[PATCH v2 wayland 00/11] Stop leaking file descriptors
daniel at fooishbar.org
Mon Dec 4 22:34:48 UTC 2017
On 13 April 2017 at 17:51, Derek Foreman <derekf at osg.samsung.com> wrote:
> Moved the test cases to the end so they're not introduced in a failed
> Reworked the removal of the global zombie singleton patch - we now
> create a wl_zombie at proxy creation time and store the number of
> fds for each opcode in that. If no signature requires fds we just
> use a NULL pointer to avoid useless malloc/free. At client
> disconnect we do a map walk to inter any zombie left behind and
> make sure we don't leak memory.
> Thanks to Pekka and Jonas for suggestions on how to handle this
> without breaking API as I'd previously done.
> To perform the map walk I added a new patch that changes the
> wl_map_for_each() function to provide the entry flags - we can't
> look up flags from the iterator callback without dereferencing the
> pointer, and with my changes we don't know what the data type stored
> in that map location is. This only changes internal callers - the
> publicly visible map walk function still acts as before.
Thanks for this. It looks fairly good to me so far. I've pushed 1, 5
and 9 as trivially correct. As said in comments to 4, I feel like a
slightly improved 4 could also replace 2 and 3 at the same time. 6, 8
and 10 all have my R-b, though a comment as to why three is the
correct number of roundtrips wouldn't go astray on 10 (it seems right
to me, mind). Unfortunately I'm not getting far into 11 without my
eyes glazing over, and I haven't given 7 a pass over for actual
lifetime handling yet, but that at least gets my A-b.
More information about the wayland-devel