* tests/core/image/complex_image.cc: Update. --- milena/ChangeLog | 6 +++ milena/tests/core/image/complex_image.cc | 57 +---------------------------- 2 files changed, 8 insertions(+), 55 deletions(-) diff --git a/milena/ChangeLog b/milena/ChangeLog index 537aa93..44e4772 100644 --- a/milena/ChangeLog +++ b/milena/ChangeLog @@ -74,6 +74,12 @@ 2008-09-23 Roland Levillain &lt;roland(a)lrde.epita.fr&gt; + Make use of the new semantics of complexes in tests. + + * tests/core/image/complex_image.cc: Update. + +2008-09-23 Roland Levillain &lt;roland(a)lrde.epita.fr&gt; + Have complexes share common data and behave as pointers. * mln/core/complex.hh (mln::internal::complex_data<D>): New class. diff --git a/milena/tests/core/image/complex_image.cc b/milena/tests/core/image/complex_image.cc index 7711f38..39e6239 100644 --- a/milena/tests/core/image/complex_image.cc +++ b/milena/tests/core/image/complex_image.cc @@ -91,60 +91,7 @@ int main() // A pset. p_complex<D, point2d> pc(c); - /* An any-face handle (on E0). - - Note that AF is built on `e0_', not `e0' (e0_ is built on - `pc.cplx()', not `c'), since the p_complex `pc' makes a copy of - the complex `c', and crossed-ownership tests doesn't work. I.e., - - pc.has(e0); - - is false. - - - FIXME: This might be a problem, since `pc.cplx()' and `c' - represent the same complex, in two different memory location (but - the former is controlled by PC, while the latter can be modified, - or even destroyed). This is a common problem for ``big'' values - that we don't want to manipulate by value (copy), or when we - don't want to use expensive, deep comparisons of pset to ensure - consistency. Here (and in graph-based images), we choose to - create a copy of the pset once, and manipulate it with a - tracked_ptr, to ensure both - - 1. perfect control of the lifetime of the pset (here, you can - delete `c', and `pc' will still be valid); - - 2. no pset duplication when creating new images based on it. - - - I (Roland) don't see elegant solutions here. A possiblity would - be to disconnect a face_handle from its complex (currently, a - face_handle is a bit like a Trivial Iterator from the C++ - Standard Library), but this means relaxed dynamic checks, and - more obscure errors. - - At least, we could have better error messages, i.e., something - like - - mln/core/image/complex_image.hh 267: - mln::complex_image<D, P, V>::operator(): Uncompatible p_complex. - - instead of - - mln/core/image/complex_image.hh:267: - typename mln::complex_image<D, P, V>::lvalue - mln::complex_image<D, P, V>::operator()(const mln::complex_psite<D, P>&) - [with unsigned int D = 2u, - P = mln::point<mln::grid::square, int>, - V = mln::value::int_u<8u>]: - Assertion `this->data_->pc_.has(p)' failed. - - (which looks even uglier in the original, non-indented version). - - Ask Akim for his improved versions of abort() and assert(). */ - face_handle<1, D> e0_(pc.cplx(), 0); - any_face_handle<D> af(e0_); + any_face_handle<D> af(e0); // An associated psite. complex_psite<D, point2d> cs(pc, af); @@ -186,7 +133,7 @@ int main() // Create and init an image based on PC. ima_t ima(pc, values); - // Check the value associated to edge E0_ (through complex psite CS). + // Check the value associated to edge E0 (through complex psite CS). mln_assertion(ima(cs) == 1u); -- 1.6.0.1