Stackoverflow segfault from reload

[quinnj]

I made a package named Segfault.jl and put this in it:

module Segfault

struct Bar
    id::String
    Bar() = new()
    Bar(nm::String) = new(nm)
end

Bar(; kwargs...) = (b = Bar(); b.nm = kargs[1]; return b)

const DEFAULT_BAR = Bar()

end # module

Then calling reload("Segfault") from the repl resulted in a segfault. If the code is loaded interactively at the repl, an appropriate StackOverflowError is reported.

julia> versioninfo()
Julia Version 0.7.0-DEV.1172
Commit 88b3137b32* (2017-07-30 03:30 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin16.7.0)
  CPU: Intel(R) Core(TM) i7-4870HQ CPU @ 2.50GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, haswell)

[musm]

Even simpler

julia> g() = nothing
g (generic function with 1 method)

julia> g(;x=nothing) = (g(); nothing)
g (generic function with 1 method)

julia> g()
ERROR: StackOverflowError:

[tkelman]

segfaulting instead of stack overflow on mac has been reported many times: #14284, #17109, #17280, #19285, #21072, #21258, #22853

I wonder if using llvm's unwinder on mac would be any more reliable here than our osxunwind fork.

[Roger-luo]

Err, is #31096 the same with this one? @KristofferC

@musm 's example works fine on 1.1. no seg faults.

NVM, I got a Stackoverflow error on Ubuntu.

[KristofferC]

It might not be the exact same thing, but crashing instead of stackoverflow on macOS is known (as the comment above shows, e.g. #17109).