Re: Praise for Gfortran (finally)

On 23 Sep., 19:20, Beliavsky <beliav...@xxxxxxx> wrote:
On Sep 18, 3:56 am, paul.richard.tho...@xxxxxxxxx wrote:
We are sharply aware of this problem.  It has come about largely
because of the inadequacy, in the past, of our testsuite.  Not having
a take-the-standard-and-write-a-testsuite yardstick to go by, ours has
grown as we have fixed bugs and regressions.  

They do exist, for example http://www.fortran.com/f95ts.html. How
much do they cost? I would donate a few hundred dollars toward buying
one, if the Gfortran started a collection.

By now the test suites are by far not as useful as before. The above-
mentioned test suite is also available from NAG. When I once asked
there how much it costs the answer was something like 15000 pounds (I
don't recall the exact amount). However, I have privately bought the
light version of it (http://www.fortran.com/f90tsl.html) for a
discount price (still $$$); using it I found just about half a dozen
bugs.

Additionally, *** Hendrickson has used his Shape95 Spackman &
Hendrickson Fortran test suite (also worth $$$$$) to check gfortran
and has reported 43 bugs of which 7 remain to be fixed. (Thanks!)

I would assume that with the F95 test suite of fortran.com/NAG one can
find a couple of bugs, but presumably too few that it makes sense to
spend $$$$$, especially since there is the problem how to make it
available to whom within the licensing constrains.

In any case Fortran 2003 is not covered by any test suite so far.
(Well, ignoring the private collection of the compiler vendors and the
public but implementation-tailored collection of gfortran.)

For iron out the remaining gfortran bugs, I think one has essentially
to rely on users which report bugs.

Tobias

PS: I know someone who co-developes a larger Fortran program and he
complained that Fortran compilers are buggy: With C errors are almost
certainly user errors, but for Fortran one can never rule out compiler
bugs; the program thus now contains a couple of workarounds.I have to
agree: It is not too difficult to hit a compiler error now and then -
and I have already found several bugs in NAG f95, ifort, g95,
gfortran, sunf95, ... For gfortran, one can have the hope that its
wide-spread use will help to make it harder and harder to encounter
compiler bugs; fortunately, most of the reported bugs are old bugs or
bugs in new features and not regressions, e.g. things which used to
work with older gfortran versions.
.

Quantcast