U++ forum: Welcome to the forum

Re: substring find [message #861 is a reply to message #859]

Mon, 06 February 2006 11:28

mirek
Messages: 14267
Registered: November 2005

Ultimate Member

No, small string optimization means that for small strings, you keep data inside String object, something like:

String {
struct Large {
const char *ptr;
....
}
union {
char data[16]
Large large_string;
}
};

If you think about the issue, unshared reference counted string (and per my research, most reference counted strings are unshared) has 16 bytes overhead (4 bytes for pointer in String, 4 bytes reference count, 4 bytes length of string, 4 bytes allocation length). Then 70% of Strings has len < 15. Means, in 70% cases SSO will store the String "for free" when compared to current implementation. In remaining 30%, SSO will just use those 16 bytes to store pointer/length/alloc previously stored in shared string (storing there just reference count). Also, you will avoid alloc/free, interlocked increment/decrement etc...

Mirek

Report message to a moderator