On Fri, Jan 17, 2014 at 03:30:49AM +0100, Andreas Beckmann wrote:
> pdl FTBFS with testsuite failures on several architectures
These testsuite failures were actually there before, but I only now made a
successful testsuite run _mandatory_ in the build process...

> (are they all big-endian architectures?):
powerpc, mips, s390x, sparc are indeed big-endian.
debian's ia64 port AFAIK uses little-endian mode.

Besides the potential endianness issues, there is a bug hidden in the depths
of PDL.
To summarize [1]:

In a couple of locations within the core, there is an assumption that
"double" is the largest datatype and some return values are casted/converted
to "double". "long long int" is indeed longer than the mantissa of a
"double", so when the conversion happens and the value of the longlong is
large, the least significant bits are lost.
The address space layout on ia64 leads to large pointer addresses, leading to
segfaults during the test suite run.
In some sense, it is just by luck that the same does not happen to pointers
on other 64bit architectures.
And, as PDL is a language used in the context of numerical computation (think
of it like MatLab/NumPy for perl), this implicit precision loss may also
affect real calculations performed with it.

[1] http://thread.gmane.org/gmane.comp.lang.perl.pdl.devel/5637/focus=5641

-- 
c u
henning


-- 
To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org

Reply via email to