Segfault with small matrix on release builds

[miakramer]

Steps to reproduce:

z = randn(5, 4)
svds(z, nsv=1)

On the debug build, this works as normal (unfortunately for debugging...). On the release build, most of the time for me this will segfault. Once it failed with a ReadOnlyMemoryError. The stacktrace isn't particularly informative, but here's the top frames in the backtrace that have names:

(gdb) info frame
Stack level 0, frame at 0x7fffffffa9b0:
 rip = 0x7fff8d0dbeb8 in dlarnv_64_; saved rip = 0x7fff7a114b44
 called by frame at 0x7fffffffaa10
 Arglist at 0x7fffffffa518, args: 
 Locals at 0x7fffffffa518, Previous frame's sp is 0x7fffffffa9b0
 Saved registers:
  rbx at 0x7fffffffa978, rbp at 0x7fffffffa980, r12 at 0x7fffffffa988, r13 at 0x7fffffffa990, r14 at 0x7fffffffa998, r15 at 0x7fffffffa9a0, rip at 0x7fffffffa9a8
(gdb) bt
#0  0x00007fff8d0dbeb8 in dlarnv_64_ () from /home/mia/.julia/artifacts/9de7cd7b565653107a5f4628320381237cb55cb8/lib/libopenblas64_.so
#1  0x00007fff7a114b44 in dgetv0_ () from /home/mia/.julia/artifacts/a07ca23fc45868aa2f1c602061add992ff482462/lib/libarpack.so
#2  0x00007fff7a11ca10 in dsaitr_ () from /home/mia/.julia/artifacts/a07ca23fc45868aa2f1c602061add992ff482462/lib/libarpack.so
#3  0x00007fff7a11e4f9 in dsaup2_ () from /home/mia/.julia/artifacts/a07ca23fc45868aa2f1c602061add992ff482462/lib/libarpack.so
#4  0x00007fff7a11f307 in dsaupd_ () from /home/mia/.julia/artifacts/a07ca23fc45868aa2f1c602061add992ff482462/lib/libarpack.so

I'm not sure if it's a problem in Arpack.jl, arpack or openblas. Let me know if I can provide any more information or if I should move the bug report there. Also happy to try to track down the problem more, but it's quite tricky since it doesn't seem to happen on debug builds.

[andreasnoack]

Is this your own source build of Julia? Could you please share the output from versioninfo() and the ] st?

[miakramer]

The segfault is on the Manjaro packaged version of Julia, versioninfo():

Julia Version 1.5.3
Commit 788b2c77c1* (2020-11-09 13:37 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-10.0.1 (ORCJIT, ivybridge)

]st:

  [7d9fca2a] Arpack v0.5.1
  [6e4b80f9] BenchmarkTools v0.5.0
  [336ed68f] CSV v0.8.3
  [42fd0dbc] IterativeSolvers v0.9.0
  [7a12625a] LinearMaps v3.2.1
  [8314cec4] PGFPlotsX v1.2.10
  [91a5bcdd] Plots v1.10.6
  [d330b81b] PyPlot v2.9.0
  [90137ffa] StaticArrays v1.0.1
  [24249f21] SymPy v1.0.40
  [3d5dd08c] VectorizationBase v0.19.0

Happy to do a release build from source and test again if that would be informative. The debug build was the same v1.5.3 tag from the source.

[andreasnoack]

The segfault is on the Manjaro packaged version of Julia

There have been many issues with julia versions distributed by the various Linux package systems. See e.g. #5 and the linked issues. I'll recommend that you try out the official binaries from https://julialang.org/.

[miakramer]

Interesting. You're right, it runs fine on a release build from source. Seems to be the same issue as #5, I will close this one.