On Thu, Oct 20, 2016 at 2:58 PM, Sage Weil <sweil@xxxxxxxxxx> wrote: > On Thu, 20 Oct 2016, John Spray wrote: >> Latest master has the MDS crashing on startup: >> http://tracker.ceph.com/issues/17636 >> https://github.com/ceph/ceph/pull/11577 >> >> My patch is kind of dumb, clearly the intent of the original code is >> to avoid any constructor overhead and it would be nice to do so, but >> the original code seems to make the assumption that no encode/decode >> implementations deferences the object they're passed. >> >> Good motivation to go and immediately familiarise myself with the new >> code this morning though :-) > > Hmm, what compiler version are you using? The intent of the code is > two-fold: avoid the ctor, as you mention, but also trigger a SEGV if the > implementation of bound_encode for any traits::bounded type actually > makes use of the (bogus) reference we're passing in. This is gcc 5.3.1 (fedora 22). > Which type(s) did it fail on? My guess is that it is either a malformed > bound_encode (taking a value instead of a reference) or the implementation > is broken (value-dependent). Ah, I see that denc_traits<inodeno_t>::bound_encode was indeed taking a value instead of a reference. John > Or, we are unlucky enough that the compiler is both turning const foo& > into a pass-by-value and not optimizing it away when the implementation > doesn't make use of it. :/ Hopefully not! > > sage > -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
