When you run a Ruby command with multiple -e arguments, those "inline scripts" should be executed to completion, each in turn.

The ordering of END and at_exit is also unexpected here. It appears that CRuby always splices END and at_exit to run at the termination of the top-level script...which in this case happens after both -e scripts have run:

$ ruby23 -e "
BEGIN { puts 0 }
puts :a
END { puts 1 }
at_exit { puts 2 }"
-e "
BEGIN { puts 3 }
puts :b
END { puts 4 }
at_exit { puts 5 }"
0
3
a
b
5
4
2
1

@enebo showed me some examples with required files that also show that END blocks are registered in the same process as at_exit blocks and run in reverse order the same way.

We match some of this ordering, but for cases where exceptional results come out of the scripts, things get a little weird:

$ ruby23 -e "END { raise '1' }; at_exit { raise '2' }" -e "raise 'b'; END { raise '4' }; at_exit { raise '5' }"
-e:1:in `block in <main>': 2 (RuntimeError)
-e:1:in `block in <main>': 1 (RuntimeError)
-e:2:in `<main>': b (RuntimeError)

Here, the second block of code does not get to the point of registering its END and at_exit blocks, so they don't run...ok. But the END and at_exit from the first block of code appear to run before the second block of code fails.

JRuby runs this differently, with the raises in END and at_exit coming after the raise in the second block of code:

$ jruby -e "END { raise '1' }; at_exit { raise '2' }" -e "raise 'b'; END { raise '4' }; at_exit { raise '5' }"
RuntimeError: b
  <top> at -e:2
RuntimeError: 2
  block in -e at -e:1
RuntimeError: 1
  block in -e at -e:1

These ordering behaviors and the multiple -e behavior probably need specs. I will be taking some MRI tests since I don't see these as common cases and they're nontrivial to fix right now.