Re: Question on compiling to bytecode

[Per Bothner <per at bothner dot com>]

Daniel Bonniot wrote:
> Doesn't Kawa also have an interpreter that duplicates the compiler 
> logic? That is, each construct in gnu.expr has both a compile and an 
> eval method, with (hopefully) identical semantics?
No.  Each construct has a compile method, but only the simple
constructs (variable reference, assignment, constant, function
call) have a non-exception-raising eval method.  If any of the
more complex expressions (inluding let-binding and lambda
expressions) are seen, that forces compilation to happen.

The idea is that simple expressions are probably most common
in interactive use, and it is easy to interpret them, so we
might as well, to avoid the (modest) overhead of compilation.
But more complex expressions, including any that create
lexical bindings, would be more complicated to interpret,
and would require duplicate machinary to interpret it (e.g.
setting up the data structures for lexical bindings), so
it is simpler to just invoke the compiler in that case,
given that the compiler overhead *is* fairly modest.  It
also helps ensure that behavior is consistent.
--
	--Per Bothner
per@bothner.com   http://www.bothner.com/per/