Re: ixampl goes Unicode

On Wednesday 17 August 2022 23:54:56 (+02:00), Adam Retter wrote:


As it's streamable to validate you need a maximum buffer of byte[3], 
although I suspect you can for tye purposes of pure validation get away 
with a buffer of byte[1]. the validation itself is a simple bit pattern 
match and is trivial - e.g. GitHub.com/digitalpreservation/utf8-validator
That was exactly the problem: I'm using a programming language that doesn't 
have the concept of 'bit pattern'.
That's why I used my solution (and needed to discover it first, because all 
solutions talk in terms of bit patterns).


Best wishes,


Steven




On Tue, 16 Aug 2022, 08:20 Steven Pemberton, <steven.pemberton@cwi.nl> 
wrote:

  > If this hasn't been written up anywhere, it would be great as a very
short paper. :)
Any suggestion of where it would be suitable to submit to?

BTW, one extra nice tidbit: the bytes of a Unicode character are just a
base64-encoding of its codepoint. Each byte is one digit, and the value of
each digit is just its position in its range. So for instance #C1 is in the
range [#C0-#DF], and so has value 1.

  > Do you have a separate check for the illegal characters?

I don't, and I wouldn't know what to do with one either. I don't have a
check for a string beginning with a continuation byte either. I just assume
the stream is already good.

  > P.S. I'd love to see the Unicode classes

https://www.fileformat.info/info/unicode/category/index.htm

  > (and generally, the entire ABC implementation) when you have a chance.

Actually this whole exercise resulted from you asking for the sources, and
me tidying them up for release... Give me a day or so.

Best wishes,

Steven

  >
  >
  > On 8/15/22 5:39 PM, Steven Pemberton wrote:
  > > It is now live.
  > > I haven't yet updated the Unicode character classes though.
  > >
  > > Steven
  > >
  > > On Monday 15 August 2022 18:07:44 (+02:00), Steven Pemberton wrote:
  > >
  > > > A weird thing happened yesterday, quite unexpected (to me): I got
ixampl working with Unicode characters.
  > > > I'd never thought it possible, because ABC has only 8 bit
characters, and they are atomic: no bit operations, no conversion
functions, and UTF-8 is always described in terms of bit patterns.
  > > >
  > > > And then yesterday, I had a brainwave. There are only 256 bytes. 
128
of them are ASCII, and they just represent themselves (that's the reason
UTF-8 exists).
  > > >
  > > > Of the other non-ASCII characters, they all play a single role in
any UTF-8 string:
  > > >
  > > > [#C0-#DF] are leading bytes of a 2 byte character
  > > > [#E0-#EF] are leading bytes of a 3 byte character
  > > > [#F0-#F7] are leading bytes of a 4 byte character.
  > > > [#80-#BF] are continuation bytes of the multibyte characters,
  > > > and [#F8-#FF] are illegal.
  > > >
  > > > What this meant was that I could make a 256 long byte array, start,
where each entry describes that role: 0 for continuations, 1 for ASCII, 2
for leading byte of 2 byte characters and so on for 3 and 4.
  > > >
  > > > In ABC the | operator delivers the first n bytes of a string
  > > >
  > > > "dishonest" | 4 = "dish"
  > > >
  > > > so to extract the next Unicode character from a string s, all I 
have
to do is
  > > >
  > > > s|start[s|1]
  > > >
  > > > Bingo!
  > > >
  > > > The new ixml is not online yet: just running the regression tests.
  > > >
  > > > Steven
  > > >
  > > >
  > >
  >
  >