[Insight-users] Image offset is giving bad pointer for large datasets (7Gb)

[Luis Ibanez]

Hi Mike,

OffsetValueType is certainly not intended to be a Pointer,
but indeed to be the difference between two pointers.

The typical use of OffsetValueType is to locate a pixel
in the block of memory of an image:

                  OffsetValueType  offset = 555;
                  image->GetBuffer()  +  offset;

The other use of OffsetValueType is to be the components
of the differences between two indices, which by extension
should make us consider using ptrdiff_t also as the component
type of the itk::Index and the itk::Size, since it is a valid use
to allocate a 1D image of 10Gb.

I'm not sure that the argument of whether some programmer
may get confused, should take precedence over the argument
of what is the correct type to use.

The first problem gets solved with three lines of comments
in the declaration of

          /**  Here we explain why...  */
          typedef   ptrdiff_t     OffsetValueType

Following basic practices of Generic Programming, in all
subsequent ITK code, we should use the type "OffsetValueType"
and there is no need to expose what OffsetValueType actually is.


Users that digg deep into the code will be already confused
by the fact that different platforms define "int" and "long" in
different ways. What is interesting about ptrdiff_t is that is
clearly defined in a platform-independent way.


The best way to protect people from ignorance,
is to actually educate them.   :-)


        Luis


--------------------------------------------------------------------------------------
On Sat, Jul 11, 2009 at 8:50 AM, Mike Jackson
<mike.jackson at bluequartz.net>wrote:

> I think the only thing I have against using ptrdiff_t is that it
> doesn't describe exactly what you are wanting in a type. ptrdiff_t is
> the result of subtraction between two _pointers_. Is OffsetValueType
> really a Pointer or an index into an array? I think if programmers
> started digging when debugging and saw ptrdiff_t then that might be
> confusing to them as they might start thinking that they can cast an
> OffsetValueType to a pointer and start doing pointer manipulations on
> it.
>   It seems to me that OffsetValueType is just that, a _value_, and
> not a pointer. Also, do we fully understand how ptrdiff_t is defined
> on all the platforms that ITK runs on? Can we be sure we take that
> into account properly? In comparison we know _exactly_ how it behaves
> because we would be using either "int" or "long long int" which also
> has the advantage of describing exactly what we need, a plain old
> signed integer. Again, when new ITK programmers start digging to find
> out what OffsetValueType and come up with a plain old integer there
> most likely will not be any misunderstanding about the type.
>
> Mike Jackson
>
> On Sat, Jul 11, 2009 at 7:02 AM, Luis Ibanez<luis.ibanez at kitware.com>
> wrote:
> >
> > What's wrong with using :     ptrdiff_t   ?
> >
> >
> >        Luis
> >
> >
> > -----------------------------------------------------------
> > On Fri, Jul 10, 2009 at 8:34 PM, Mike Jackson <
> mike.jackson at bluequartz.net>
> > wrote:
> >>
> >> Probably should have googled that first. Seems to be a POSIX extension
> >> so probably isn't that portable.
> >>
> >>  From an earlier email:
> >> ------
> >> On recent realization:
> >>
> >>             "size_t"    is unsigned       :-/
> >>
> >> and we need the OffsetValueType to be signed,
> >> since we use it to compute differences...
> >> ------
> >> so why doesn't ITK just define it to int or long long int depending on
> >> the size of size_t? That test could be run during CMake time like all
> >> the others?
> >>
> >> #if size_of_size_t == 4
> >> typedef signed int OffsetValueType;
> >> #elif size_of_size_t == 8
> >> typedef signed long long int OffsetValueType;
> >> #endif
> >>
> >> Of course we would probably want to use the predefined 64 bit type
> >> that the ITK/CMake tests come up with: __int64, _int64_t, int64_t or
> >> whatever they are..
> >>
> >> Just my thoughts?
> >> Mike
> >>
> >> On Fri, Jul 10, 2009 at 8:23 PM, Mike
> >> Jackson<mike.jackson at bluequartz.net> wrote:
> >> > Is there a "ssize_t" type? Or is that some contrived type on some of
> >> > the projects that I work on?
> >> >
> >> > Mike
> >> >
> >> > On Fri, Jul 10, 2009 at 6:29 PM, Luis Ibanez<luis.ibanez at kitware.com>
> >> > wrote:
> >> >>
> >> >> On recent realization:
> >> >>
> >> >>              "size_t"    is unsigned       :-/
> >> >>
> >> >> and we need the OffsetValueType to be signed,
> >> >> since we use it to compute differences...
> >> >>
> >> >> It seems that what we need is the type
> >> >>
> >> >>                         "ptrdiff_t"
> >> >>
> >> >> which is supposed to represent the differences
> >> >> between two pointers, and therefore should be
> >> >> capable of measuring distances between any
> >> >> two locations in memory.
> >> >>
> >> >> I'm now rerunning the Experimental with
> >> >> "ptrdiff_t" instead of "size_t".
> >> >>
> >> >>
> >> > Luis
> >> >
> >
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20090711/1579eb32/attachment-0001.htm>