[J3] [EXTERNAL] Re: Exception handling - Golang proposal
Vipul Parekh
parekhvs at gmail.com
Fri Jul 12 22:45:31 EDT 2019
On Fri, Jul 12, 2019 at 2:26 PM Bill Long via J3
<j3 at mailman.j3-fortran.org> wrote:
>..
>
> Unfortunately, we’ve already had this same problem with features that start as a TS. Some of the Coarray TS specifications got changed in F2018, for example. Vendors have learned - don’t implement the TS, or at least expect to have to redo some of it later when the final standard comes out.
> ..
Prima facie, TS appears the reasonable approach toward developing
exceptions handling feature in Fortran, especially considering the
concerns with impact on performance as well as the strong criticism in
computing literature on existing facilities in other languages, both
of which suggest an appreciable likelihood of getting this wrong in
Fortran unless it is pursued with a well-thought out set of use cases
to guide the development of the formal requirements.
A (nearly) concurrent "trial implementation" to go along with the (TS)
feature development is exactly what Fortran needs: I think this
feature rather too important for scientific and technical computing to
be ignored for any longer, so the question is whether one can motivate
and influence the open-source platforms - GCC/LLVM - and/or some
vendor(s) to "step up to the plate" and "bat" for Fortran.
Nonetheless, the real challenge with exceptions handling in Fortran
appears to remain with the "standard" process of feature development:
with putting together *succinct and valid* use cases instead of LONG
PREAMBLEs that can lead to sound formal requirements which can then
help formulate the specifications.
Toward this, I think the best places to review and study are NOT other
languages and their proposals but the *actual codes* of scientific and
technical applications (high-performance/large-scale simulations,
multiphysics modeling, etc.) developed using competing paradigms to
Fortran such as with C++, Microsoft .NET, Python to understand what
works with exceptions handling when it comes to scientific and
technical computing and would likely help in Fortran also and which
aspects to avoid.
Regards,
Vipul Parekh
More information about the J3
mailing list