patchUnicode: update of combining code points

Unicode 6.3 defines the following code as combining or accents, git_wcwidth() should return 0.

Earlier unicode standards had defined these code point as "reserved":

358 COMBINING DOT ABOVE RIGHT 359 COMBINING ASTERISK BELOW 35A COMBINING DOUBLE RING BELOW 35B COMBINING ZIGZAG ABOVE 35C COMBINING DOUBLE BREVE BELOW 487 COMBINING CYRILLIC POKRYTIE 5A2 HEBREW ACCENT ATNAH HAFUKH, 5BA HEBREW POINT HOLAM HASER FOR VAV 5C5 HEBREW MARK LOWER DOT 5C7 HEBREW POINT QAMATS QATAN 604 ARABIC SIGN SAMVAT 616 ARABIC SMALL HIGH LIGATURE ALEF WITH LAM WITH YEH 617 ARABIC SMALL HIGH ZAIN 618 ARABIC SMALL FATHA 619 ARABIC SMALL DAMMA 61A ARABIC SMALL KASRA 659 ARABIC ZWARAKAY 65A ARABIC VOWEL SIGN SMALL V ABOVE 65B ARABIC VOWEL SIGN INVERTED SMALL V ABOVE 65C ARABIC VOWEL SIGN DOT BELOW 65D ARABIC REVERSED DAMMA 65E ARABIC FATHA WITH TWO DOTS 65F ARABIC WAVY HAMZA BELOW

This commit touches only the range 300-6FF, there may be more to be updated.

Signed-off-by: Torsten Bögershausen <tboegi@web.de>
---
 utf8.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

Torsten Bögershausen:

Is there a script that generates this code from the Unicode database files, or did you hand-update it?

On 15.04.14 21:10, Peter Krefting wrote:

Some of the code points which have "0 length on the display" are called "combining", others are called "vowels" or "accents". E.g. 5BF is not marked any of them, but if you look at the glyph, it should be combining (please correct me if that is wrong).

If I could have found a file which indicates for each code point, what it is, I could write a script.

So yes, it is updated by hand.

On 16/04/2014 07:48, Torsten Bögershausen wrote:

Indeed it is combining (more specifically it has General Category "Nonspacing_Mark" = "Mn").

The most complete and machine-readable data are in these files:

http://www.unicode.org/Public/UCD/latest/ucd/UnicodeData.txt http://www.unicode.org/Public/UCD/latest/ucd/EastAsianWidth.txt

The general categories can also be seen more legibly in:

http://www.unicode.org/Public/UCD/latest/ucd/extracted/DerivedGeneralCategory.txt

For docs, see:

http://www.unicode.org/reports/tr44/ http://www.unicode.org/reports/tr11/ http://www.unicode.org/ucd/

The existing utf8.c comments describe the attributes being selected from the tables (general categories "Cf","Mn","Me", East Asian Width "W", "F"). And they suggest that the combining character table was originally auto-generated from UnicodeData.txt with a "uniset" tool. Presumably this?

https://github.com/depp/uniset

The fullwidth-checking code looks like it was done by hand, although apparently uniset can process EastAsianWidth.txt.

Kevin

On 2014-04-16 12.51, Kevin Bracey wrote:

Excellent, thanks for the pointers. Running the script below shows that "0X00AD SOFT HYPHEN" should have zero length (and some others too). I wonder if that is really the case, and which one of the last 2 lines in the script is the right one.

What does this mean for us: "Cf Format a format control character"

#!/bin/sh

if ! test -f UnicodeData.txt; then
  wget http://www.unicode.org/Public/UCD/latest/ucd/UnicodeData.txt
fi &&
if ! test -f EastAsianWidth.txt; then
  wget http://www.unicode.org/Public/UCD/latest/ucd/EastAsianWidth.txt
fi
if ! test -f DerivedGeneralCategory.txt; then
  wget http://www.unicode.org/Public/UCD/latest/ucd/extracted/DerivedGeneralCategory.txt
fi &&
if ! test -d uniset; then
  git clone https://github.com/tboegi/uniset.git
fi &&
(
  cd uniset &&
  if ! test -x uniset; then 
    autoreconf -i &&
    ./configure --enable-warnings=-Werror CFLAGS='-O0 -ggdb'
  fi &&
  make
) &&
UNICODE_DIR=. ./uniset/uniset --32 cat:Me,Mn,Cf
#UNICODE_DIR=. ./uniset/uniset --32 cat:Me,Mn

On 16/04/2014 22:58, Torsten Bögershausen wrote:

Maybe dig back through the Git logs to check the original logic, but the comments suggest that "Cf" characters have been viewed as zero-width. That makes sense - they're usually markers indicating things like bidirectional text flow, so won't be taking space. (Although they may be causing even more extreme layout effects...)

Soft-hyphen is noted as an explicit exception to the rule in the utf8.c comments. As of Unicode 4.0, it's supposed to be a character indicating a point where a hyphen could be placed if a line-wrap occurs, and if that wrap happens, then it can actually take up 1 space, otherwise not. So its width could be either 0 or 1, depending. Or, quite likely, the terminal doesn't treat it specially, and it always just looks like a hyphen... Thus we err on the safe side and give it width 1.

See http://en.wikipedia.org/wiki/Soft_hyphen for background.

The comments suggest adding "-00AD +1160-11FF" to the uniset command line for that tweak and for composing Hangul. (The +200B tweak isn't necessary any more - Zero-Width Space U+200B became Cf officially in Unicode 4.0.1:

http://en.wikipedia.org/wiki/Zero-width_space http://www.unicode.org/review/resolved-pri.html#pri21 )

All of this is only really an approximation - a best-effort attempt to figure out the width of a string without any actual communication with the display device. So it'll never be perfect. The choice between double and single width in particular will often be unpredictable, unless you had deeper locale knowledge.

Actually, while doing this, I've realised that this was originally Markus Kuhn's implementation, and that is acknowledged at the top of the file:

http://www.cl.cam.ac.uk/~mgk25/ucs/wcwidth.c

Good, because he knows what he's doing.

Kevin

Torsten Bögershausen:

All combining characters has a non-zero combining class in http://www.unicode.org/Public/UNIDATA/UnicodeData.txt (fourth field, called Canonical_Combining_Class in http://www.unicode.org/reports/tr44/ ). For instance, the aforementioned U+05BF is defined as follows:

   05BF;HEBREW POINT RAFE;Mn;23;NSM;;;;;N;;;;;

The combining class is 23, so this is a combining character.

There is a difference between non-spacing combining marks ("Mn" in the third column (General_Category)) and others ("Mc" for spacing marks and "Me" for enclosing marks), so they might need specifial handling. Additionally, you have the "zero-width" characters, such as U+200B Zero Width Space. These have the "Cf" class, although it also contains visible characters IIRC.