Re: Fusion-ifying proto parse trees

"Eric Niebler" <eric@xxxxxxxxxxxxxxxxxxxx> writes:

> Not to worry, we're not stuck in a groove. :-) I asked you why you
> needed flat sequences because, when I get hung up on a design problem,
> I often find it useful to go back to the beginning, start from first
> principles, and make sure that I haven't overlooked something. That's
> all. I'm not trying to tell you what you do or don't need, I'm just
> trying to understand the problem.


Well, I don't mind sounding obnoxious, so -- admittedly without having
followed this discussion closely -- I'll tell you that you don't need
flattening :).

I see no reason that Spirit's performance (runtime or compile-time)
should depend on generating a "flat structure" in its expression
trees.  After all, you're just talking about the type structure.  The
result of evaluating the expression tree is fundamentally a struct
containing references to the objects in play.  It doesn't really
matter where the angle brackets are.

Just to be provocative, I'll even go further and claim that flattening
is evil :).  For the maximum performance in most linear fusion
algorithms, you actually want to impose a tree structure on flat data.
That is, divide the sequence in two (for example), do your algorithm
on the first half, and then continue on the second half, recursively.
Among other things, this approach makes the most of the compiler's
inherent inlining depth limitations.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com



-------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642