Segfault on method error

[iblislin]

Got segault on master (39f668c) and 0.6.1-pre.92

julia -e 'using Flux; struct A end; A() * 1'

I expect it will throw MethodError.

[removed spurious issue – mike]

[SimonDanisch]

That's pretty much expected. Internals like pointer arithmetic uses *(::Int, ::Int), so redefining that to something else pretty much needs to segfault...

[martinholters]

Dup of #20945, #22132, #23605, #21944, #24078

[iblislin]

hmm 🤔 ... the original case is

julia -e 'using Flux; struct A end; A() * 1'

but I do not think Flux redefine *(::Int, ::Int)

[KristofferC]

So, what was the reason you put that in your first post?

[iblislin]

just found it crashed as well, and guess the cause is the same...

[MikeInnes]

Lol, unfortunately Julia will segfault for more than one reason. Reopening to reflect the real issue.

[iblislin]

└─[iblis@kaladbolg Oops]% julia -e 'using Flux; struct A end; A() * 1'               
                                                                                   
ERROR:                                                                               
signal (11): Segmentation fault                                                      
while loading no file, in expression starting on line 0                              
jl_egal at /home/iblis/git/julia06/src/builtins.c:114                                
show at ./show.jl:1099                                                               
show at ./show.jl:208                                                                
unknown function (ip: 0x7f10f7d0ea1f)                    
jl_call_fptr_internal at /home/iblis/git/julia06/src/julia_internal.h:339 [inlined]
jl_call_method_internal at /home/iblis/git/julia06/src/julia_internal.h:358 [inlined]
jl_apply_generic at /home/iblis/git/julia06/src/gf.c:1926                            
print at ./strings/io.jl:29                                                          
unknown function (ip: 0x7f10e8062e26)                    
jl_call_fptr_internal at /home/iblis/git/julia06/src/julia_internal.h:339 [inlined]
jl_call_method_internal at /home/iblis/git/julia06/src/julia_internal.h:358 [inlined]
jl_apply_generic at /home/iblis/git/julia06/src/gf.c:1926
#print_to_string#229 at ./strings/io.jl:102
jl_call_fptr_internal at /home/iblis/git/julia06/src/julia_internal.h:339 [inlined]
jl_call_method_internal at /home/iblis/git/julia06/src/julia_internal.h:358 [inlined]
jl_apply_generic at /home/iblis/git/julia06/src/gf.c:1926
string at ./strings/io.jl:120
jl_call_fptr_internal at /home/iblis/git/julia06/src/julia_internal.h:339 [inlined]
jl_call_method_internal at /home/iblis/git/julia06/src/julia_internal.h:358 [inlined]
jl_apply_generic at /home/iblis/git/julia06/src/gf.c:1926
show_method_candidates at ./replutil.jl:511
unknown function (ip: 0x7f10e806187d)
jl_call_fptr_internal at /home/iblis/git/julia06/src/julia_internal.h:339 [inlined]
jl_call_method_internal at /home/iblis/git/julia06/src/julia_internal.h:358 [inlined]
jl_apply_generic at /home/iblis/git/julia06/src/gf.c:1926
showerror at ./replutil.jl:415
#476 at ./replutil.jl:213
...

[martinholters]

Reminds me of #22842.

[martinholters]

With Julia 0.6.1-pre.92 and Flux 0.3.2, I get the expected MethodError, no crash.
With Julia 0.7.0-DEV.2157 (or 0.7.0-DEV.2136 corresponding to 39f668c) and Flux 0.3.2, using Flux errors (due to an assertion in Hiccup).
Can you provide the versions of Flux and its dependencies you were using?

[iblislin]

Julia 0.6.1-pre.92 and Flux master (83cc77c) got randomly crashing.
(Sometimes it pop MethodError correctly.)

julia> Pkg.status.(["DataFlow", "Juno", "MacroTools", "NNlib", "ForwardDiff", "Requires"])
 - DataFlow                      0.3.0+             master
 - Juno                          0.3.1                                   
 - MacroTools                    0.3.7                                   
 - NNlib                         0.1.2+             act-elu
 - ForwardDiff                   0.6.0
 - Requires                      0.4.3
6-element Array{Void,1}:
 nothing
 nothing
 nothing
 nothing
 nothing
 nothing

julia> versioninfo()
Julia Version 0.6.1-pre.92
Commit 389b23cf6e* (2017-10-07 01:18 UTC)
Platform Info:                                                                                                                                    
  OS: Linux (x86_64-pc-linux-gnu)                                                                                                                 
  CPU: Intel(R) Xeon(R) CPU E3-1231 v3 @ 3.40GHz                                                                                                  
  WORD_SIZE: 64                                                                                                                                   
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)                                                                                
  LAPACK: libopenblas64_                                                                                                                          
  LIBM: libopenlibm                                                                                                                               
  LLVM: libLLVM-3.9.1 (ORCJIT, haswell)

[iblislin]

With Julia 0.7.0-DEV.2157 (or 0.7.0-DEV.2136 corresponding to 39f668c) and Flux 0.3.2, using Flux errors (due to an assertion in Hiccup).

well, got same error during using here.

[martinholters]

Hm, I still don't see the crash with the versions you gave, where DataFlow master is 417105f, except for NNlib which I left at 0.1.2. Meanwhile, does using Flux; methods(*) crash for you? Can you try applying f4619d0 locally to see whether that fixes it?

[iblislin]

Meanwhile, does using Flux; methods(*) crash for you?

Before patching, it does not crash... I ran it more than 10 times, no crashing.

Can you try applying f4619d0 locally to see whether that fixes it?

After patching, it does not run into any crash with using Flux; struct A end; A() * 1. 👍
Will this patch be shipped with 0.6.1 releasing?

[martinholters]

Will this patch be shipped with 0.6.1 releasing?

#23831 has the backport pending tag and should be trivial to backport, but it was probably too late to be considered in #23670. OTOH, 0.6.1 isn't out, so... @ararslan?

[ararslan]

Won't make it into 0.6.1 but we can always do a 0.6.2.