Re: [Tinycc-devel] Local procedures

[Antti-Juhani Kaijanaho]

On Wed, Sep 12, 2007 at 02:10:33PM +0200, Zdenek Pavlas wrote:
> Right.  It was just an example how to implement a code semantically 
> equivalent to nested functions.  Had the second argument to closure() been 
> not a stack reference but a literal or a malloc()ed struct, 'fun' would 
> become a real closure, a first-class object that could be return'ed, stored 
> somewhere, etc.

Still not right.  A real closure operator captures its (lexical)
environment in the closure that it creates.  Anything you do in C
without compiler magic will have to explicitly name the variables that
are captured.

You solution reminds me of the way closures are emulated in C++ as
function objects.  It's close, but not the real thing.

>> You *do* need compiler support for them.
>>   
> Syntax sugar, perhaps?

Unfortunately, you do need more compiler support than that.

BTW, the only limitation of nested functions related to real closures is
that you can't safely use a nested function closure after the function
that created it has returned.  There are still quite a number of
applications of it that don't need to do that (the canonical example is
the comparison function you give to qsort).

-- 
Antti-Juhani Kaijanaho, Jyväskylä
http://antti-juhani.kaijanaho.fi/newblog/
http://www.flickr.com/photos/antti-juhani/