[core-dev] Re: [codepatch] PATRICIA

LimeWire doesn't strip the quotes, BearShare seems to strip them and  
for GnucDNA spammer it is the signal to return even more junk. :)

GGEP is a good idea. One phrase per query and all results OR'ed?  
Everything else screams for an automaton IMO.

Patricia is indeed cool, to cite a childish guy: "Roger loves  
Patricia". :) Although it is not necessarily required for phrase  
searching. It would even work with the current Trie if you split the  
Strings in a similar fashion as I'm doing it in the Keyword class, or  
rather switch Trie from String type keys to CharSequence to narrow  
the memory requirements down.

cu
  Roger


On May 12, 2005, at 4:28 AM, Sam Berlin wrote:

> Hmm -- do existing clients strip invalid characters from queries?   
> So if a
> search goes out like "LimeWire Rocks" and it includes the quotes,  
> would
> existing LimeWires match this against entries with 'LimeWire' and  
> 'Rocks'
> (without the quotes)?
>
> If so, it shouldn't be a problem to just start sending searches  
> that include
> quotes and let older servents send bad results (which we can filter  
> on the
> client side).
>
> If not, I don't think adding more entries into QRP is the way to  
> go, because
> the tables already are too populated.  Perhaps a GGEP field in the  
> search
> that includes the index of the begin & end quotes (so that a search  
> can be:
> "LimeWire Rocks" p2p, and find all results that have 'p2p AND  
> "LimeWire
> Rocks"').  The plain-text query would be 'LimeWire Rocks p2p', and  
> the GGEP
> field would be something like '0, 15', meaning that there's a begin  
> quote at
> 0th character & an end quote at the 15th character.
>
> Aside from this QRP problem (which is a big problem, and is part of  
> the
> reason that searching for phrases is difficult), your patch is  
> awesome.
> Super awesome.
>
> Thanks,
>  Sam
>
>
>> -----Original Message-----
>> From: Roger Kapsi [mailto:[email protected]]
>> Sent: Wednesday, May 11, 2005 6:25 PM
>> To: Sam Berlin
>> Cc: [email protected]
>> Subject: Re: [codepatch] PATRICIA
>>
>> ... or I forgot the QRP! :)
>>
>> OK, I have two ideas.
>>
>> #1 Ultrapeers check if a query starts and ends with quotes
>> and if so they ignore them. Simple but doesn't conform to the
>> QRP spec.
>>
>> #2 Additionally to the pure keywords we could add all
>> permutations with quotes to the QRP. Not necessarily all
>> permutations, for files for example only the file name and
>> only certain combinations.
>>
>> some file.txt
>>
>> some
>> file
>> txt
>> "some
>> some"
>> "file
>> file"
>> "txt
>> txt"
>>
>> The combinations "some", "file" and "txt" make no sense IMO.
>>
>> Opinions, comments!?
>>
>> cu
>>   Roger
>>
>>
>> On May 11, 2005, at 10:17 PM, Sam Berlin wrote:
>>
>>
>>> This looks fantastic.  Will the searching for quoted
>>>
>> phrases work with
>>
>>> QRP already (and just be slightly more inefficient because
>>>
>> there'll be
>>
>>> more false positives) or...?
>>>
>>> Thanks,
>>>  Sam
>>>
>>> On May 11, 2005, at 3:10 PM, Roger Kapsi wrote:
>>>
>>>
>>>
>>>> Hi,
>>>>
>>>> this patch switches FileManager from Trie to PatriciaMap, adds
>>>> support for queries with quoted phrases and reduces the number of
>>>> IntSet instances.
>>>>
>>>> cu
>>>>  Roger
>>>>
>>>> <Archive.zip>_______________________________________________
>>>> codepatch mailing list
>>>> [email protected]
>>>> http://www.limewire.org/mailman/listinfo/codepatch
>>>>
>>>>
>>>
>>>
>>>
>>
>>
>
>

_______________________________________________
core-dev mailing list
[email protected]
http://www.limewire.org/mailman/listinfo/core-dev