Trailing semi-colon

[prescod]

Hello! KDL looks interesting!

One of the benefits of having closing curly bases should be that it is obvious when a statement ends.

Unfortunately the KDL grammar seems to force me to explicitly end the statement (at least in my experiments and my skim of the grammar:

object1 {object2;}

Rather than:

object1 {object2}

I had thought that the use of semi-colon as terminator, rather than separator, had gone out of fashion. At least in languages where semicolons are usually optional.

What benefit does the end-user get from the grammar requiring a statement terminator? Would you consider making it optional, even at a cost of complicating the text of the grammar?

[hkolbeck]

That's only the case when placing the terminating brace on the same line as a child node. I think it could be worth considering to allow the latter example, but encouraging placement of children on the same line feels like it could be an anti-pattern. At the end of the day, requiring a single character extra in a corner case doesn't feel like it's terribly user-impacting.

[jihchi]

Just FYI. I found there is an issue discussing the same thing: kdl-org/kdl-rs#11 .

The issue seems like a spec compliance bug if I understand the issue above correctly.

[hkolbeck]

I'm a little confused, I don't see how node-terminator := single-line-comment | newline | ';' | eof allows a child on the same line to not have an explicit ; terminator?

[hkolbeck]

kdl4j will also reject node {child} currently, I'll add a test to the suite whichever direction is decided on.

[zkat]

What's the proposed grammar for this?

[larsgw]

'{' nodes unterminated-node? '}'

Maybe whitespace at the end.

'{' nodes (unterminated-node ws*)? '}'

[hkolbeck]

After some additional thought, the definition I gave is a terrible idea, because it makes it ambiguous which of nodes and unterminated-node? would match the lone child node with a semicolon. This seems like it could cause issues creating parsers for KDL in many parser generators. If I'm mistaken in that, then I withdraw my withdrawl.

[larsgw]

No, that's probably true. I can look into combining EOF and "no terminator" later today.

[hkolbeck]

I want to make it explicit again that I think this change to the language is unnecessary, and unless it can be done pretty simply I'm not in favor of making it.

[zkat]

I agree. I just don't think we should do it. It's not worth the breaking change, and it's not worth the increased implementation complexity and lookahead requirement.