On 16/03/2020 20:27, Henri Menke wrote:
>> Now, there is more data being loaded today than when I did that work, and
>> some of it is in LuaTeX so could be done Lua-only. It's also possible that
>> the Perl script was sub-optimal, or that as part of a general 'install'
>> function the time would not really show. However, XeTeX needs the data, so
>> one is still looking at having to explicitly pre-process in Lua. Moreover,
>> most of the time taken for format-building is not about reading Unicode
>> data. With LuaTeX, pre-loading expl3 does cut out a slight 'stall' when
>> loading everything for case-changing, but having a LuaTeX and a XeTeX path
>> separately is not attractive.
> 
> Is there any distribution that doesn't have LuaTeX in the default
> installation?  (Apart from exotic things like TeX Live infra-only)  Then
> it would be conceivable to just make LuaTeX a hard requirement and
> process the Unicode data on the fly instead of going via CTAN.

There's KerTeX ... but that's not got XeTeX either I think (and doesn't 
have the pdfTeX utility primitives). Certainly one could pre-digest at 
the format-building stage, or perhaps as a post-update action for the 
Unicode data files. However, there's still the fact that even reading a 
'pre-digested' file takes non-zero time.

>> As format-building is all about saving time for 'normal' runs, I'm not
>> seeing there is a massive need to speed up the process. I know there is one
>> engine in development that doesn't use format files, so that might be a
>> place to consider things, but I think we'd need a strong case to alter the
>> approach for XeTeX/LuaTeX (pdfTeX, ...).
> 
> Are you referring to JSBox?  I doubt that this will every be public.

Yes, JSBox is what I have in mind.

Joseph