Re: [Texmacs-dev] Bigloo?

[Joris van der Hoeven]

> But if you want a good GC, take the one of OCaml, it is under LGPL. The
> only big issue is that this GC needs integer to be stored with bit 0 set
> to 1 (to avoid to mistake them for pointers). And of course, I doubt it
> could work with languages like C & C++.

Yes, I just read it. I think that the idea of two-generational GC's is
very useful for computer algebra (I was actually thinking about such
a thing for Mathemagix). Notice that, when the compiler is really
intelligent, it will *know* which routines only need a copy & compactify
phase (no need to keep track of possible major->minor pointers).

> The problem with TeXmacs is the problem with languages like C/C++ that
> were not designed for the ground-up to interact with a GC. This is not
> the case for Scheme and I suppose the designer of the compiler has
> incorporated relevant assumptions in the compiler to make the GC work
> well.

We should ask Boehm if his GC has been designed to make this possible
(without what Bigloo does not do that). I know that Guile and most
other Schemes come with their own mark & sweep GC.

My major concern for TeXmacs is what we should do in order to
implement a better GC.
  1) Does this require a modification in all our pointer handling,
     or can we apply some tricks?
  2) Can we hope for a compactifying GC (that would be nice,
     because it would make memory consumption proportional to
     the total size of the currently opened documents + cache).
  3) We definitely need at least some debugging tools in order to
     track memory leaks down.