Re: [eigen] How to resize a partially fixed matrix

[Benoit Jacob <jacob.benoit.1@xxxxxxxxx>]

Thanks, patch applied.

The anonymous namespace solution gave me:
/home/bjacob/eigen2/Eigen/src/Core/util/Constants.h:246: warning:
‘Eigen::<unnamed>::NoChange’ defined but not used

I put a EIGEN_UNUSED before, which for GCC expands to
__attribute__((unused)), this solves the problem for GCC but this is
only a good solution if for every compiler that will issue such a
warning, there is an equivalent of GCC's __attribute__((unused)).

If we hit a dead end, we can always revert to named enums.

Benoit

2009/6/25 Patrick Mihelich <patrick.mihelich@xxxxxxxxx>:
> The syntax is inherited from C; it just looks weird because the brackets are
> empty. But I was being too clever earlier. We don't need static
> initialization here since it's just the type we care about. Here's a more
> sensible version:
>
> struct NoChange_t {};
> namespace {
>   NoChange_t NoChange;
> }
>
> The anonymous namespace avoids multiple definition troubles.
>
> Cheers,
> Patrick
>
> On Wed, Jun 24, 2009 at 3:52 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx>
> wrote:
>>
>> 2009/6/24 Patrick Mihelich <patrick.mihelich@xxxxxxxxx>:
>> > I like the NoChange API, but I get a little leery seeing overloads
>> > disambiguated on integral types (even though it seems OK here). A safe
>> > approach in general is to use a statically initialized tag object, that
>> > is
>> > change
>> >
>> > enum NoChange_t { NoChange };
>> >
>> > to
>> >
>> > struct NoChange_t {} NoChange = {};
>>
>> Hehe, that's why I initially proposed an approach with a struct
>> NoChange... but I retracted it because I was afraid of "multiply
>> defined object" issues as this is in a header file. I'll be honest: I
>> didn't know about the C++ syntax that you propose here. Do you confirm
>> it's standard c++98, and takes care of the "multiply defined object"
>> issue? If yes i'm all for your solution.
>>
>> Cheers,
>> Benoit
>>
>> >
>> > Cheers,
>> > Patrick
>> >
>> > On Wed, Jun 24, 2009 at 1:37 PM, Markus <grey_earl@xxxxxx> wrote:
>> >>
>> >> Am Mittwoch, 24. Juni 2009 schrieb Benoit Jacob:
>> >> > First of all the variant proposed:
>> >> > >> So I'm strongly for your code change, but changing the assertion
>> >> > >> to
>> >> > >> EIGEN_STATIC_ASSERT_EXACTLY_ONE_DYNAMIC_DIM(Matrix)
>> >> >
>> >> > doesn't make me comfortable, again because a method that works for
>> >> > (partially)dynamic matrices should also work on fixed-size matrices
>> >> > of
>> >> > the same size, I am very reluctant here.
>> >> >
>> >> > So here's what I did (changeset b030f9866bae):
>> >> >     * add resize(int, NoChange) and resize(NoChange, int)
>> >> >     * add missing assert in resize(int)
>> >> >     * add examples for all resize variants
>> >> >     * expand docs (part of which is from Tim Hutt's e-mail)
>> >> >
>> >> > Now if you want we can add another name (or completely rename) the
>> >> > new
>> >> > resize(int, NoChange) and resize(NoChange, int).
>> >> > But perhaps it's good enough like this? i don't have a strong
>> >> > opinion.
>> >> > resizeHoriz(size) and colwise().resize(size) look nice but they
>> >> > aren't
>> >> > much shorter to type than resize(NoChange,size);
>> >> >
>> >> > Cheers,
>> >> > Benoit
>> >> >
>> >>
>> >> Does this also work for fully dynamic matrices now?
>> >>
>> >> So if I have
>> >>
>> >> Matrix<double, Dynamic, Dynamic> mat;
>> >> mat.resize(3,3);
>> >> mat.resize(NoChange, 5);
>> >>
>> >> does this work? If yes, everything fine with me.
>> >>
>> >> Markus
>> >>
>> >>
>> >
>> >
>>
>>
>
>