[tlbuild] xetex on sparc-solaris dumps core using opentype fonts

[Karl Berry]

    I cannot find any difference between 4.4 and 4.6 codes.

Clearly they are *supposed* to be equivalent.

    Bug in compiler? 

It's a natural supposition, but when it fails on every BigEndian
machine, regardless of compiler (original gcc, Apple gcc, Sun cc), it
seems improbable.

Maybe ICU never tested the 4.6 release on a BigEndian machine?
Also seems improbable, but who knows.

    Does (value << 8) & 0xff00 + (value >> 8) & 0xff work?

On powerpc-darwin, yes, it does.  (Well, I used & 0x00ff in the second
clause for symmetry, but that's just gotta be the same.)

That looks like a less intrusive change than using the 4.4 code.

I expect swapLong needs similar treatment.  Well, hopefully Jonathan
will have some input before committing anything.

Thanks,
k