do .. done: report "unclosed" error if done is missing

[nojb]

• Allow do done (equivalent to do () done)
• Allow "shortcut" attributes, extensions: do%foo ... done, do[@foo] ... done.
• Give "unclosed" error when done is missing.

[xavierleroy]

Do we write so many busy waiting loops

for i = 1 to n do () done

that we need a special syntax to save three keystrokes? :-)

[alainfrisch]

Do we write so many busy waiting loops

for i = 1 to n do () done

that we need a special syntax to save three keystrokes? :-)

I often write while (...) do () done (as a way to do do-while or similar loops). (I'm not claiming I cannot live with the ().)

[nojb]

No problem taking the do done sugar out of the PR, but do we agree that there is no reason to reject the other two tweaks?

[xavierleroy]

I'm not looking for reasons to reject, just for reasons to accept.

The "unclosed" error explanation is nice to have.

Attributes and extensions on do, I'll let someone more competent than me say.

do done as a shortcut for do () done doesn't seem very useful to me, even for while where it makes more sense than for for. There's something to be said in favor of regularity in the syntax. Plus, all Algol-like languages have explicit empty statements (skip in IMP, a lone ; in C), and () works reasonably well as OCaml's explicit empty statement.

[gasche]

What is the meaning of do[@foo] e done or do%foo e done? If I understand the parsing rule correctly, this is considered as equivalent to do e[@foo] done or do [%foo e] done, which I find a bit confusing. Are we sure that this is the right behavior, is there a precedent for this? Ohter uses of ext_attributes I could easily find apply to the first keyword of the construct (here for, while where they are already supported), not inside.

(I guess I'm not convinced by the "attributes" part of this PR either.)

[nojb]

Are we sure that this is the right behavior, is there a precedent for this?

It is similar to what we do for begin%foo e end and begin[@foo] e end:

ocaml/parsing/parser.mly

Lines 2369 to 2370 in 85a3d4d

	| BEGIN ext = ext attrs = attributes e = seq_expr END
	{ e.pexp_desc, (ext, attrs @ e.pexp_attributes) }

[nojb]

I updated this PR with only the "unclosed" error, which should be uncontroversial :)

[gasche]

Clearly a nice improvement. Thanks!