Name: bb33257			Date: 02/27/98


This bug is an addendum to bug #4113734.  You can close that bug as a duplicate of this one.

I've gotten back the following additional information from the Arabic people at IBM
in response to a note from me regarding the errors in the code-conversion tables for
Arabic:

* * *

>I looked over the code-conversion tables you sent me and tried to compare them
>with Javasoft's, and then I filed a bug report with them.  I'll let you know
>what happens with this.
>
>I didn't see many discrepancies.  In the ByteToChar conversion tables, =
>there
>were four to six mistakes per code page.  They were as follows:
>
>In ByteToCharCp420.java:
>  0x45 should map to \u200c
>  0x77 should map to \ufeb1
>  0x80 should map to \uf8f5

correction, 0x80 should map to \ufeb5

>  0x8b should map to \ufeb9
>  0x8d should map to \ufebd
>In ByteToCharCp1256.java:
>  0x80 should map to \u0080
>  0x8a should map to \u008a
>  0x8f should map to \u008f
>  0x98 should map to \u0098
>  0x9a should map to \u009a
>  0x9f should map to \u009f
>In ByteToCharCp864.java
>  0x9f should map to \u200c
>  0xd7 should map to \ufec3
>  0xd8 should map to \ufec7
>  0xf1 should map to \ufe7c

in 864 there are some more differences:
 0x1a should map to \u001c
 0x1c should map to \u007f
 0x7f should map to \u001a
 0x25 should map to \u0025

>And code page 1089 doesn't exist at all.

code page 1089 is the IBM number for ISO8859-6. I have checked the
differences and they are as follows:
 0x30 should map to \u0030
 0x31 should map to \u0031
 0x32 should map to \u0032
 0x33 should map to \u0033
 0x34 should map to \u0034
 0x35 should map to \u0035
 0x36 should map to \u0036
 0x37 should map to \u0037
 0x38 should map to \u0038
 0x39 should map to \u0039

>I suspect there are other things which also need to change in the CharT=
>oByte
>conversion tables, but I haven't looked into this.

I will try to look into them, however, the format of those tables are bit
complex.

>
>If you know of anything I've missed or gotten wrong here, I'd like to k=
>now
>about it, and I'd appreciate more background as to why we're suggesting=
> the
>changes we're suggesting (are the old tables just plain wrong in these =
>cases,
>or is this more of a judgment call?), but basically I just wanted to le=
>t you
>know I'd filed the bug and I'll let you know what happens with it.

There are two issues related to why there are differences in the mappings:
1- code pages 420 and 864 are basically font pages and coming from old
   implementations. The characters is question (except for the control
   characters in 864) are displayed in two cells vs. one cell in other
   code pages (ISO8859-6, 1256, unicode), the two cells are divided into,
   what we call 3/4 of the character which identifies that character and
   the other 1/4 is the tail which is common to all of them.
   So in IBM tables we map the 3/4 characters to the full shape in unicode.

2- in other cases like code 0x25 and 0x30-0x39, IBM tables map those
   characters to the US ASCII part of unicode for proper processing
   by BIDI unaware applications.

* * *

If I get back additional information regarding the inverse conversions from
Unicode back to SBCS, I'll pass it along in the form of another bug report.
======================================================================