Missed stackoverflow on large tuple due to lack of stack probing

[tkf]

I get a segfault with

foldl(|>, repeat([x -> (x, x, x, x, x, x, x)], 8); init = (1,))

rr trace: https://s3.amazonaws.com/julialang-dumps/reports/2020-06-06T04-32-54-tkf.tar.zst

It creates a huge tuple so I'm not sure if I should be reporting this as a bug. But just in case you need yet another piece of code that is hostile to the compiler...

[Keno]

The issue is that this allocates a huge tuple on the stack, which then skips over the guard page, so we don't see it as a stack overflow. LLVM has "probe-stack" support which could mitigate this.

[c42f]

Seems related to #28577 — we discussed probe-stack there as well.