If you look at a heap snapshot, all the edges coming from an array all report an index of [0]:

Looking at the .heapsnapshot file produced by the current C++ code, it looks like we are constructing edges with garbage values. Here is a snippet from one .heapsnapshot file I produced:

0,2,1
0,3,1
0,5,2
0,7,3
1,8,5
2,18446744073709551615,6
2,18446744073709551615,7
2,18446744073709551615,8
2,18446744073709551615,7
2,18446744073709551615,9
2,18446744073709551615,7
2,18446744073709551615,10
2,18446744073709551615,7
2,18446744073709551615,11
2,18446744073709551615,7

The first field is the edge type, and 2 is "element", meaning this is an array item.

The second field is supposed to be the array index, but all of them have the same value: 18446744073709551615, which is -1 in Int64:

julia> Unsigned(18446744073709551615)
0x0000000000000000ffffffffffffffff

julia> Unsigned(18446744073709551615) == typemax(UInt64)
true

This value is coming from gc_slot_to_arrayidx(from, to), here:

So somehow the gc_slot_to_arrayidx value is wrong for all of the edges:

cat /Users/nathandaly/src/julia/37833_1019974912726708.heapsnapshot.edges | grep '^2,' | grep -v '18446744073709551615' | wc -l
       0

Is it valid to call gc_slot_to_arrayidx(from, to) on the parent/child objects? The callsites look like this, for example:

Asking for support here from @gbaraldi, @apaz-cli, @d-netto. Thanks in advance!