Rather than discussing whether MathML is a failed standard, web or otherwise, I recommend
we discuss specific, constructive topics. I suggest the discussion be in the context of
MathML where appropriate, not because I want to defend MathML but because it is an
existing standard. It is a place to start. If the solutions we reach replace MathML all or
in part, so be it. Let's not start by throwing it out but by addressing its problems.
We can certainly create a new standard if MathML can't be fixed. Finally, if this is
the wrong venue for this topic or any other, please suggest a better one. If there are
other parties that need to know about the discussion, please let them know.
Assuming others agree, let’s start with perhaps an important issue. Should Presentation
MathML dictate a specific rendering or leave formatting choices up to the renderer. Peter
says, "I have the impression people generally expect consistent rendering across
browsers. But anecdotal evidence is, well, anecdotal." I would agree with this
statement. People do expect this. I believe they get that expectation from TeX but it does
make sense. Why would a user want a different rendering in a different browser?
The reason I said "no" to this before was because the MathML spec leaves a lot
of rendering decisions up to the implementation. Someone reading the MathML spec should
NOT expect all renderings to be the same. In fact, the spec doesn't specify the
rendering at the required level of detail. Doing so would be difficult. TeX doesn't
specify its rendering in detail either except via the code itself. In other words, the
only proper rendering of TeX is that done by TeX itself.
We could create a MathML 4 in which the graphical rendering is specified in writing and in
detail. Implementations would be constrained much more than by the current spec. Another
way to achieve this goal is to create a reference implementation. This would be the TeX
way, or close to it.
We could even map MathML onto TeX somehow and then defer to TeX's rendering. The
MathML spec would be annotated by TeX templates (perhaps macros) that serve to define the
rendering. The reference implementation would consist of a MathML-to-TeX convertor and the
TeX engine itself. Implementations that intend to abide by the MathML 4 spec could use the
reference implementation or roll their own.
When I say rendering above, I only mean graphical rendering. When we talk about audio or
braille rendering, things are much less clear. The state of the art in MathML-to-speech
has certainly not reached a point where everyone can agree. Besides, there is personal
taste of the reader and multiple languages to consider.
Ok, I'll stop there and take a breath.
Paul