SystemStackError is silenced

[ptarjan]

In this code:

require 'interception'
Interception.listen do |a, b|
  puts a, b
end

def foo
  foo
end

foo

puts 'done'

It silences the error and totally eats any debug messages:

$ ruby /tmp/a.rb
$ echo $?
1

We expected to get something similar to what you get without your gem:

$ cat /tmp/a.rb
def foo
  foo
end

foo

puts 'done'
$ ruby /tmp/a.rb
/tmp/a.rb:2:in `foo': stack level too deep (SystemStackError)
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:2:in `foo'
         ... 10067 levels...
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:2:in `foo'
        from /tmp/a.rb:5:in `<main>'

[zenazn]

At a guess, during a SystemStackError we have extremely few stack frames to work with (since the stack is yet to be unwound), and something (my guess is https://github.com/ConradIrwin/interception/blob/master/lib/interception.rb#L104-L106) in interception is causing us to blow the stack again, re-entering our raise TracePoint.

Since that call occurs in a self.rescueing = true block, all further exceptions are nop'd.

Thoughts?

[headius]

Stack errors are almost always fatal; they can happen inside teardown code, leaving resources dangling. Intercepting stack errors almost never works like you expect.

[russell-stripe]

It looks like the root cause is TracePoint, which Interception relies on; I'm able to reproduce with TracePoint alone:

TracePoint.new(:raise){}.enable
def foo; foo; end
foo

This bug has been filed with Ruby for a year, but so far no response: https://bugs.ruby-lang.org/issues/11667.

Welp.