I am very pleased we are making progress. If we can get this right, it 
is going to save me having to retype all the Hebrew! The text certainly 
looks much better but the problem is not as yet fully resolved. There 
remain circles in the text (see attachment). If I try to delete them, 
something is lost from the word, either a vowel or a dagesh. I have also 
taken a screen shot of the Emacs version. Please confirm this is the 
correct one I should be using.

On 2019/06/13 11:44, Eli Zaretskii wrote:
>> Cc: 36171@debbugs.gnu.org
>> From: Raoul Comninos <revrari@mweb.co.za>
>> Date: Thu, 13 Jun 2019 10:53:24 +0200
>>
>> I have attached a sample document with problematic Hebrew taken from a
>> few of the files.
> Ah, I see.  This document uses precomposed Hebrew characters from the
> Alphabetic Presentation Forms block, and we didn't include those in
> the regular expressions used for finding sequences we need to compose
> for display.  I've now fixed this in Git, you should be able to
> rebuild after syncing with Git, and then the problem should go away.
>
>> (xft xfthb x)
>>
>>> Emacs needs HarfBuzz the library, not the program.  What does the
>>> following produce inside the latest Emacs you built from Git:
>>>
>>>     M-: (frame-parameter nil 'font-backend) RET
> Yes, xfthb uses HarfBuzz to perform text shaping.
>
> Thanks.