Re: bug in make check

[John Ketchum <johnk at qualcomm dot com>]

According to this

http://www.bero.org/gcc296.html

2.96 is actually less buggy than 2.95, and  things that are attributed to 
bugs in 2.96 are actually due to more strict adherence to standards.  In 
particular, one of the examples of broken code fragments given that have 
stopped working under more recent versions of gcc due to more strict 
standards adherence is

struct a {
         int a;
         int b;
};
[...]
struct a *s=(struct a *)malloc(sizeof(struct a));
memset(s + sizeof(int), 0, sizeof(int));
assert(s.b==0);

This looks suspiciously like the structure used for complex numbers in 
gsl.  Is it possible that somewhere in the hermitian code, there are 
references by computed offsets rather than explicit reference to the 
relevant structure element, and which are not strictly standards compliant 
according to the above?

At 09:32 PM 4/10/2002 +0100, Brian Gough wrote:
>Juergen J. Zach writes:
>  > Contrary to the claim:
>  >  " 2) With gcc-2.95/2.96 the tests fail in the eigen/ directory.
>  > This is due to a compiler optimization bug which causes errors in
>  > the manipulation of complex numbers.  Do not use the library if you
>  > encounter this problem.
>  >  This is fixed in more recent versions of gcc.  Compiling without
>  > optimisation will work around the bug.  ",
>  >  the same thing still happens with gcc 3.0.4. Any hints as to what
>  > might be done about that?
>
>For reference, what platform are you using?
>
>Brian

John Ketchum
Qualcomm Inc.
9 Damonmill Square Suite 2A
Concord, MA 01742
Office: 60915
Direct: 781-276-0915
Fax: 781-276-0901
Mobile: 617-312-7023