This adds the Pnt{1,2,3,4,5,6} structures.
This adds the traits:
− AnyPnt
− FloatPnt
− PntExt
− FloatPntExt
− Orig (to return the zero point)
− PntAsVec
− VecAsPnt
This adds operator overloading:
− Pnt + Vec
− Pnt - Vec
− Pnt * Scalar
− Pnt / Scalar
− Pnt + Scalar
− Pnt - Scalar
− Iso * Pnt
− Rot * Pnt
− Pnt * Iso
− Pnt * Rot
This changes some behavior:
− Iso multiplication with a Vec does not translate the vector any more.
− ToHomogeneous adds a 0.0 at the end of a Vec and a 1.0 at the end of a Pnt.
− FromHomogeneous performs w-normalization on a Pnt, but not on a Vec.
− The Translate<Vec> trait is never implemented (i-e. a Vec is not to be translated).
cc #25
Version of rustc: 0.11.0-pre-nightly (faa7ba7 2014-05-31 01:06:40 -0700).
Main changes:
* `cmp::Ord` -> `cmp::PartialOrd`
* `cmp::Eq` -> `cmp::PartialEq`
Note that `na::PartialOrd` is not the same as `cmp::PartialOrd`
(which lacks a lot of partial ordering operators).
The ColSlice implementation for fixed size matrices returns a DVec,
while this is probably not optimal performance-wise, the dynamic nature
of the result makes this necessary. Using a data type presenting the
ImmutableVector trait would solve this, but it looks like a non-trivial
change.
This allows the implementation of householder reflection without relying
on knowledge of DVec. This required a new member in the Indexable trait:
the shape() function, which returns the maximum index available.
This is to make people prefer the functional style.
Things like `a.dot(b)` dont make sense per se (there is no reason for `a` to have a different
status than `b`). Using static methods avoid this.
In-place methods are left unchanged.
Before, it was too easy to use an out of place method instead of the inplace one since they name
were pretty mutch the same. This kind of confusion may lead to silly bugs very hard to understand.
Thus the following changes have been made when a method is available both inplace and out-of-place:
* inplace version keep a short name.
* out-of-place version are suffixed by `_cpy` (meaning `copy`), and are static methods.
Methods applying transformations (rotation, translation or general transform) are now prefixed by
`append`, and a `prepend` version is available too.
Also, free functions doing in-place modifications dont really make sense. They have been removed.
Here are the naming changes:
* `invert` -> `inv`
* `inverted` -> `Inv::inv_cpy`
* `transpose` -> `transpose`
* `transposed` -> `Transpose::transpose_cpy`
* `transform_by` -> `append_transformation`
* `transformed` -> `Transform::append_transformation_cpy`
* `rotate_by` -> `apppend_rotation`
* `rotated` -> `Rotation::append_rotation_cpy`
* `translate_by` -> `apppend_translation`
* `translate` -> `Translation::append_translation_cpy`
* `normalized` -> `Norm::normalize_cpy`
* `rotated_wrt_point` -> `RotationWithTranslation::append_rotation_wrt_point_cpy`
* `rotated_wrt_center` -> `RotationWithTranslation::append_rotation_wrt_center_cpy`
Note that using those static methods is very verbose, and using in-place methods require an
explicit import of the related trait.
This is a way to convince the user to use free functions most of the time.
Everything changed, hopefully for the best.
* everything is accessible from the `na` module. It re-export
everything and provides free functions (i-e: na::dot(a, b) instead of
a.dot(b)) for most functionalities.
* matrix/vector adaptors (Rotmat, Transform) are replaced by plain
types: Rot{2, 3, 4} for rotation matrices and Iso{2, 3, 4} for
isometries (rotation + translation). This old adaptors system was to
hard to understand and to document.
* each file related to data structures moved to the `structs` folder.
This makes the doc a lot more readable and make people prefer the
`na` module instead of individual small modules.
* Because `na` exists now, the modules `structs::vec` and
`structs::mat` dont re-export anything now.
As a side effect, this makes the documentation more readable.
