On Sun, Nov 30, 2014 at 07:17:46PM +0100, gregor herrmann wrote: > On Sun, 30 Nov 2014 17:36:04 +0100, Alessandro Ghedini wrote: > > > On dom, nov 30, 2014 at 03:06:55 +0100, Tobias Frost wrote: > > > Am Sonntag, den 30.11.2014, 00:21 -0800 schrieb Tom Lee: > > > > > Also, I feel like the "serious" severity is overstating the issue > > > > given that 0.11.0-4 builds fine in buildd/sbuild. Alessandro pointed > > > > out the periodic rebuilds would have revealed this issue otherwise. > > > > > > > > > > > If there are no objections I'd like to propose we adjust the severity > > > > of this bug to "normal" & leave the fix for this particular bug until > > > > after the Jessie freeze. > > > > > > Here I can reprodcue the FTBFS locally with pbuilder 0.215+nmu3, so > > > I disagree. It maybe has not been detected *yet*? > > > > What does this "yet" even mean? Except inside pbuilder, hiredis builds fine > > [1]. > > The fact that it fails *only* inside pbuilder (and the fact that hiredis is > > not > > the only package in this situation) suggests that this is indeed a pbuilder > > bug. > > I really don't see how this is release critical in any way on the part of > > the > > hiredis package. > > While I tend to agree in general, here's an additional data point: > I rebuilt 0.11.0-4 in my sid amd64 cowbuilder chroot, which has > USENETWORK=yes (due to #753944) but firewalls off everything except > localhost during build. And in this environment I see a test failure:
Ok, so I think the whole problem is that /etc/resolv.conf lists non-local name servers, but the build environment can't actually reach them (because of USENETWORK=no or the firewall). This is, I think, the exact same problem as #759799 (which is btw severity: important). If the consensus is that this should be fixed in the affected packages (e.g. by disabling the tests), I'm all for it, but I really think that the effort should go into fixing pbuilder, since who knows how many packages are actually affected by this. A simple and stupid solution would be to turn off DNS name lookups completely inside the build environment if USENETWORK=no (e.g. by fiddling with /etc/nsswitch.conf), but I'm not sure if that's really applicable to pbuilder. > Not sure what the best way forward is; adding a test for "Temporary > failure in name resolution" might be an option (and works > unsurprisingly): > > #v+ > --- a/test.c > +++ b/test.c > @@ -286,7 +286,8 @@ > c = redisConnect((char*)"idontexist.local", 6379); > test_cond(c->err == REDIS_ERR_OTHER && > (strcmp(c->errstr,"Name or service not known") == 0 || > - strcmp(c->errstr,"Can't resolve: idontexist.local") == 0)); > + strcmp(c->errstr,"Can't resolve: idontexist.local") == 0 || > + strcmp(c->errstr,"Temporary failure in name resolution") == 0)); > redisFree(c); > > /*test("Returns error when the port is not open: "); > #v- > > But maybe there are better ways to fix this. That would make the test kinda useless, but I guess it's no worse than disabling it completely. Cheers
signature.asc
Description: Digital signature