Regression with --compile=min for 1.2/master compared to 1.0/1.1

[fredrikekre]

The following segfaults on 1.0.3, works fine on 1.1.0, and errors on master:

julia --compile=min -e 'using Pkg; pkg"add https://github.com/JuliaLang/Example.jl"'

1.0.3:

signal (11): Segmentation fault
in expression starting at no file:0
jl_isa at /home/fredrik/julia10/src/subtype.c:1251
jl_typemap_entry_assoc_exact at /home/fredrik/julia10/src/typemap.c:169
jl_typemap_assoc_exact at /home/fredrik/julia10/src/julia_internal.h:883 [inlined]
jl_lookup_generic_ at /home/fredrik/julia10/src/gf.c:2135 [inlined]
jl_apply_generic at /home/fredrik/julia10/src/gf.c:2181
do_call at /home/fredrik/julia10/src/interpreter.c:324
eval_value at /home/fredrik/julia10/src/interpreter.c:430
eval_stmt_value at /home/fredrik/julia10/src/interpreter.c:363 [inlined]
eval_body at /home/fredrik/julia10/src/interpreter.c:682
jl_interpret_call_callback at /home/fredrik/julia10/src/interpreter.c:777
unknown function (ip: 0xfffffffffffffffe)
...

master (b126788):

ERROR: this intrinsic must be compiled to be calledfatal: error thrown and no exception handler available.
ErrorException("type MethodInstance has no field inferred")
rec_backtrace at /home/fredrik/julia-master/src/stackwalk.c:94
record_backtrace at /home/fredrik/julia-master/src/task.c:210 [inlined]
jl_throw at /home/fredrik/julia-master/src/task.c:417
jl_errorf at /home/fredrik/julia-master/src/rtutils.c:77
jl_field_index at /home/fredrik/julia-master/src/datatype.c:903
jl_f_getfield at /home/fredrik/julia-master/src/builtins.c:731
getproperty at ./Base.jl:20 [inlined]
lookup at ./stacktraces.jl:127
#process_backtrace#647 at ./errorshow.jl:607
jfptr_#process_backtrace#647_5792 at /home/fredrik/julia-master/usr/lib/julia/sys.so (unknown line)
jl_apply_generic at /home/fredrik/julia-master/src/gf.c:2191
do_call at /home/fredrik/julia-master/src/interpreter.c:326
eval_value at /home/fredrik/julia-master/src/interpreter.c:414
eval_stmt_value at /home/fredrik/julia-master/src/interpreter.c:365 [inlined]
eval_body at /home/fredrik/julia-master/src/interpreter.c:775
jl_interpret_call_callback at /home/fredrik/julia-master/src/interpreter.c:857
Interpreter frame (ip: 0)
...

[fredrikekre]

Works on the 1.0.4 backport branch.

[KristofferC]

Seems the problem with the backtrace is the inferred field access at:

julia/base/stacktraces.jl

Lines 126 to 127 in b126788

	if ip.code isa Core.MethodInstance && ip.code.def isa Method
	codeinfo = ip.code.inferred

[fredrikekre]

Hmm, I could reproduce the segfault on 1.0.3 consistenly 10 times in a row before opening the issue, but not anymore so I guess this is just a problem on master then...

[fredrikekre]

Same error on the 1.2 backport branch as on master.

[JeffBezanson]

The initial problem is the atomic_add! for the refcount in LibGit2 constructors.

[fredrikekre]

Here is another example which results in the same segfault as in my original post:

bug.jl:

using Plots
x = range(0, stop=6π, length=1000)
y = sin.(x)
r = plot(x, y)

function display_dict(x)
    @info "in display_dict"
    buf = IOBuffer()
    show(IOContext(buf, :limit=>true, :color=>true), MIME("text/plain"), x)
    return String(take!(buf))
end

str = display_dict(r)
println(str)

Now,

$ julia --compile=min bug.jl

crashes with the same stracktrace as the original post on Julia 1.0.3. Seems to crash from show.

[JeffBezanson]

I can't reproduce the segfault. I either get ERROR: this intrinsic must be compiled (on master) or it works (on the branch of my PR).

[fredrikekre]

Yea only happens on 1.0.3

[fredrikekre]

@JeffBezanson do you have any idea what might have fixed #31965 (comment) ? Worth a bisect for a backport to 1.0?