https://svn.lrde.epita.fr/svn/oln/trunk/milena Index: ChangeLog from Roland Levillain &lt;roland(a)lrde.epita.fr&gt; Add some comments on mln::canvas::morpho::algebraic_union_find<F>. * mln/canvas/morpho/algebraic_union_find.hh (mln::canvas::morpho::algebraic_union_find<F>::make_set) (mln::canvas::morpho::algebraic_union_find<F>::do_union): Add FIXMEs on the non genericity of this canvas w.r.t. the type of the attribute. algebraic_union_find.hh | 34 ++++++++++++++++++++++++++++++++++ 1 file changed, 34 insertions(+) Index: mln/canvas/morpho/algebraic_union_find.hh --- mln/canvas/morpho/algebraic_union_find.hh (revision 1982) +++ mln/canvas/morpho/algebraic_union_find.hh (working copy) @@ -120,6 +120,28 @@ void make_set(const psite& p) { parent(p) = p; + /* FIXME: What if the value_type of DATA (i.e., A) were not + based on a accu::count_<mln::pix>? Currently, nothing + enforces this, but the code below expects this line to be + valid: + + data(p).take_as_init(make::pix(f.input, p)) + + which probably restricts the kind of input images. + + If we want to be more generic, the initialization should + read something like: + + init_data(p); + + i.e., the functor for the initialization of data should + be passed as an argument to the canvas' ctor. + + Of course, we might want to restrict attributes to the + accumulator accu::count_<mln::pix> (which is perfectly + acceptable), but then this class should statically check + the conformance of the template parameter A to this + constraint. */ data(p).take_as_init(make::pix(f.input, p)); // FIXME: algebraic so p! } @@ -149,6 +171,18 @@ { if (equiv(r, p)) { + /* FIXME: Same remark as above concerning the + initialization of data(p); instead of + + data(p).take(data(r)); + + we should (or could) have + + unite_data(p, r); + + so as to keep the generic aspect of this canvas + (as long as the set of acceptable types for the + template parameter A is not bound). */ data(p).take(data(r)); parent(r) = p; }