This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: sso tradeoffs
- From: Benjamin Kosnik <bkoz at redhat dot com>
- To: Howard Hinnant <hhinnant at apple dot com>
- Cc: libstdc++ at gcc dot gnu dot org
- Date: Thu, 3 Nov 2005 11:02:16 -0600
- Subject: Re: sso tradeoffs
- References: <69BD9B30-F3BE-4128-997D-67B67C256FD7@apple.com> <4369DD61.7000606@suse.de> <9DC948BF-4BA1-406E-89FE-5FAE173DDA0A@apple.com>
> Thanks for your thoughtful comments Paolo. Getting at least two
> brains (and hopefully a few more) thinking about these tradeoffs was
> exactly what I was hoping for. I'm not sure what time frame is going
> to work for me to submit an alternative sso, but it is on my to-do
> list. Perhaps with two alternative sso's that we can easily switch
> between, and/or tweak tuning parameters, we can get some of that real-
> world data you speak of.
It's my hope that versa string can be taught to use different data
encapsulation techniques, and with a bit of effort, support multiple
types. I haven't looked at this code in detail yet, however.
I would prefer that approach, if possible, to a bunch of different
extension string classes.
One thing that might be useful would be to come up with docs that
describe what would be optimal, strategy-wise, for different types of
string usage. We have known cases where empty strings should be small,
cases for long strings, short strings, MT strings, etc etc.
-benjamin