
16 Jun
2020
16 Jun
'20
2:14 p.m.
On Tue, Jun 16, 2020 at 08:07:10 -0400, John Ferlan wrote:
Since 1e2ae2e31, changes to use the automagic free logic didn't take into account that one path uses posix_memalign and the other uses VIR_ALLOC_N - the former requires using VIR_FREE() and not g_free() to free the memory.
I don't really think this makes sense. VIR_FREE is now using g_clear_pointer(&PTR, g_free) internally, so g_free will be used regardless. Is the problem really in g_free or rather some kind of problem with the __attribute__(cleanup)) feature? Additionally this fix looks weird. I'd prefer if autofree is dropped here and VIR_FREE is used directly in both cases if this in fact fixes the problem as there is no reason to keep g_autofree just in one case.