On Tue, Apr 23, 2019 at 9:44 AM Adrien Destugues
<pulkomandy@xxxxxxxxxxxxx> wrote:
This should be in Gerrit discussions, and if the issue is non-obvious,
possibly lead to a comment
in the code. Now it's lost in IRC logs which are not easily related to the
code in question.
Admittedly there is still a problem that this depends on internal FIFO
behavior on destruction, and
there is still a small race related to buffer management. How to resolve
that, I don't really know
as it would probably require a much more substantial rework of this code.
This change is at least
much more correct and should stop the KDLs people have been getting.
So jessicah is right, this is still incorrect, even if it is less likely to
fail.
You cannot safely test if an object has been deleted, because another object
of the same type may be
allocated at the same address (bad luck, but it can happen).