[Insight-users] Examples wiki

[David Doria]

On Mon, Jul 13, 2009 at 1:41 PM, Bill Lorensen<bill.lorensen at gmail.com> wrote:
> My concern is that these examples would not be compiled, regression
> tested, valgrind'ed, style checked, etc. Will they compile on all
> platforms? If we can come up with mechanism to test them, then it
> sounds like a good idea.
>
> I looked at one example on the vtk wiki:
> http://www.vtk.org/Wiki/Get_the_names_of_all_of_the_data_arrays
> A few things I noticed:
> 1) It uses hard tabs. This code could not be checked into the itk repository.
> 2) main uses exit() rather than return().
> 3) It creates a vtk object with ::New, but does not do a Delete().
>
> 1) would be caught at checkin.
> 2) might be picked up as a warning by one of the itk compilers.
> 3) would be detected by valgrind.
>
> Bill
>

Bill -

Certainly all valid concerns. My intention, however, has been to
provide something simply to demonstrate the ideas. There could
certainly be a disclaimer "This code is intended to demonstrate
concepts and is not guaranteed to be memory-leak free and does not
necessary adhere to VTK coding standards". After all, they are not
part of VTK, but simply a cut and paste of a users code (which,
obviously, is not usually going to follow the standards!).

As a user who has gone through the learning curve (well, GOING
through, haha), personally I would have benefited from such things
immensely - in fact, they are directly the result of my learning VTK!
They are kind of like pointers from ideas that the user pre-VTK
understands, e.g. "I want to make a cube" to vtk classes/ideas, e.g.
vtkCubeSource. If a new user digs through the examples that ship with
the software, it is sometimes difficult to find such a pointer.

I guess my concern with actually checking in this kind of example is
that it would cause a lack of "ability/desire to contribute". That is,
"non-developers" may not want to take the time to learn the coding
styles and practices enough to get their examples into shape to add to
the repository, but they would be much more willing to "dump" a page
of code onto the wiki. I guess I was kind of counting on the "wiki"
phenomenon to clean up the code - that is, someone like yourself
quickly caught a couple of problems in the
Get_the_names_of_all_of_the_data_arrays example and then they would be
fixed. Also, they would not want the responsibility of ensuring the
examples compile cross-platform / with different versions of VTK, etc.
- the point is just "it took me an afternoon to figure X out - maybe
this will save you some trouble". That all said - maybe we can't count
on this type of contribution, either, as not a SINGLE example has been
added to the examples section I started by another user - but it DOES
seem like people are looking at them judging by some of the hit
counters being > 500 (assuming this is accurate and "bot free").

To your point about regression testing - again this would rely on the
wiki mentality. There are plenty of other places that code appears
which is not regression tested (the Insight journal,
xxSoftwareGuide.pdf, etc) and, clearly, it is not even possible to
test (as it is in a document format). When I find a source like that
and then realize it is not compatible with the current release, there
is basically nothing that can be done. With the wiki, if a user finds
an error in the code due to regression, it can simply be modified.
Obviously the check for this is more manual than an actual regression
test, but, because of my "lack of contribution" notes above, it seems
like a reasonable alternative.

Maybe there could be some sort of "Examples sandbox" which could have
public svn access rights for people to "dump" code into, and then as
experienced users with time on their hands come along they could
slowly be cleaned up into actual, regression testable examples? I just
fear something like that would not get as much visibility as the wiki
- where from the VTK homepage the user can simply find "Help->Wiki"
and many examples are visible right away. And if a user downloads the
release version of the code (non-cvs), these would clearly be
inaccessible.

Please let me know if we should continue this conversation / if you
have any comments on these ideas.

Thanks,

David