Project

General

Profile

Misc #8962

Updated by nobu (Nobuyoshi Nakada) over 10 years ago

=begin 

 Is there any chance to reap the benefit of new Generational GC merits for 
 C-extension library authors? 

 == Background 

 First of all: RGenGC is great.    Ko1 showed that it could make significant 
 performance improvement at RubyKaigi2013. (especially P82) 
 ((<RubyKaigi2013-ko1.pdf|URL:http://www.atdot.net/~ko1/activities/RubyKaigi2013-ko1.pdf>)) 

 http://www.atdot.net/~ko1/activities/RubyKaigi2013-ko1.pdf  

 I guess the improvement is triggered by marking most (or all?) of built-in 
 classes as WB-protected struct to work with Generational GC. 

 == Motivation 

 As an a extension library author, I want to try to get the performance merit. 
 But there is no document or guide to enable it. 

 The PDF says "Inserting WBs step by step, and increase performance gradually", 
 and I believe it is the greatest point of RGenGC, but there is no guide to 
 proceed with the steps for now. It's sad. 

 == Subject 

 Could you write about it as a document, ko1 or anyone? 
 I guess it's good to be written at ((%README.EXT*%)). 
 README.EXT*.  
 (The case of ((%iseq.c%)) iseq.c may be used as an example.) 

 I'm glad to see the documents are written before Ruby 2.1 release. 

 == Restriction 

 Sorry for the absence of my knowledge.    Because I'm not good at 
 RGenGC, I could't write the document by myself but could only request. 

 PS 
 I guess this issue depends on #3064 (sorry, in Japanse), the request of 
 documenting (({RTypedData})), RTypedData, because there is no interface to specify 
 (({FL_WB_PROTECTED})) FL_WB_PROTECTED flag with traditional (({RData})). RData. 

 =end 

Back