avahi-daemon exhibits increasing memory consumption due to libdbus-1 FIXME

[Thomas Fitzsimmons]

Hi Simon,

Simon McVittie <simon.mcvittie at collabora.co.uk> writes:

> On 29/01/13 18:04, Thomas Fitzsimmons wrote:
>> For example on my machine (avahi-0.6.31,
>> dbus-1.4.16), after avahi-daemon has reached steady-state:
>>
>> Would the D-Bus maintainers be open to a patch to fix this long-standing
>> FIXME?
>
> I would be happy to receive a patch: please open a bug on
> bugs.freedesktop.org under dbus/core.

I filed a bug report and attached a patch for review:

https://bugs.freedesktop.org/show_bug.cgi?id=60176

> Depending how large and complicated the patch is, and how likely to
> cause regressions it appears to be, I might only be willing to apply it
> to master (dbus 1.7), or I might also be willing to apply it to the
> current stable branch, dbus 1.6. I am unlikely to apply it to the 1.4
> old-stable branch, which is only supported for security fixes.

OK.  I noticed that two public functions can return references to
childless unregistered nodes.  So this patch will subtly change the
semantics of the public API, but in ways that consumers of libdbus-1 may
not be relying on.

For example, dbus_connection_list_registered lists childless
unregistered nodes.  With this patch, existence checks for dangling
parents will fail, where before they would succeed.  I didn't survey
usage patterns to see if anyone is doing that.

Likewise, when dbus_connection_dispatch called the default Introspect
handler (handle_default_introspect_and_unlock), it used to return
dangling parents in its results list, but with the patch it won't.

> Please note that the current 1.4 release is 1.4.24. You should only use
> 1.4.16 if your distribution provides backported security fixes.

OK, thanks.  We maintain our own distribution based on OpenEmbedded.
We're in the process of doing an update that should include a newer
D-Bus release.

Thomas