CDATA case-insensitivity question

[numist]

cmark's inline raw HTML and HTML block matchers accept the literal <![CDATA[ case-insensitively, treating <![cdata[, <![CDATa[, <![cDaTa[ etc. as valid CDATA opens. CommonMark 0.31.2 §6.6 specifies the exact string <![CDATA[ (case-sensitive), and that wording inherits the underlying XML 1.0 (Fifth Edition) production [19] CDStart ::= '<![CDATA[', which is case-sensitive (XML is case-sensitive throughout).

cmark's matcher has matched case-insensitively since the initial commit, but the commit history doesn't show any deliberation about CDATA specifically; is this leniency intended (in which case the spec needs amending) or incidental (in which case the matcher should tighten)?

[jgm]

I can't say I recall the reasoning or research that led to it.

The spec actually doesn't say anything about case-sensitivity, as you note.

So perhaps we should tighten this. Am I missing something, anyone?

[nwellnhof]

The HTML5 spec only recognizes <![CDATA[ in upper case. (But CDATA sections are only relevant in "foreign content" (SVG or MathML), otherwise they're a parse error. So there's no good reason to treat them as start condition at all.)

So perhaps we should tighten this.

Yes, I think we should.

[jgm]

I think we should keep it as a start condition; conventionally, markdown support for raw HTML isn't limited to HTML5 and includes e.g. xhtml.

[nwellnhof]

markdown support for raw HTML isn't limited to HTML5

I'd say it is. But allowing CDATA as start condition doesn't hurt, so there's no reason to change the spec.

[jgm]

I'd say it is.

Well, the original Markdown description by John Gruber dates from 2004, before HTML5 was a thing. And interesting, the page serving it has an xhtml doctype!

[jgm]

I looked into this. We invoke re2c with:

	re2c -W -Werror --case-insensitive -b -i --no-generation-date \

which makes all of the scanners case-insensitive. So, fixing this rough edge for CDATA is not really trivial. We'd have to remove the --case-inensitive flag and change the regexes for regular HTML tags to be explicitly case-insensitive.

[nwellnhof]

From the re2c documentation:

"foo"

Case-sensitive string literal.

'foo'

Case-insensitive string literal.

So this wasn't hard to fix.