
On Wed, Nov 02, 2011 at 02:19:07PM -0600, Eric Blake wrote:
On 11/02/2011 12:57 PM, Guido Günther wrote:
Built fine on most Debian architectures:
https://buildd.debian.org/status/package.php?p=libvirt&suite=experimental
The built failure on amd64 is due to virnetsockettest failing with:
5) Socket UNIX Accept... libvir: RPC error : Path /build/buildd-libvirt_0.9.7~rc1-1-amd64-EGXZTE/libvirt-0.9.7~rc1/debian/build/tests/virnetsockettest-test.sock too long for unix socket: Cannot allocate memory
since the socket path doesn't fit in UNIX_PATH_MAX. Since exceeding the path shouldn't't be fatal I'm using the attached patch.
From: =?UTF-8?q?Guido=20G=C3=BCnther?=<agx@sigxcpu.org> Date: Wed, 2 Nov 2011 19:02:42 +0100 Subject: Skip socket test if we exceed UNIX_PATH_MAX.
As seen on the amd64 buildd with:
s/buildd/build/
This one was actually intentional. It's the build daemon, short buildd.
+++ b/tests/virnetsockettest.c @@ -205,11 +205,13 @@ static int testSocketUNIXAccept(const void *data ATTRIBUTE_UNUSED) if (progname[0] == '/') { if (virAsprintf(&path, "%s-test.sock", progname)< 0) { virReportOOMError(); + ret = EXIT_AM_SKIP; goto cleanup; }
Wrong place to be checking - virAsprintf() only fails on OOM (malloc failure), not on size fit, so your skip is unlikely to be hit. I agree that it's okay to skip the test if run in a subdirectory so deep that a Unix socket cannot be created with a name that long, but that should be done strlen() check just before virNetSocketNewListenUNIX, not by looking for malloc failure. And since we're most likely not out of memory, that may mean we also have a bug to fix in our error reporting quality within virNetSocketNewListenUNIX.
Yes - the analysis was correct, the patch was horribly broken though. New versions attached. I wen't for a tempdir though to make sure the test can succeed. -- Guido