https://svn.lrde.epita.fr/svn/oln/trunk/milena Index: ChangeLog from Roland Levillain &lt;roland(a)lrde.epita.fr&gt; Solve infinite recursive calls in linear::impl::convolve_. * mln/linear/convolve.hh (convolve_(any, const I&, const Weighted_Window<W>&, any, O&)): Move the body of this function... (convolve_(const I&, const Weighted_Window<W>&, O&)): ...here (new function). Use it to factor... (convolve_(Speed_I, const I&, const Weighted_Window<W>&, Speed_O, O&)) (convolve_(any, const I&, const Weighted_Window<W>&, any, O&)): ...the implementations of these functions. convolve.hh | 62 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 61 insertions(+), 1 deletion(-) Index: mln/linear/convolve.hh --- mln/linear/convolve.hh (revision 1647) +++ mln/linear/convolve.hh (working copy) @@ -1,4 +1,4 @@ -// Copyright (C) 2007 EPITA Research and Development Laboratory +// Copyright (C) 2007, 2008 EPITA Research and Development Laboratory (LRDE) // // This file is part of the Olena Library. This library is free // software; you can redistribute it and/or modify it under the terms @@ -65,6 +65,54 @@ namespace impl { + /* FIXME: We must clean up the interface of + mln::linear::impl::convolve_: + + - either allow certain patterns of speed traits (e.g., + any/any, fastest/fastest, fastest/any, etc.). In this + case, the generic version should abort at compile time; + + - or accept all combinations (which is the current case), and + default to the slowest one (presumably any/any). + */ + + // Fwd decl. + template <typename I, typename W, typename O> + inline + void convolve_(const I& input, + const Weighted_Window<W>& w_win_, + O& output); + + /// Default version, delegating to the generic version. + template <typename Speed_I, typename I, typename W, + typename Speed_O, typename O> + inline + void convolve_(Speed_I, const I& input, + const Weighted_Window<W>& w_win_, + Speed_O, O& output) + { + /* Don't delegate using such a call: + + \code + impl::convolve_(trait::image::speed::any(), input, + w_win_, + trait::image::speed::any(), output); + \endcode + + since it would end up with infinite recursion. The reason + is that the compiler would select this function (in which + you read this very comment), instead of the next one (with + input and output speed traits set to `any'), to resolve the + call. This is because C++ overloading rules favor the + generic function over the more specialized one. And we + cannot use explicit partial specialization, since it just + doesn't exist for functions. + + Hence the chosen solution: create and call another + overloading for mln::linear::impl::convolve_, with no + unnatural selection behavior. */ + impl::convolve_(input, w_win_, output); + } template <typename I, typename W, typename O> inline @@ -72,6 +120,18 @@ const Weighted_Window<W>& w_win_, trait::image::speed::any, O& output) { + // Delegate the call to the generic version. + impl::convolve_(input, w_win_, output); + } + + /// A factored implementation of the most generic version of + /// mln::linear::impl::convolve_. + template <typename I, typename W, typename O> + inline + void convolve_(const I& input, + const Weighted_Window<W>& w_win_, + O& output) + { const W& w_win = exact(w_win_); mln_piter(I) p(input.domain());