On 01.04.2008, at 11:44, Andreas Raab wrote:
> Bert Freudenberg wrote:
>> I was not implying that - I was referring to the many abandoned  
>> projects, not the active, flourishing ones. Again, Marcus would  
>> have to be more specific what specifically he had in mind. One  
>> example of "the perfect is the enemy of the good" are the series of  
>> JIT compiler implementations, none of which were finished, so  
>> Squeak still pretty much has the same interpreter it had ten years  
>> ago.
>
> Uhm, isn't *that* just a precise argument to the opposite?  
> Incremental improvements instead of half-finished research projects?

Err, I think you read me backwards, or I was imprecise. Yes, that is  
exactly the argument for incremental improvement. The Squeak VM is  
quite fast for a pure interpreter.

What we will never know is if the first Jitter had been incrementally  
improved rather than being abandoned like all its successors, it may  
have surpassed the current interpreter performance by far. The  
downside is that it would inherently be much more complex - the  
interpreter strikes a nice balance here.

Anyway, we seem to agree that incremental improvements is precisely  
what gets you usable near-term pink-plane results, and I think that  
was the gist of Marcus' message, too.

- Bert -

> Instead of rewriting the VM everytime and not quite finishing it  
> ever, the speed f the VM has doubled over the last ten years:
>
> 0 tinyBenchmarks; Squeak1.1:
>   '82740788 bytecodes/sec; 3818244 sends/sec'
>
> 0 tinyBenchmarks; Squeak3.8:
>  '191760299 bytecodes/sec; 5460228 sends/sec'
>
> Coincidentally, some of these improvements (like the at-cache) are  
> direct results of the more researchy efforts.
>
> Cheers,
>  - Andreas


>> The positive way of expressing CADT would be "burn the disk packs"  
>> and in particular in a research environment that is indeed the best  
>> you can do. It's simply a different motivation - do you build to  
>> have, or do you build to know? The latter does not require  
>> completion to be successful.
>> - Bert -
>>> Best regards,
>>> Nikolay Suslov
>>>
>>>
>>> On Mon, Mar 31, 2008 at 10:13 AM, Bert Freudenberg <bert at freudenbergs.de 
>>> > wrote:
>>>
>>>
>>> Well, Marcus would have to answer that himself. But there are a  
>>> lot on
>>> unfinished projects around Squeak, like the half dozen or so GUI
>>> builders, and you can certainly think of more. OTOH this problem is
>>> not specific to Squeak, it is just the reality of a lot of unpaid
>>> development work which people do for fun, a.k.a. CADT:
>>>
>>>       http://www.jwz.org/doc/cadt.html
>>>
>>> - Bert -
>>>
>>>
>>>
>>>
>
>