Update character widths from Unifont 8.0.01

[jiahao]

Ref #50

[stevengj]

The Ruby oddness strikes again. A difference in the utf8proc_stage2table, it seems like, but looking at the Ruby script I'm not sure why this would be non-deterministic.

@StefanKarpinski, you used to be a Ruby hacker, right? Any idea why the data_generator.rb script would be non-deterministic?

[stevengj]

Note that e.g. u+06de now has a charwidth of 2 with this patch. This seems odd to me: on my terminal, ۞ seems to have a charwidth of 1.

[jiahao]

Unifont 8.0.01 does define U+06DE to have fullwidth (1024 units); 7.0.06 gave it a negative RBearing of -1024 units so that it had width 0.

[StefanKarpinski]

Hashes are unordered in Ruby 1.8.x (but not, I believe in 1.9.x). Perhaps that could affect things?

[stevengj]

@StefanKarpinski, I figured that the order of a hash would be undefined, but I'm not seeing where the script would depend on this. What am I missing?

[stevengj]

Has anyone been able to reproduce this failure locally? I tried using the same Ruby version as on Travis (ruby 1.9.3p551, from #48), but it still passes.

[stevengj]

My inclination is to merge this and to open a separate issue for the data-generation Travis test, since that was a problem prior to this patch.

[ScottPJones]

How difficult is this Ruby code to understand? We've got a much better language right here!

[stevengj]

I don't have the time to rewrite it from scratch right now, do you? Also, it is frustrating to rewrite otherwise-working code rather than fix something that is probably a one-line fix, if only I could reproduce the problem.

[ScottPJones]

It is really that complicated? I'm just asking, not trying to pressure you (or anybody) into rewriting it.
I was just wondering if the effort of finding that bug would be more than getting something in Julia that would save time in the future (and not require a person with Ruby expertise).

[StefanKarpinski]

The Ruby code is pretty straightforward, but it's also unclear why it would be nondeterministic. I don't think completely rewriting it is the answer.

[stevengj]

Merged since the Ruby problem on Travis is not a regression in this patch.