Package: gnupg2
Version: 2.0.26-1
Severity: normal
gnupg2 fails some tests:
here is the log
http://buildd.debian-ports.org/status/fetch.php?pkg=gnupg2&arch=arm64&ver=2.0.26-1&stamp=1408548814
and here is a snippet from the end:
FAIL: genkey1024.test
FAIL: conventional.test
> IDEA FAIL: conventional-mdc.test
This can be reproduced with:
( cd tests/openpgp/ && make check TESTS=genkey1024.test )
It took quite some work to track down this issue, which was done by Edmund
Grimley-Evans.
The tests that were failing involved gpg-agent. The problem was not
reproducible on Ubuntu Trusty.
Building the software on Trusty (where it worked OK), then copying the
binaries over to Debian, it still didn't work, suggesting an issue
in the environment/libraries, not the code itself.
Further debugging eventually tracked this down to a problem in glibc,
corrupting the stack.
Edmund wrote:
--------
"My test software suggested that registers d8, d10, d12, d14 are
restored correctly, but d9, d11, d13, d15 are not.
I believe this is the bug that is preventing gnupg2 from working,
because I have observed pth_select_ev +corrupting those odd-numbered
registers, and libpth20 does use getcontext, setcontext, ...
This bug may have been fixed by Will Newton's patch in April, which is
mentioned in Debian's changelog:
http://metadata.ftp-master.debian.org/changelogs//main/g/glibc/glibc_2.19-9_changelog
If I've understood correctly (please help!) this patch is in Debian's
source package +(debian/patches/arm64/submitted-setcontext.diff) but
is only applied in "experimental". "
I'll file another bug on glibc for this, and block on that so that we
know when gnupg2 can pass its tests on this arch. There isn't actually
a fault in gnupg2 itself, so far as we know.
--
Wookey
--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]
