begin  quoting boblq as of Thu, Feb 03, 2005 at 01:09:04AM -0800:
> On Thursday 03 February 2005 12:54 am, Stewart Stremler wrote:
> > begin  quoting Andrew P. Lentvorski, Jr. as of Wed, Feb 02, 2005 at 
> > 10:36:05PM -0800:
> > > On Feb 2, 2005, at 8:48 PM, Stewart Stremler wrote:
[snip]
> > Naturally.  It's an obvious thing to optimize on.  And if the optimization
> > is relatively transparent and doesn't change the syntax of the language,
> > it's conceptually irrelevent ... everything behaves /as if/ it were all
> > cons-cells.
> 
> Not really. The guys who knew what was going on were often using
> that implementation detail to dig deeper into the underlying data
> structure. This leads to (as you might expect) a huge number of
> difficulties when the implementation is changed. 

Indeed.

Optimization makes code faster as the expense of making it brittle.

> This problem was a killer for a lot of FORTH code that 
> was from a modern perspective too integrated, i.e. the 
> implementation was seamlessly optimized. 

It's good to have boundaries beyond which you pretend not to know what
is going on.  A few "shear planes"* in the code is a good thing...

> But it made a lot of sense to those who understood it 
> at the time. Not really much different from the early 
> LISP code.

Cleverness can sometimes be counter-productive.

* I hope it's obvious what I mean by this.

-Stewart "Cleverness is not always better than maintainability" Stremler
-- 

KPLUG-List mailing list
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list

Reply via email to