[XeTeX] A LaTeX Unicode initialization desire/question/suggestion

[David Carlisle]

On Mon, 13 Jan 2020 at 03:41, Doug McKenna <doug at mathemaesthetics.com> wrote:

> >| Are there architectural reasons preventing you from having a
> >| format file, or is it simply that you hope to make loading quick
> >| enough that you do not need it?
>
> Yes, and yes.  JSBox does not depend on an internal array of integers (mem[] or whatever).  Doing so makes it essentially impossible to use a modern-day debugger to examine data structures.  Every data structure that is allocated is done so (indirectly) via malloc() or whatever the equivalent might be on some system.  This makes it harder to create a \dump format file, though not impossible.  But it wouldn't be (or need to be) compatible with anything in the official TeX world.  Regardless, my goal is to see how far one can get without needing format files.

You could probably look at what luatex is doing as it is possibly
closer to you here than xetex: the sources are in C and use dynamic
memory allocation replacing almost all of tex's fixed arrays, but they
do manage to dump formats.

> Also, see below.
>
> >| The pressure to load more into a
> >| format is likely to increase rather than decrease, people often
> >| routinely make custom formats preloading large packages like tikz or
> >| pstricks for example.
>
> True, but there is a fundamental difference between what I'm working toward, and what the TeX infrastructure does.  In the TeX world, every job is a single process.  Every time a TeX job is done, a process is launched, the job gets done, and the program ends.  It's the Unix/command-line way.  So the format has to be loaded (fast) on every job.  Makes perfect sense.



This seems really quite strange to me I can think of very few user
level programs that are compiled from source on every use, it is
placing a big extra burden on the programmer (the latex team in this
case) to write the program with all code paths optimised for run time
speed rather than being able to offload expensive testing to compile
time so it is only done once.

David