Hi
Yes, thanka for this patch. I'm reluctant to consider that binaries
built on armhf failing on arm64 is an _RC Bug_, but its definitely
undesirable.
I'll forward the patch to upstream, and let them optimize it better.
regards
Alastair
On 25/01/2019 18:02, Steve Langasek wrote:
Package: adios
Followup-For: Bug #919763
User: ubuntu-de...@lists.ubuntu.com
Usertags: origin-ubuntu disco ubuntu-patch
Hi Alastair,
I'm reopening this bug report because I see that you have closed it by
disabling the testsuite.
This is a wrong fix for the issue given that the tests are correctly
identifying CPU incompatibility in the code. This only results in willingly
shipping binaries that will not work on more recent ARM CPUs (and even older
CPUs, when using a kernel in a particular mode).
If you as maintainer don't want to support adios across the range of CPUs
that the Debian armhf port is expected to work on, then it would be better
to remove the adios binaries for this architecture instead.
(And btw, this issue would also affect sparc64 as an architecture, not only
armhf, due to similar alignment constraints.)
However, I've just prepared a patch that fixes the unaligned access problems
in adios's code. Would you consider applying this as a solution instead?
It could stand to be improved - in particular, for cases where we are
byteswapping, the code currently writes the variable twice where this could
be optimized to be a single write. But I presume the byteswap case is only
of interest on big-endian architectures, so not of great concern on modern
targets.
Thanks for considering,
--
Alastair McKinstry, <alast...@sceal.ie>, <mckins...@debian.org>,
https://diaspora.sceal.ie/u/amckinstry
Misentropy: doubting that the Universe is becoming more disordered.
