[Cuis-dev] Character ClassificationTable

Juan Vuletich JuanVuletich at zoho.com
Tue Nov 14 07:44:53 CST 2017


On 09/11/2017 07:13 p.m., Phil B via Cuis-dev wrote:
> I think this is something we got from the Spur migration and while it 
> makes sense for implementations of 16-bit characters and greater where 
> the space savings is considerable, it's a needless performance hit for 
> Cuis with 8-bit chars.  Essentially, to consolidate to a single lookup 
> table, calls to #isLetter etc and case conversion is 5-10% slower for 
> what is probably a 10-20KB total savings in the image.  (I.e. 
> essentially every lookup became a lookup and decode which involves a 
> bitAnd: and primitive call in the case of #isLetter). This adds up 
> very quickly because the methods using the lookup table are heavily 
> used in the inner loop of virtually any parser including the built-in 
> Smalltalk, number and date parsers for example.  Any objections to 
> going (back) to multiple lookup tables for the improved performance?
>

I'm Ok with this. I think you're in a better position than I to judge :)

Thanks,

-- 
Juan Vuletich
www.cuis-smalltalk.org
https://github.com/Cuis-Smalltalk/Cuis-Smalltalk-Dev
@JuanVuletich





More information about the Cuis-dev mailing list