Is there any examples where this could be used in the stdlib? I think they would help a lot in making a decision on this PR.

There are quite a few:

• compiler/crystal/codegen/cache_dir.cr:81
• compiler/crystal/macros/macros.cr:196
• compiler/crystal/semantic/top_level_visitor.cr:618
• compiler/crystal/tools/doc/highlighter.cr:11
• file_utils.cr:218
• file_utils.cr:313
• http/common.cr:222
• kernel.cr:132

My gut reaction was: "oh, looks sweet". It fits with the def/rescue/end syntax we already have. Looking at some of the stdlib examples, I believe it would be nice sugar.

Would the following work?

spawn do
  something
rescue ex
  logger.error(ex)
end

@straight-shoota Thank you very much!! And we should fix them after merged this and released next version.

@ysbaddaden Yes, of course!

Can the formatter handle this?

@asterite Of course. I did't touch formatter code, but formatter worked fine. That's fantastic.

Would be nice if we cleaned up the places where rescue was in a block in the stdlib.

My first thought was that it doesn't look right because I'm not used to it, but its consistent with the other block types like def and begin, so I think it's fine. Its a bit more visually noisy than the def and begin variants in the language already though.

Oh, and a formatter spec would be nice even if it wasn't broken before.

Just a quick thought: What about while? It's used very similar to a block, especially compared to each or loop (which is implemented using while), and perhaps it would be nice if it had similar features...
It could however lead to some misunderstandings regarding break and next. It would be reasonable to assume that both trigger ensure, or only next or maybe even none of them - depending on the context. That's why I'm skeptical, but maybe there are other thoughts about this.

It could however lead to some misunderstandings regarding break and next. It would be reasonable to assume that both trigger ensure, or only next or maybe even none of them - depending on the context. That's why I'm skeptical, but maybe there are other thoughts about this.

Note: there's also next for blocks

True. And break also. So this is also relevant with respect to blocks.

@RX14 Added formatter spec for this.

I know Ruby is adding this for 2.5, but after thinking about this a bit more, I wonder if this code is clear:

[1, 2, 3].each do |x|
  # do something
ensure
  # clean up
end

Is the ensure executed after all the items, or once for each item? I know it's after each item, but maybe it's a bit confusing. For a method there's no problem because it's done after the whole method. But for a block that can be executed many times, I don't know...

ensure runs against each items, it seems clear to me because ensure appears inside block and the block runs against each items.

I still 👍 this

My 2c is that I don't find it that obvious (at a glance) that the ensure will be executed for each block invocation. I like the syntax for blocks that'll execute exactly once, for sure.

I think we should be sure to very clearly and explicitly document this behavior.