[J3] C TS 18661

[Keith Bierman]

On Wed, Jan 30, 2019 at 1:29 PM Vipul Parekh via J3 <
j3 at mailman.j3-fortran.org> wrote:

> On Wed, Jan 30, 2019 at 11:09 AM Steve Lionel via J3
> <j3 at mailman.j3-fortran.org> wrote:
> >
> > ..  A case for the other intrinsics mentioned would
> > have to be made using different arguments. ..
> > ..
> ...e BESSEL functions
> introduced in Fortran 2008 available some place?
>
>
There's a lot that goes into such things. Getting the fp details right is
hard (fdlibm is still probably the closest portable library), and as Van
noted, it took him 17 years to find a bug. Since Oracle seems to have
destaffed the project (long after my departure) I'm not sure if/when anyone
will fix *that* bug. No doubt others may emerge....

Compiler writers are seldom really good at fp implementation. So making
more things intrinsic may well result in *worse* results ... the compiler
writers need *something* when you've made it part of the Standard.,.and
users typically rely on the compiler getting intrinsic things correct ....
so it sets up an easily missed failure mode.

Adding lots of really low level things to high level languages seems
unfortunate to me.  Modern optimizing compilers often detect/move things
pretty long distances (especially if it's in the same basic block).

I appreciate that these (and many more) routines are important and popular,
and probably deserving good modular interface(s) and perhaps standards of
their own.

I don't favor the C solution of exiling *everything* to libraries, but just
adding everything useful because it is useful is not likely to really make
life better in the long run.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.j3-fortran.org/pipermail/j3/attachments/20190130/c233566b/attachment-0001.html>