Crash right before Stack Overflow with ntuple construction

[jballanc]

This is occurring on v0.5 (Could be related to #17109)

If calling ntuple(f::F, n::Int) with n leads to a Stack Overflow, then calling with n-1 leads to a crash. For example:

julia> ntuple(x -> 1, 436521);
ERROR: StackOverflowError:
 in ntuple(::##35#36, ::Int64) at ./tuple.jl:61 (repeats 79994 times)

julia> ntuple(x -> 1, 436520);
fatal: error thrown and no exception handler available.
ErrorException("task switch not allowed from inside gc finalizer")
rec_backtrace at /Users/jballanc/Source/julia/src/stackwalk.c:83
record_backtrace at /Users/jballanc/Source/julia/src/task.c:228
jl_error at /Users/jballanc/Source/julia/src/builtins.c:42
jl_switchto at /Users/jballanc/Source/julia/src/task.c:380
yieldto at ./event.jl:89
unknown function (ip: 0x107c00bb0)
jl_call_method_internal at /Users/jballanc/Source/julia/src/./julia_internal.h:86
wait at ./event.jl:122
task_done_hook at ./task.jl:161
unknown function (ip: 0x315a1b5c9)
jl_call_method_internal at /Users/jballanc/Source/julia/src/./julia_internal.h:86
jl_apply at /Users/jballanc/Source/julia/src/./julia.h:1388

[pabloferz]

This one should be fixed after #17707

[KristofferC]

This specific example yes but probably not the more general segfault at StackOverflow - 1.

[pabloferz]

Of course, but that is also reported in #17109, right?

[KristofferC]

Oh sorry, I mixed the two of them up.

[jballanc]

I had initially opened this separate from #17109 because there were some suggestive differences in the stack traces between the two. In other words, #17109 crashes in libunwind suggesting a possible off-by-one on frame creation, but this one crashes in a GC finalizer.

I suppose it might be possible that they are still the same (not knowledgeable enough myself to know if the GC finalizer would be called during stack unwind after a stack overflow), but regardless if this variant of the bug is no longer reproducible, I suppose it could be closed.

[yuyichao]

but this one crashes in a GC finalizer.

Why do you say that?

[jballanc]

I was assuming this was the case based on this line from the stack trace:

ErrorException("task switch not allowed from inside gc finalizer")

But also, in this case the Julia process crashes with an uncaught exception, which is different than the segfault in #17109 (though, again, that doesn't mean they don't both have the same root cause).

[JeffBezanson]

Closing as dup of #17109; this manifestation should be fixed.