gnupic: Re: [gnupic] is OO any use?
Subject:
Re: [gnupic] is OO any use?
From:
David McNab ####@####.####
Date:
19 Aug 2005 22:48:20 +0100
Message-Id: <4306539A.4010508@rebirthing.co.nz>
David P Harris wrote:
> David-
> Is this written up somewhere? Sounds interesting. Being able to choose
> between early and late binding is very useful, and let's you tune your
> application. David
It's fairly thoroughly documented in the original (slow but compact)
pic18forth - http://pic18forth.sourceforge.net/doc/oo.html
Features and syntax will be the same in the new fast-but-bloaty version.
Cheers
David
>
> David McNab wrote:
>
>> Thanks all for comments.
>>
>> As it happens, I've just migrated over the OO layer from my slow/compact
>> forth to the new fast/bloaty forth.
>>
>> The OO uses a vtable at the start of each object's data area, to 'tag'
>> what class the object is, and facilitate late binding.
>>
>> Method and attribute accesses get compiled to code which passes a 2-byte
>> hash of the method/attribute name, for faster comparison during the
>> search.
>>
>> Even with this optimisation, the cpu usage for late-binding method
>> lookup is weighing in at 218 cycles minimum. that's 43.6us for an 18f
>> running at 20MHz. This will grow further for methods further down the
>> vtable, and grow even more harshly for methods higher in the class'
>> ancestry. Eek.
>>
>> Fortunately, the OO also supports early binding, which cuts this down to
>> 2 cycles - the cost of a CALL. The syntax is uglier though, and needs
>> the user to know the object's class ancestry.
>>
>> Anyway, what the hell, it's done. I might find a use for it, or I may
>> not. But I'd better warn users in the doco that accessing objects using
>> late binding cops a cruel speed penalty.
>>
>> Cheers
>> David
>>
>>
>> David Santo Orcero wrote:
>>
>>
>>> Hello, David!
>>>
>>>
>>>
>>>
>>>> This is not a troll, so please forgive me if it comes across as such.
>>>>
>>>
>>> I don think so. :-)
>>>
>>>
>>>
>>>
>>>> What I'm asking here is - is there any real practical value to be
>>>> gained
>>>>
>>>> from supporting OOP in a PIC runtime environment?
>>>
>>>
>>>
>>> Let's divide the question in two parts:
>>>
>>> OOP as a design metodology: so far you are confortable with OOP
>>> design, I don't feel any problem with using it. If it allows you to a
>>> better design, and a cleaner organization, you will have a code more
>>> easy to understand and maintain. And if you use assembler -and most
>>> of us use it-, you will need that kind of code. There are tricks well
>>> known to obtain good non-OOP code for a OOP design, so don't be
>>> afraid to mix OOP design with final assembler code.
>>>
>>> OOP programming languages: teorically, it is possible; but I don't
>>> know any OO language compiler for PICs. Anyway, I strongly doubt that
>>> it is usefull. There are some issues related with dynamic linking and
>>> function calling that are heavier in a OO languaje due to polymorfism
>>> and inherance that could make the call of a function too heavy for
>>> obtaining good performance for a PIC. More powerfull microcontrolers
>>> -and I think here in something like a picojava core and higher- would
>>> allow an OO languaje, but the energy needs will be higher, and the
>>> cost of the components and the development time of the hardware will
>>> also increase.
>>>
>>> Yours:
>>>
>>> David
>>>
>>> ---------------------------
>>> David Santo Orcero
>>> http://www.orcero.org/irbis
>>> ---------------------------
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: ####@####.####
>>> For additional commands, e-mail: ####@####.####
>>>
>>>
>>>
>>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: ####@####.####
>> For additional commands, e-mail: ####@####.####
>>
>>
>>
>>
>>
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: ####@####.####
> For additional commands, e-mail: ####@####.####
>
>
>
