Jeff King <peff@xxxxxxxx> writes: > On Sat, May 10, 2014 at 10:02:59AM -0400, Jeff King wrote: > >> On Fri, May 09, 2014 at 02:08:27PM -0700, Junio C Hamano wrote: >> >> > > 3. Just disable the http tests when run as root. >> > > >> > > I think I'd favor 3. But I'd like to be sure that being root is the >> > > problem. >> > >> > I agree with both the conclusion and the precondition. >> >> Here's the patch. >> >> The problem starts in v1.9.2, not in v2.0.0, so it's not technically a >> regression in this cycle. And we are awfully late in the -rc period. But >> it is just a change in the test script, and one that seems rather >> unlikely to produce unexpected side effects. I'll leave it you whether >> you want to queue it for v2.0.0, or for the next maint release. > > Hrm, sorry, I was wrong about this. I had thought the > auto-network-testing had gone into v1.9.2, but it didn't. So this _is_ a > potential regression in v2.0.0. It's still relatively minor, affecting > only the test suite, so it can probably wait for post-v2.0.0 if you > don't want to do another -rc. A bit more disturbing is that I did not get the impression that we know the exact reason why these http tests, especially the getwpuid call, fail for Fabio when run as root. And if my impression is correct, then "do not run tests as root" applied as a "solution" to the failure report, would merely be sweeping the problem under the rug, even though it is a very good advice in general. Is it a bug in the server itself that it fails to do getpwuid, or is it because the system Fabio's on is somehow screwed up? Until the latter can be ruled out, we might be better off not doing anything, which may give interested parties an easier way to dig deeper into the real cause of getpwuid failing, no? Such a digging may even result in a better solution (e.g. finding a specific pattern of misconfigured systems and stop tests only on them). Personally, I do not think running our tests as root is an interesting enough problem to warrant the effort from us to dig only to come up with such a "better solution", and I would be perfectly happy to apply the "do not run this test as root" patch, or even a broader "we do not let any test run as root, unless individual tests explicitly ask us to allow it" somewhere in test-lib.sh included by everybody. That may sweep the issue under the rug as a side effect, but it is OK because it is not the primary mission of our tests to find issues in either httpd binaries or the system configuration that may cause the httpd server misbehave in the first place. So... -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html