On May 9, 11:11 am, Steve Lionel <Steve.Lio...@[EMAIL PROTECTED]
> wrote:
> On Thu, 8 May 2008 19:56:55 -0700 (PDT), Damian <dam...@[EMAIL PROTECTED]
>
wrote:
> >I found this code that I *think* was submitted with the bug re****t
> >(can't recall if this was for the first bug re****t, which was fixed,
> >or for the second, which was not fixed in the initial release of 10.1
> >but might be fixed in a more recent release):
>
> Ok - I found it and remember the case now. This particular problem is
not
> fixed in a 10.1 update but will be fixed in the next major release, as
we want
> to make sure that the fix gets adequate external testing. It is a
rather
> unsusual combination of things to show the problem, but it is indeed an
> outstanding bug in 10.1.
> --
> Steve Lionel
> Developer Products Division
> Intel Cor****ation
> Nashua, NH
>
> For email address, replace "invalid" with "com"
>
> User communities for Intel Software Development Products
> http://softwareforums.intel.com/
> Intel Fortran Sup****t
> http://sup****t.intel.com/sup****t/performancetools/fortran
> My Fortran blog
> http://www.intel.com/software/drfortran
I
Regarding the re****ted bug being "a rather unusual combination of
things to show a problem," the first leak I re****ted was simpler.
When that was fixed, I had to go to a slightly more complicated
example to demonstrate that the fix was not sufficiently universal.
Even that example is relatively simple compared to what is common in a
certain class of object-oriented scientific software projects.
I've been working on a scientific software design methodology that
relies upon heavy use of user-defined assignments and operators with
arguments that contain nested derived types. This methodology
facilitates writing "software abstractions that resemble blackboard
abstractions" to quote my colleague Kevin Long. Specifically, it
facilitates code that closely resembles equations in the physical
sciences (scalar, vector & tensor field equation). Numerous research
groups have done similar work in C++, e.g. the Sundance project at
Sandia National Laboratories, the Overture project at Lawrence
Livermore National Laboratories, and the Sophus project at the
University of Bergen.
I believe Fortran will eventually outpace these C++ efforts in several
respects (not the least of which being the automatic memory
deallocation features we're discussing), but this will only happen if
cases like the one I constructed are considered of central im****tance
rather than unusual cases.
Damian
|
