Re: [libvirt] [PATCH 8/9] virobject: Check if @parent is the first member in class

On Fri, Apr 13, 2018 at 04:47:15PM +0200, Michal Privoznik wrote: If a class is missing 'parent' memeber, it's a bug in the definition of the struct/class, therefore there should be a static assertion rather than a runtime check. Also, I don't like this extra parameter, which really shouldn't be needed; you created a macro which hides this parameter, but that doesn't mean that design-wise it makes sense to have it there, think of it as a constructor, you don't pass a constructor an offset of the class' member, because it shouldn't have need for it, but you do, solely for the purpose of checking whether we have a particular member in place. So, to start a discussion about this (I also think Dan posted something related to this recently, but I don't seem to be able to find it in the archives - do I even archive?!!!), I came up with my first compile-time hack ever, it seems to work like expected, but I'd like to hear your opinions both the macro itself and the approach we're going to take, so here's my replacement patch: diff --git a/src/util/virobject.h b/src/util/virobject.h index 92dd51239..2a973d401 100644 --- a/src/util/virobject.h +++ b/src/util/virobject.h @@ -75,8 +75,12 @@ virClassPtr virClassForObjectRWLockable(void); # define VIR_PARENT_REQUIRED ATTRIBUTE_NONNULL(1) # endif +# define VIR_CLASS_HAS_PARENT(name) \ + !sizeof(char[0-offsetof(name, parent)]) + # define VIR_CLASS_NEW(prnt, name) \ - virClassNew(prnt, #name, sizeof(name), name##Dispose) + VIR_CLASS_HAS_PARENT(name) ? \ + virClassNew(prnt, #name, sizeof(name), name##Dispose) : NULL Notes: - I suppose mingw would handle this hack the same way it handles VIR_TYPEMATCH, IOW it works... - it also doesn't need to be a ternary, I suggested extending VIR_CLASS_NEW to do the complete assignment in [Patch 7/9], like this: # define VIR_CLASS_NEW(prnt, name) \ if (!(name##Class) = virClassNew(prnt, #name, sizeof(name), name##Dispose)) return -1; Regards, Erik