Re: ddd-3.3.9-test2 build feedback

[Arnaud Desitter]

----- Original Message ----- 
From: "Tim Mooney" <address@hidden>
Newsgroups: gmane.comp.debugging.ddd.general
Sent: Monday, May 17, 2004 3:06 AM
Subject: Re: ddd-3.3.9-test2 build feedback


> In regard to: Re: ddd-3.3.9-test2 build feedback, Arnaud Desitter said
(at...:
>
> >>> Tim Mooney wrote:
> >>
> >> The performance improvements are hard to miss.
> >
> > There was a lot of potential for "easy" improvements. Most changes
between
> > ddd 3.3.7 and ddd 3.3.9 relate to speed up.
> >
> > Going futher will be more difficult.
> > The string class used by ddd is not reference counted. Switching to
> > std::string in the few locations shown by a profiler could improve
speed.
>
> Would there be any benefit to completely switching ddd to std::string, and
> getting rid of the custom/GNU string class?  Would that be a difficult
> process, or would it just be a lot of relatively straightforward code
> replacement?

I would be a big job as the GNU string class does more than std::string.
The danger is to spend a lot of time working around the various std::string
implementation problems. A potential benefit is to benefit from more
advanced algorithms such as reference counted buffer/copy-on-write which
may speed up the multiple copies.
In my opinion, a simpler approach is to locate the most used GNU strings
and replace these by std::string and see if it makes it better with respect
to
speed and memory.
Anyway, the decision will depend on who does the job.

> > Anyway it is good to know that you noticed.
>
> I personally don't care for "artificial version inflation", but the
> changes seem significant enough so that you may want to call the official
> release 3.4.0 or perhaps 3.5.0.

I do not mind except that somebody as to generate a new splash screen.
Ask Andrew.

Regards,