[gobo-eiffel-develop] Cost of Unicode strings for ASCII data

[Colin A. <col...@ho...>]

I came across very slow behaviour for gexslt compared with saxon when 
dealing with a 21MB string.

I decided to investigate whether UC_UTF8_STRING performance was a major 
contributor to this, or not.

Well, I havent been able to answer this definitively, but I do have an 
interesting measure.

In a test Eiffel program, I read the entire file into a STRING_8, and I then 
created a new string from that data. The elapsed time of the program depends 
upon the dynamic type of the resulting string. Here are my timings:

STRING_8: 7 seconds
STRING_32: 13 seconds
UC_UTF8_STRING: 14 seconds.

Note that the STRING_8 was created by a call:

create {STRING_8} l_text.make_from_string (l_text). Is this doing a copy? If 
it is, then the overhaed is very heavy.

_________________________________________________________________
Tell Hotmail about an email that changed your life!  
http://www.emailbritain.co.uk/