On Sat, Sep 4, 2010 at 2:09 AM, Scott McMurray <me22.ca+boost@gmail.com>wrote:
On Fri, Sep 3, 2010 at 10:09, Dave Abrahams <dave@boostpro.com> wrote:
Can someone explain to me, in English, what this technique allows me to express that I couldn't otherwise?
Is this just about allowing generic composition of lambdas?
It feels like it to me.
What I've understood so far is that _0_n and _n are essentially the same,
I've changed to 1 based indexing for both the abstraction number and the argument number. _n in a boost.bind that isn't enclosed in another boost.bind is the same as _1_n in DBB. _n in an unprotected boost.bind that is enclosed in 1 other boost bind is the same as _2_n in DBB. _n in an unprotected boost.bind that is enclosed in an unprotected boost bind that is enclosed in yet another unprotected boost bind is the same as _3_n in DBB.
but that _0_n requires a lambda to scope them explicitly (sortof like protect).
Right now for the particular case of _1_n, an enclosing lam is required, but that need not necessarily be the case. It is similar to protect only in that they enclose other terms, that's where the similarity ends.
The advantage of _y_n is then that you can refer to the _ns in enclosing lambdas (aka outside of y protects).
Am I on the right track, David?
Yes, that is the advantage of De Bruijn indices when compared to protects. The enclosing lambdas' arguments can be both referred to and mixed. David -- David Sankel Sankel Software www.sankelsoftware.com 585 617 4748 (Office)