Bug 472263

Created attachment 160291 [details]
host and client stacktraces

SUMMARY
Valgrind can abort with errors along the lines of

iselVecExpr (amd64, subarch = amd64-cx16-lzcnt-rdtscp-sse3): can't reduce
PermOrZero8x16(t291,t102)
vex: the `impossible' happened:
   iselVecExpr_wrk
vex storage: T total 1856835088 bytes allocated
vex storage: P total 512 bytes allocated

(Full log tail attached.)

valgrind: the 'impossible' happened:
   LibVEX called failure_exit().


STEPS TO REPRODUCE (not so minimal, sorry; additionally requires Boost.Test headers)
1. git clone https://github.com/ncbi/ncbi-vdb
2. cd ncbi_vdb
3. git checkout 3.0.6
4. mkdir Release
5. cd Release
6. cmake -DCMAKE_BUILD_TYPE=Release ..
7. make -s -j5
8. ln -s ../interfaces .
9. mkdir -p linux/x86_64/release
10. ln -s lib linux/x86_64/release/
11. cd ../..
12. git clone https://github.com/ncbi/ncbi-cxx-toolkit-public
13. cd ncbi-cxx-toolkit-public
14. ./cmake-configure --without-debug --with-dll -DNCBI_ThirdParty_VDB=${PWD%/*}/ncbi-vdb/Release
15. cd CMake-*/build
16. make -s -j5 test_wgs_loader
17. valgrind --tool=memcheck --suppressions=../../scripts/common/check/valgrind.supp ../bin/test_wgs_loader --run_test=TestReplacedProtein1

OBSERVED RESULT

Valgrind terminates with the above report and the attached stack trace.

EXPECTED RESULT

Valgrind proceeds to completion.

SOFTWARE/OS VERSIONS
At minimum, Valgrind 3.20.0, 3.21.0, and revision cb684b50e.

ADDITIONAL INFORMATION
I built and tested everything on a CentOS 7 system using a third-party GCC 7.3.0 installation.   FWIW, the problem is new as of ncbi-vdb 3.0.6, even though it occurs with no active VDB calls.