RCR 336: Allow rescue, else, and ensure clauses on do/end blocks

Setting up an exception handler for every block invocation will be very expensive.

---

Setting up an exception handler for every block is *not* necessary -- only blocks with rescue clauses will have exception handling.

---

When I first saw a method def with a rescue I actually assumed that a rescue clause was supported for all block constructs. I wasn't until a couple of days later when I tried it on a if/end and it failed that I realized that it was a special case for method and class defs. Since then, I have often wished I could add rescue clauses to block structures that are currently do not support it.

I like adding optional rescue clauses to do/end but I think that rescue clauses should be supported on all syntax block constructs. I think that the same arguments for having rescue clauses on methods and class defs can be applied to having them on any block construct. And having the everywhere makes it easier to explain and think about (for me at least).

Peter

---

Peter: there's a problem with if and unless blocks and rescue clauses.

  # this is valid code
  begin
    # do something
  rescue
    # if there's an exception
  else  # <---- uh-oh
    # if there are no exceptions
  end

See the `else' part? That would make if and unless blocks look odd:

  if foo == bar
    # do something
  else
    # do something else
  rescue
    # if there's an exception
  else
    # if there are no exceptions
  end

Daniel Schierbeck

---

This should not apply to just do/end, but all *block* forms (which would exclude the if/unless syntax forms).

---

 This should not apply to just do/end, but all *block* forms (which would exclude the if/unless syntax forms).

I agree with this and would love to see it. Good idea Daniel.

Brian

---

I'm in agreement about adding this to do..end ,and to the {..} form of block syntax as well.

The problem with priority of else in an if .. [ else] .. end is real but is a separate question.

Getting back to do..end

Notice that a block with just a single statement with a rescue modifier like:

  x.call do 
            something_scary rescue something_not_so_scary
         end

is legal, even though

  x.call do
            something_scare
            rescue
                something_not_so_scary
         end

is not.

This makes me think that this is purely a syntactic restriction, and that there are no semantic arguments against this.

Rick DeNatale

---