CXX2 violates YAML 1.2 grammar

[hvr]

test-case CXX2 considers

--- &anchor a: b

a valid yaml-stream; but in fact, the grammar doesn't allow for block collection nodes to appear on the --- line

This also excludes the simpler

--- a: b

or also sequences such as

--- - x

The reason becomes obvious when we follow/trace the productions:

[208]  l-explicit-document  ::=  c-directives-end  l-bare-document

[207]  l-bare-document  ::=  s-l+block-node(-1,block-in)

n = -1
c = block-in

[196]  s-l+block-node(n,c)  ::=  s-l+block-in-block(n,c) | s-l+flow-in-block(n)

[198]  s-l+block-in-block(n,c)  ::=  s-l+block-scalar(n,c) | s-l+block-collection(n,c)

[200] s-l+block-collection(n,c) ::= ( s-separate(n+1,c) c-ns-properties(n+1,c) )?
                                    s-l-comments
                                    ( l+block-sequence(seq-spaces(n,c)) | l+block-mapping(n) )   

[79]    s-l-comments    ::=     ( s-b-comment | /* Start of line */ ) l-comment* 

[77]    s-b-comment     ::=     ( s-separate-in-line c-nb-comment-text? )?  b-comment

[76]  b-comment  ::=  b-non-content | /* End of file */ 

[30]  b-non-content  ::=  b-break

so the problem is that in order to match s-l+block-collection(n,c) (rule 200), we must match s-l-comments, but since we are not on column 1 (because we're 3 characters into the line due to ---), we must instead match s-b-comment; and that demands to match a line-break via b-break (or an EOF).

Hence, we cannot match a s-l+block-collection on the same line right after the directives-end marker ---.

QED :-)

---

PS: One obvious workaround would be to make matching s-l-comments optional; but that would make other currently disallowed cases such as

k: - x
   - y

(for which iirc there's at least one tests which tests that YAML parser refuse it as a a parsing error). I don't know if there's an easy fix to make the grammar accept CXX2 without at the same being too liberal and allowing other syntax that isn't intended to be valid.

[perlpunk]

looking at that, it seems you're right. thanks for listing the productions. see also the related #29 @ingydotnet @flyx makes sense? should we mark the test as an error?

[hvr]

Btw, what's valid however is something like

--- &foo !!seq
- x
- y

or also

--- &foo
a: b

but iirc we got tests for that...

[flyx]

I think I overlooked this detail until now. It's a pretty nasty gotcha that something named s-l-comments can force a newline ;).

The test input is indeed illegal according to the productions and should be an error.

[perlpunk]

@flyx cool. I'll modify the test to be an error.

[perlpunk]

thanks @hvr!