Re: [Sbcl-devel] SBCL server to compile code

[Tito L. <tit...@gm...>]

On Tue, Jul 02, 2013 at 01:25:38PM -0400, Martin Cracauer wrote:
> Tito Latini wrote on Tue, Jul 02, 2013 at 08:41:47AM +0200: 
> > Martin Cracauer wrote:
> > > As said, forking, compiling in the child and then loading the fasl in
> > > the parent, is an option.
> > 
> > ok, useful to know there aren't esoteric techniques to resolve this
> > problem. I'll weigh the sacrifice of the portability and the size of
> > the dynamic space in the choice of the strategy. To avoid the use of
> > the disk will be a priority.
> 
> Where does the disk come in as a problem? The garbage from compilation
> will stop your world via the global GC.  Are you running the compiler
> in a separate thread?

There is a rt-thread, a nrt-thread, a fast-nrt-thread and a thread for
the REPL. The communication with the rt-thread is lock-free. The
expansion of DEFSYNTH is a big block and the compiler triggers the gc
that stops the world. A possibility to reduce the effect of the gc is
to use a separate instance of SBCL to compile the monster, avoiding
the use of the disk to save/load the fasl. I deduced there is not a
common recipe for this task, therefore I can use what I want.

I appreciated all the replies, now it is the time to think.

Tito Latini