[Insight-developers] Doc updates for ImageIOBase

[Bill Lorensen]

Steve,

I think making an empty implementation in ImageIOBase is appropriate.
Forcing developers to provide a method that is not required is bad
practice. Better documentation is always needed. Thanks for spending
the time on this.

As for the number of folks providing their own ImageIO's, I know that
when I was at GE, we had at least two of our own ImageIO's.  The
Slicer projects, www.slicer.org, has two, itkMRMLIDImageIO.h and
itkMGHImageIO.h. These are both discovered at run time.

A quick search on Google Code Search, http://www.google.com/codesearch
, shows that the Orfeo Toolbox,
http://sourceforge.net/projects/orfeo-toolbox/ , and
http://cvs.openmicroscopy.org.uk have implemented their own
ImageIO.Orfeo seems to have implemented at least 5 private ImageIO's.
Actually, one of the Orfeo IO's, otbONERAImageIO, implements a
non-trivial WriteImageInformation.

Worldwide, I suspect that the number is fairly significant.

Happy New Year,

Bill

On Sat, Jan 3, 2009 at 12:21 AM, Steve M. Robbins <steve at sumost.ca> wrote:
> Hi Bill,
>
> You met my expectations.  :-)
>
> On Fri, Jan 02, 2009 at 11:45:14PM -0500, Bill Lorensen wrote:
>
>> We don't know whether any of our customers use WriteImageInformation
>> and m_Initialized. Many customers have written their own IMAGE IO
>> classes and it would not be fair to them to change the API. Since we
>> have released that code, we cannot remove it.
>
> I was not seriously proposing to remove these things.  But I do wonder
> how to document them at the abstract class level since the
> documentation that exists is wrong.
>
> For example: should WriteImageInformation() be documented as obsolete
> and become an empty virtual function in ImageIOBase (i.e. existing
> implementations can override it, but new ImageIOBase subclasses
> needn't bother)?  Or should it be documented to actually write the
> info and bug reports created for each subclass that doesn't do so?  Or
> simply be documented as "some subclasses may implement this" (the
> status quo)?  Or ... ?
>
> Thanks,
> -Steve
>
> P.S.  Do you have some data on how many customers have private ImageIO
> classes?
>
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.9 (GNU/Linux)
>
> iD8DBQFJXvXN0i2bPSHbMcURArP8AJ4tpfHldc5SYRbCYGI6y6s8n5tL9ACgsKdN
> KPPyoBn4PlcGV98/MdCe+Dg=
> =/6KB
> -----END PGP SIGNATURE-----
>
>