347370 - normalize() combining text nodes can change MathML rendering

Status: RESOLVED WORKSFORME

(Core :: MathML, defect)

Description

[Jesse Ruderman]

This testcase demonstrates that you can have a fraction whose arguments (children) are just two adjacent text nodes.  If normalize() is called, the two text nodes are combined, and the fraction code decides the DOM is suddenly invalid.  This seems wrong to me, although I can't find something in the DOM spec specifically saying that languages must be designed so that normalize() will not affect rendering.

Is this a fundamental conflict between the MathML and DOM specs?  A bug in Gecko, where text nodes should not be accepted as arguments to <mfrac>?  I think it's the latter; section 2.3.5 of the MathML spec seems to be telling me that non-whitespace characters are not allowed outside of token elements.

Comment 1

[Jesse Ruderman]

Attachment: testcase

Comment 2

[Jesse Ruderman]

> A bug in Gecko, where text nodes should not be accepted as arguments to <mfrac>?

A more compatible fix would be to treat multiple adjacent text nodes (where at least one includes non-whitespace) as a single text node for the purpose of determining the argument-to-mfrac index.

Comment 3

[rbs]

>section 2.3.5 of the MathML spec seems to be telling me
>that non-whitespace characters are not allowed outside of token elements.

That's right. So the first markup in the testcase is invalid to begin with. It is only a coincidence that it is even rendered. Since Gecko is non-validating, we don't claim (nor attempt) to detect/report all possible invalid cases that breach the spec. Let's resolve this one therefore as INVALID.

Comment 4

[Jesse Ruderman]

This is now WFM.  Which makes me happier than INVALID :)