[tex-live] XeLaTeX Not enough memory (part 2)

[Pander]

Karl Berry wrote:
>     - How can one configure in TeX Live the order of inclusion of font 
>     types? E.g. firstly use typ1, secondly user ttf, ... etc.
> 
> There is no general mechanism for ordering search of font types across
> programs.  I don't know if XeTeX has some specific way to control it.  I
> don't recall ever seeing such a thing.  But since you are making a huge
> font catalogue it is not surprising to me that it takes a huge amount of
> memory.
> 
>     - Do I abuse fontspec in anyway for XeLaTeX to require so much memory or 
>     is fontspec not optimised in this regard?
> 
> I hope someone who knows about fontspec will answer.

Since the kerning tables especially take up much of the memory and I 
only use a certain font on a certain page, would it be possible to 
provide a command on the next page to free memory by manually forcing 
unloading of that font?

> 
>     - Could once again the default value of fonts_mem_size be doubled?
> 
> The default value of font_mem_size is 3,000,000 and I don't especially
> want to hugely increase it, since that memory is malloc-ed at startup as
> far as I can tell.

Hmmm, very 1970's.

> 
> However, the hard limit is defined by sup_font_mem_size in tex.ch.  I
> increased that from 4,000,000 to 147,483,647, so you can change
> font_mem_size in texmf.cnf yourself and remake your format.
> 
> I trust someone will correct me if I'm wrong -- it appears to me that
> font_mem_size is simply an integer, so 2^31-1 should be the actual
> maximum (on 32-bit architectures).  That would have been 2,147,483,647,
> but I just didn't want to go that far, so I lopped off the 2 :).
> 
>     Why is this limited anyway?
> 
> Because nearly everything in TeX & derivatives (except for LuaTeX) is
> allocated as a static limit.  That's how Knuth wrote it, and making them
> dynamically increasable is not simple.
> 
> Perhaps you should consider making your font catalogue using LuaTeX.  (At
> least, I think Taco has eradicated most of these hard limits by now?)
> 
> k