<cough>

From: "Dimiter" <[EMAIL PROTECTED]>


> Ahem, ahem... :)
> 
> This is the most funny idea i've heard from a long time ago, it reminds
> me that when I was at age 12 I tought tat I cah compress sompressed file
> and then compress it again and again until I fit all my games on a
> single flopy disk :) 

Ah those were the days! You belittle yourself to right off these 
childhood imaginings! Perhaps if a floppy disk had fallen on your head
while you were taking a bath, you would have had the appropriate 
epiphany. Of course you are constrained in your example by singularity 
of media (only floppy disks), but a plethora of options abound in our 
case. See below...

> The compression is achieved because low information content (or highly
> correlated signal if you prefer) the loseless JPEG compresses
> photographic images (which are highly corelated data) at 10% in the best
> case. Apart from solving the problem how would you split the code in 8x8
> pixels (as prescribed in the JPEG standard) the transformed signal will
> have approximately the same size.

I use JPeg and fractal compression merely as stepping stones to provoke
a more lateral approach. Consider the elements that we are storing in 
memory (at the simplest level we have code and data segments) We simply
identify the data elements as having either lossy or lossless storage
requirements, and handle them appropriately. 

In either case depending on the frequency of access requirements, data 
segment elements could sometimes be stored to disk rather than memory.
(Why waste memory on what are predominately static entities anyway?)

As far as compressing code using lossy methods goes, consider using
CRC back checking to derive the original code from the lossy 
decompressed version. In other words use lossy compresion on the code 
and include well-engineered CRCs (or similar code cracking devices) to 
act as hints for resolving the losses encountered during decompression. 
i.e. the combination of lossy code with CRC hints are applied to an 
algorithm that iteratively adjusts the code until the original CRC is 
attained, and voila!, lossless code.
 
> BTW both the fractal and JPEG compressions are used in their lossy
> variants which makes them inappropriate for code. The memory compresion
> is done mostly with loseles statistical algorithms like the good old
> Huffman. It gives you maximum ratio of 5. if you have 512MB dedicated
> for IDEA (in fact if you had it you wouldn't have the trouble) it means
> 1.5G "waiting buffer". 

As stated before 

  alg(lossy(code) + hints + crc) = lossless(code)

This approach, I expect, goes beyond the scope of this list, but does 
serve to illustrate how we seem to spend a lot of time trying to squeeze 
the contents of a pear shaped swimming pool into a leaky red bucket, 
simply because we don't have a tropical beach at our disposal.

> Assuming that on my 256MB heap the garbage
> collections occur at least every hour, it would mean that the 1.5GB
> would sufice for approx 6 hours... so you have to go to the beach really
> often :)

This is not a bad thing, and waterproof laptops that can be mounted
on bodyboards will no doubt soon be more popular than portable mp4 
palm players and their ilk.

They screwed up the material for my mantle. Its looks more like sackcloth 
than anything I've seen at the royal drapiers. And for some reason they 
have shaven my head. Oh these Spanish and there inauguration rituals!

WPE
Madrid Februias 33.3 LP MONO

PS. What with all our noses being on the moon, is it any wonder nobody
wants to wake up and smell the coffee.

> 
> =QUOTED=============================================
> 
> Ahem...
> 
> If you look at the source of Runtime.gc() (line 488) you'll see that it
> can be overridden. So, all that needs doing is to implement JPeg or
> Fractal decompression on the memory stack. In javart.dll (or its 
> Linux equivalent - I think its called javalnx.so) the octal dump is
> 
>   :
> 017 AF02 00001100100100010001
> 018 AD02 0000100100701000!111
>   :
> 
> which is the generic machine code for BITSET RAMPUSH
> 
> So as long as memory is stored compressed (because it is just 
> zeroes and one's remember) I see no problem deferring garbage 
> collection until at least Tuesday.
> 
> =END=QUOTED=============================================


_______________________________________________
Eap-features mailing list
[EMAIL PROTECTED]
http://www.intellij.com/mailman/listinfo/eap-features

Reply via email to