[gambit-list] redefining symbols with 'load'?

James Long longster at gmail.com
Sun Sep 2 02:07:57 EDT 2007


I want to make sure I understand how 'load' works in Gambit.  I have
basic build system using a slightly modified version Snow up and
running, and I'm making a few assertions that I want to make sure are
true.

In Gambit, a 'load' and 'include' are obviously quite different
because it's using machine-compiled code for compiled objects, not
bytecode.  I have heard (I have not tested this yet, although I will
soon) that it's quite expensive to call code across module boundaries,
which makes reasonable sense because it cannot simply use the
'goto'-style jump.  So, for any high-performance application, you want
to inline code when compiling if you don't  need a shared library.

Now, I have my system separated into a few shared libraries (.o1
files), but there are cases when I want to inline code that is common
between them.  For example, I have a utilities package which is called
quite often in all of the modules, and it makes sense to include this
file rather than load it (for performance).  However, doing so would
include this file in all modules, so when the modules are loaded
together, the same symbols are redefined.

If code is of the form

(load "lib")
(include "util.scm")

(call-utility-function 'with 'some 'params)

and util.scm is also included in the "lib" library, is it same to
assume that when the above code is compiled in Gambit that
call-utility-function with resolve to the local version with a
'goto'-style jump?  I don't see how it wouldn't since it has no idea
about lib's version of call-utility-function until run-time, unless it
does some sort of check at runtime?  I know you can't redefine symbols
in Gambit at compile-time, so I'm assuming it does a sort of static
binding.

I'll be the first to admit that this may be a major case of
pre-optimization.  so slap me if I shouldn't be worrying about this as
much.



More information about the Gambit-list mailing list