Re: guile-vm 0.4

[Rob Browning]

Michael Livshin <address@hidden> writes:

> might as well not invent a new API and define it like this (CL-style):
> 
> (compile-file "foo.scm") -> compiles foo.scm, writing foo.go
> (load "foo") -> loads foo.go if that exists and is not older than
>                 foo.scm, else loads foo.scm and maybe gives a warning.
> (require 'foo) -> ditto.

This seems fine with me, though it doesn't address the case where you
want to generate and run byte-compiled/jit-compiled code, but don't
want to write any files to disk (i.e. you only keep .scm files on
disk).

In that case, a friend here suggested a solution (one I may have
proposed a day or so ago, but I forget -- I've been kinda sick this
week).  Perhaps an optional argument to compile/load, saying how to
handle recursive loads:

  (compile/load "foo" . compile-sub-files?)

Presuming the byte and jit compilers don't *need* to leave .go files
behind in order to function, this would allow you to compile/load from
read-only filesystems, tiny filesystems, etc.  I'm also thinking of
architectures like handhelds here where memory may be more plentiful
than persistent storage.

-- 
Rob Browning <address@hidden> PGP=E80E0D04F521A094 532B97F5D64E3930