🚀 go-pugleaf

For better and worst, we will most likely follow whatever ANSI C++
says.  If you really want to see different behavior from standard
behavior then, I suggest you see about getting your opinions heard by
the ANSI C++ committee.  See comp.std.c++ for more details.

> Date: Mon, 14 Sep 92 18:17:51 +0200
> From: dak@pool.informatik.rwth-aachen.de (David Kastrup)
> Message-Id: <9209141617.AA01219@messua>
> To: bug-g++@prep.ai.mit.edu

> There was some discussion going on whether the above construction
> was legal.
> Although it is not explicitly stated in ARM, it is a very important
> construction, because:
> Templates cannot be nested. Suppose you have a sort template that
> should work as fast as possible on linked lists. It would look something
> like:
> template <class T, T * T::*link>
> You could modify the link parameter to be a parameter for the
> subroutine, but then you are replacing a compile time constant
> by a variable which would usually assume one value anyhow.

> Consider defining a Fourier transform working on arbitrary fields
> with roots of 1.

> Then you could use something like
> template <class Field, unsigned N, Field * Nthroot>
> and use it something like
> Complex const NR = exp(2*pi*Complex(0,1)/N);
> fft<Complex, 1024, &NR>.transform(...
> or
> Modular<64513> NR = 9; //Value is wrong
> fft<Modular<64513>, 1024, &NR>.transform(...

> Although it would work with complex numbers just as fast as usual,
> it would without modification work even faster with other fields
> as a base.

> Other classes and functions might need other information
> necessarily not included in the basic type, but being constant.

> PLEASE let the template scopes begin immediately after their declaration,
> not after the closing >.