[CameronBrooks11]

Adds 17 conformance cases for the questions #63 settled around U+0000, and for CommonMark 0.31.2 §6.6's unquoted attribute value. They were written from SPEC.md at 5fda73e by an author who had not seen the implementation, and land byte for byte as written (sha256 of every file checked after copying). Two of them failed against the reference. That is #64, fixed here in its own commit.

Commits

1. test(conformance): add blind cases around U+0000 and control characters: the cases alone.
2. fix(reference): allow control characters in an unquoted attribute value: closes reference: unquoted attribute value refuses control characters CommonMark allows #64.
3. test(conformance): raise the case floor to 525: MIN_CASES 508 → 525.
4. docs(conformance): record the provenance of the U+0000 cases: the README provenance paragraph, as earlier sets have.

Every case, and its outcome against the reference

"Before" is the reference at 5fda73e, unchanged. "After" is with commit 2.

case	before	after
resolve/a-u0000-in-a-heading-is-in-no-slug	pass	pass
anchor/record/refused-for-a-raw-u0000-in-the-anchor-line	pass	pass
anchor/record/refused-for-an-escaped-u0000-in-leaf-text	pass	pass
anchor/record/resolves-with-a-raw-ufffd-in-the-anchor-line	pass	pass
anchor/suggestion/writes-back-over-a-u0000-inside-the-span	pass	pass
anchor/suggestion/writes-back-in-a-heading-over-a-u0000-inside-the-span	pass	pass
anchor/transclusion/refused-for-a-raw-u0000-in-its-pin-line	pass	pass
anchor/transclusion/resolves-with-a-raw-ufffd-in-its-pin-line	pass	pass
anchor/transclusion/resolves-a-destination-holding-a-u0000	pass	pass
refusal/link-reference-definition/accepts-a-bare-destination-holding-a-u0001	pass	pass
refusal/link-reference-definition/accepts-an-unclosed-angle-destination-holding-a-u0000	pass	pass
refusal/link-reference-definition/refuses-a-bare-destination-holding-a-u0000	pass	pass
refusal/link-reference-definition/refuses-an-angle-destination-holding-a-u0000	pass	pass
refusal/raw-html/accepts-a-double-quote-in-an-unquoted-attribute-value	pass	pass
refusal/raw-html/refuses-an-unquoted-attribute-value-holding-u0001	FAIL: expected refuse with 'raw-html', got accept	pass
refusal/raw-html/refuses-an-unquoted-attribute-value-holding-u001f	FAIL: expected refuse with 'raw-html', got accept	pass
refusal/raw-html/refuses-an-unquoted-attribute-value-of-plain-letters	pass	pass

The three transclusion cases name their target file p, U+FFFD, .md, and use that name in a {{blob:PATH}} placeholder. The author asked whether the runner takes a non-ASCII file name. It does: all three pass, so the runner needs no change.

Two cases not added

The set has 19 cases. Two of them use names that cases already on main have, with different contents:

• anchor/suggestion/writes-back-keeping-a-u0000-outside-the-span
• anchor/suggestion/writes-back-in-a-heading-keeping-a-u0000-outside-the-span

The ones on main came from #42's set. They use other fixtures, record ids and spans ([42, 51] on main against [9, 11] here). Copying the new cases byte for byte would overwrite them, and the tree's rule is add, do not edit. So these two are left out. Run unchanged from a separate directory under other names, both pass the reference with and without commit 2, so leaving them out hides no finding. No other case in the set depends on them. If they should land, they need new directory names, and the owner should decide that, since the name is part of a case.

The reference fix (#64)

reference/blockspec/document.py's unquoted attribute value was [^"'=<>\x00-\x20]+. §6.6: "An unquoted attribute value is a nonempty string of characters not including spaces, tabs, line endings, ", ', =, <, >, or `` ``." It is now [^ \t\n\r"'=<>]+`. A sweep of `<a b=x` + c + `y>` over U+0001 to U+0020, U+007F and U+FFFD gives `raw-html` for every c except U+000A and U+000D. Those two end the line, so the tag is incomplete there. A space or tab ends the value, and `y` is then a second attribute, so the tag is still complete. pandoc 3.1.11.1 (`-f commonmark`) reads ``, `` and `` as raw HTML blocks too.

Gates

Each run on its own, exit code checked directly:

just check    exit 0   All checks passed!
just test     exit 0   181 passed in 14.60s
just conform  exit 0   0 failure(s) across 525 case(s) in the fixture tree
just tree     exit 0   525 cases: 165 accept, 82 refuse, 278 anchor / 0 problem(s)
just mutants  exit 0   reader:   93 killed, 0 survived, 0 equivalent, 0 stale, 0 broken
                       resolver: 123 killed, 0 survived, 1 equivalent, 0 stale, 0 broken

just mutants at 5fda73e gives the same totals, so nothing survived before for these cases to kill. The heading write-back mutant (U+0000 unmapped in a heading's write-back) was already killed on main, by the existing heading case. The new cases add kills to mutants that were already dead: U+0000 unmapped in write-back goes from 2 cases to 4, U+0000 unmapped in a heading's write-back from 1 to 2, and broken pins ignored from 9 to 10.

Questions the author recorded, with no case

1. A pin whose path spells \u0000 matches no transclusion (§3). What the transclusion with no pin, and the pin with no transclusion, then resolve to is open (§12).
2. Write-back with a span at the edge of a heading's content. §11.7's list of raw characters that no content code point comes from does not name a heading's # runs or the spaces around them. The heading cases keep the span in the middle of the content.
3. The cases README does not say whether a non-ASCII fixture name or placeholder path is supported. The runner takes one (above).

Refs #62