Fwd: Re: testsuite failure

[Tobias Burnus]

I accidentally only replied to Steve and not to the list. My email is 
below. In a reply he confirmed my suspicion: He recently enabled the 
debugging capabilities of FreeBSD's jemalloc, which fills the malloced 
(and freed) memory with random bits.

Tobias

On 08/10/2012 06:55 PM, Steve Kargl wrote:
> Has anyone reported a testsuite failure with realloc_on_assign_5.f03?

I did so in February 2010: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47674
And I still see the issue.

> It seems someone may have broken deferred character length scalars.

Do you have something like malloc perturb? i.e. a malloc which doesn't
return nullified memory but something else? If I recall correctly, the
issue only surfaces on Linux/GLIBC with MALLOC_PERTURB_ being set.

If I recall correctly, the problem is that the dependency machinery
doesn't realize that for "a = a(1:3)" a temporary is needed. (Well,
here, a simple realloc would do as well.)

Tobias