fffed794803 in 
boost::object_cache<boost::re_detail_106600::cpp_regex_traits_base<char>, 
boost::re_detail_106600::cpp_regex_traits_implementation<char> 
>::do_get(boost::re_detail_106600::cpp_regex_traits_base<char> const&, unsigned 
long) () from /home/wesm/cpp-toolchain/lib/libboost_regex.so.1.66.0
#3  0x00007fffed79e62b in boost::basic_regex<char, boost::regex_traits<char, 
boost::cpp_regex_traits<char> > >::do_assign(char const*, char const*, unsigned 
int) () from /home/wesm/cpp-toolchain/lib/libboost_regex.so.1.66.0
#4  0x00007fffee58561b in boost::basic_regex<char, boost::regex_traits<char, 
boost::cpp_regex_traits<char> > >::assign (this=0x7fffffff3780, 
    p1=0x7fffee600602 
"(.*?)\\s*(?:(version\\s*(?:([^(]*?)\\s*(?:\\(\\s*build\\s*([^)]*?)\\s*\\))?)?)?)",
 
    p2=0x7fffee60064a "", f=0) at 
/home/wesm/cpp-toolchain/include/boost/regex/v4/basic_regex.hpp:381
#5  0x00007fffee5855a7 in boost::basic_regex<char, boost::regex_traits<char, 
boost::cpp_regex_traits<char> > >::assign (this=0x7fffffff3780, 
    p=0x7fffee600602 
"(.*?)\\s*(?:(version\\s*(?:([^(]*?)\\s*(?:\\(\\s*build\\s*([^)]*?)\\s*\\))?)?)?)",
 f=0)
    at /home/wesm/cpp-toolchain/include/boost/regex/v4/basic_regex.hpp:366
#6  0x00007fffee5683f3 in boost::basic_regex<char, boost::regex_traits<char, 
boost::cpp_regex_traits<char> > >::basic_regex (this=0x7fffffff3780, 
    p=0x7fffee600602 
"(.*?)\\s*(?:(version\\s*(?:([^(]*?)\\s*(?:\\(\\s*build\\s*([^)]*?)\\s*\\))?)?)?)",
 f=0)
    at /home/wesm/cpp-toolchain/include/boost/regex/v4/basic_regex.hpp:335
#7  0x00007fffee5656d0 in parquet::ApplicationVersion::ApplicationVersion (
Python Exception <class 'gdb.error'> There is no member named _M_dataplus.: 
    this=0x7fffee8f1fb8 
<parquet::ApplicationVersion::PARQUET_251_FIXED_VERSION>, created_by=)
    at ../src/parquet/metadata.cc:452
#8  0x00007fffee41c271 in __cxx_global_var_init.1(void) () at 
../src/parquet/metadata.cc:35
#9  0x00007fffee41c44e in _GLOBAL__sub_I_metadata.tmp.wesm_desktop.4838.ii ()
   from /home/wesm/local/lib/libparquet.so
#10 0x00007ffff7dea1da in call_init (l=<optimized out>, argc=argc@entry=2, 
argv=argv@entry=0x7fffffff5d88, 
    env=env@entry=0x7fffffff5da0) at dl-init.c:78
#11 0x00007ffff7dea2c3 in call_init (env=<optimized out>, argv=<optimized out>, 
argc=<optimized out>, 
    l=<optimized out>) at dl-init.c:36
#12 _dl_init (main_map=main_map@entry=0x13fb220, argc=2, argv=0x7fffffff5d88, 
env=0x7fffffff5da0)
    at dl-init.c:126
{code}

This seems to be caused by static initializations in libparquet:

https://github.com/apache/parquet-cpp/blob/master/src/parquet/metadata.cc#L34

We should see if removing these static initializations makes the problem go 
away. If not, then statically-linking boost_regex in both libraries is not 
advisable.

For this reason and more, I really wish that Arrow and Parquet shared a common 
build system and monorepo structure -- it would make handling these toolchain 
and build-related issues much simpler. 



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to