Re: mathvariant vs. plane 1

Paul Libbrecht wrote:
> 
> Le 25-juin-09 à 18:31, Bruce Miller a écrit :
>> [...] So, for something like LaTeXML, I would have
>> to map some combinations of mathvariant + letter
>> to plane-1, but other random combinations to
>> places in plane-0 blocks.
>>
>> For search indexing, you'd likely need to do the
>> reverse, since people will expect a W to match
>> a W in any mathvariant or plane-1 block.
> 
> Could this be considered yet another requirement to be addressed In 
> MathML3?

Actually, I think section 7.5 pretty much addresses
everything it should.  It states the equivalence
between mathvariants & plane-1, but doesn't recommend
one form over the other --- and, at this point, I don't
think it should.

> It's probably too late but the fact that you quote the search-engine's 
> analyzers as yet another place where it would be impacted really makes 
> me feel it would be worth to strengthen that paragraph.

Perhaps, but it's almost a red-herring in this context:
similar issues arise with, eg., text accents even without
bringing math into it. How you index, and how you parse
queries, depends on how you anticipate users formulating
a query: for what, in principle should be an accented vowel,
they may use unicode combining accents, combined forms,
ascii-art (or LaTeX) commands like \"o, or (most likely)
just drop the accent entirely. Then you've got
things like the German umlaut that adds an "e" when you
dumb-down.  The Math mapping is easy in comparison!

[Before anyone asks: I haven't attempted to solve the accent issue
yet, although likely some people have]

> Do you have that table in LaTeXML (the "random hidden tweaks"?) ?

Not yet. Currently it just derives an appropriate
mathvariant from the font info. I had imagined eventually
adding an option to map to plane-1 for completeness.

But the sketchy implementation of mathvariant got me
thinking of doing it sooner rather than later.
Researching the state of plane-1 implementation
showed that only Firefox supports all of plane-1.
That got me wondering whether I needed even finer-grain control:
to select specific mathvariants to map or not!

OTOH, the fact that all agents support at least a
subset of plane-1 suggests that they could probably
do it all. Likewise, the fact that firefox now has access to
the needed fonts suggests that they could implement
the extra mathvariants.  In both cases, of course,
provided the interest & programming manpower were available.

So that's what led to the question:
Whether it is Sane to use a mixed mathvariant & plane-1
approach; whether folks recommend sticking to
one or the other set? (which?).
Or should I just Shut up & Wait...

-- 
bruce.miller@nist.gov
http://math.nist.gov/~BMiller/

Received on Friday, 26 June 2009 14:25:26 UTC