increment a single element of matrix/vector

increment a single element of matrix/vector

[Patrick Alken]

I was thinking of adding two routines to GSL, to increment a single 
element of a matrix or vector by some amount, since I'm always doing 
things like:

gsl_vector_set(v, i, gsl_vector_get(v, i) + x);

which I think is ugly and annoying. Something like:

gsl_vector_add_element(v, i, x);
or
gsl_matrix_add_element(m, i, j, x);

would be really nice. Does anyone have any better suggestions for the 
function name? Or also is there another solution to this problem that I 
don't know about?

Patrick

Re: increment a single element of matrix/vector

[Rhys Ulerich]

> I was thinking of adding two routines to GSL, to increment a single element
> of a matrix or vector by some amount, since I'm always doing things like:
>
> gsl_vector_set(v, i, gsl_vector_get(v, i) + x);
>
> which I think is ugly and annoying....

Agreed.  This has often bugged me.

> Or also is there another solution to this problem that I don't know about?

How about using

INLINE_DECL double * gsl_vector_ptr (gsl_vector * v, const size_t i);

to say *gsl_vector_ptr(v, i) += x ?  Never knew about this until I
just went fishing for it.

- Rhys

Re: increment a single element of matrix/vector

[Rhys Ulerich]

> *gsl_vector_ptr(v, i) += x

Of course, make sure warnings about expressions without side effects
are turned on.  Otherwise it may be a long debugging session before
you discover a

    gsl_vector_ptr(v, i) += x

mistake for integer-like types.

- Rhys

Re: increment a single element of matrix/vector

[Patrick Alken]

On 01/27/2014 07:14 PM, Rhys Ulerich wrote:
>> *gsl_vector_ptr(v, i) += x
> 
> Of course, make sure warnings about expressions without side effects
> are turned on.  Otherwise it may be a long debugging session before
> you discover a
> 
>     gsl_vector_ptr(v, i) += x
> 
> mistake for integer-like types.
> 
> - Rhys
> 

Nice find - unfortunately since it returns a pointer to the element, you
need to do:

*(gsl_vector_ptr(v, i)) += x;

This could lead to trouble so its probably better to have a function
which does range checking, etc. I'll think a little more on a good way
to do this.

Thanks,
Patrick

Re: increment a single element of matrix/vector

[Patrick Alken]

The functions gsl_vector_inc and gsl_matrix_inc are now in the git 
repository to accomplish this

On 01/27/2014 08:06 PM, Patrick Alken wrote:
> On 01/27/2014 07:14 PM, Rhys Ulerich wrote:
>>> *gsl_vector_ptr(v, i) += x
>> Of course, make sure warnings about expressions without side effects
>> are turned on.  Otherwise it may be a long debugging session before
>> you discover a
>>
>>      gsl_vector_ptr(v, i) += x
>>
>> mistake for integer-like types.
>>
>> - Rhys
>>
> Nice find - unfortunately since it returns a pointer to the element, you
> need to do:
>
> *(gsl_vector_ptr(v, i)) += x;
>
> This could lead to trouble so its probably better to have a function
> which does range checking, etc. I'll think a little more on a good way
> to do this.
>
> Thanks,
> Patrick

Re: increment a single element of matrix/vector

[Rhys Ulerich]

>>>> *gsl_vector_ptr(v, i) += x

>> *(gsl_vector_ptr(v, i)) += x;

Why the extra parens?  They hurt nothing, but the former statement
without them works just fine for me.  Unless I'm mistaken, the
function application happens before the dereference.  And the
dereference is much, much higher than anything like += or *= that the
user might want.

>> This could lead to trouble so its probably better to have a function
>> which does range checking, etc.

There is range checking in gsl_vector_ptr, so long as someone doesn't
start doing pointer arithmetic on its result.

Lastly, the pointer-based solution is nicely general: *(i > j ?
gsl_vector_ptr(v, i) : gsl_vector_ptr(j)) += 1

- Rhys

Re: increment a single element of matrix/vector

[Patrick Alken]

On 01/28/2014 02:16 PM, Rhys Ulerich wrote:
>>>>> *gsl_vector_ptr(v, i) += x
>>> *(gsl_vector_ptr(v, i)) += x;
> Why the extra parens?  They hurt nothing, but the former statement
> without them works just fine for me.  Unless I'm mistaken, the
> function application happens before the dereference.  And the
> dereference is much, much higher than anything like += or *= that the
> user might want.

Yes sorry I didn't see the * the first time I read that
>>> This could lead to trouble so its probably better to have a function
>>> which does range checking, etc.
> There is range checking in gsl_vector_ptr, so long as someone doesn't
> start doing pointer arithmetic on its result.
>
> Lastly, the pointer-based solution is nicely general: *(i > j ?
> gsl_vector_ptr(v, i) : gsl_vector_ptr(j)) += 1
>
> - Rhys

Hmm ok you convinced me, I'll reverse those last few commits