On 7/9/2012 10:07 AM, Anders Langlands wrote:
attach() and detach() are what I was using before, but they never get
called for the 'transient' Ops that are created during updates. To be
clear about what's happening here, if I'm sending data from my server
(which in turn receives data from a renderer) to my Op, I'll be
calling asapUpdate() a lot. When this happens, Nuke sometimes creates
an Op (or several) to render the image. This "transient" Op gets
_validate called, but does not call attach() or detach().
attach() and detach() therefore work fine for opening/closing my
server connection for the firstOp(), but not for the transient Ops,
which is why I tried using the ctor/dtor instead, which is how I ended
up worrying about why the dtor is not always called...
Perhaps the server could implement a timer that cleans up shared mem
(resets it to zero size) after a while if nothing calls it, and reallocs
it if something calls it again
if (region.get_size()==0) realloc ...
void * addr = region.get_address();
Seems maybe I can't win? :)
-----------------------
Anders Langlands
x8382/+447789206593
------------------------------------------------------------------------
*From:* Steve Booth [[email protected]]
*Sent:* 09 July 2012 17:16
*To:* 'Nuke plug-in development discussion'
*Subject:* RE: [Nuke-dev] Op destructor does not get called?
Gotcha. The _close solution is going to be problematic because _close
is called after every render, not just when the node is being removed.
How about the 'detach()' method of Op? If your node is detached (which
surely happens when it's deleted), then you could safely remove the
shared memory?
Steve
*From:*[email protected]
[mailto:[email protected]] *On Behalf Of
*Anders Langlands
*Sent:* Monday, July 09, 2012 9:04 AM
*To:* Nuke plug-in development discussion
*Subject:* RE: [Nuke-dev] Op destructor does not get called?
I'm on linux right now. It's not just on termination that I want to
clean up my shared memory though - imagine if the user has a nuke open
for several days, creating and deleting nodes, then they're going to
rack up a huge chunk of shared memory that is not actually being used.
Overriding the _close() method looks kinda promising - it seems like
it's always called when the user deletes a node (I'd have to handle
rehooking up the resources in case of an undo I imagine, but that
shouldn't be too hard), and it looks like it gets called after the
given timeout for the 'transient' nodes that sometimes get created
when I'm spamming asapUpdate(). I'm seeing some strange behaviour
though so this needs further investigation... ultimately neither of
these things seems like a great solution.
-----------------------
Anders Langlands
x8382/+447789206593
------------------------------------------------------------------------
*From:*Steve Booth [[email protected]]
*Sent:* 09 July 2012 16:28
*To:* 'Nuke plug-in development discussion'
*Subject:* RE: [Nuke-dev] Op destructor does not get called?
Which OS are you using? Windows, Linux, or OSx? There are
process-termination hooks that you can utilize in each OS.
*From:*[email protected]
<mailto:[email protected]>
[mailto:[email protected]] *On Behalf Of
*Anders Langlands
*Sent:* Monday, July 09, 2012 7:46 AM
*To:* Nuke plug-in development discussion
*Subject:* RE: [Nuke-dev] Op destructor does not get called?
Hi Steve, my buffers are allocated in shared memory by a server
process that then sends data to Nuke. If I don't tell the server that
my Op is no longer using the shared memory, then the server process
will still hold a reference to the shm and it will never be freed (at
least I think so...). What I think I need therefore is to have a hook
that's guaranteed to be called when Nuke closes so I can release the
resource.
-----------------------
Anders Langlands
x8382/+447789206593
------------------------------------------------------------------------
*From:*Steve Booth [[email protected]]
*Sent:* 09 July 2012 15:42
*To:* Nuke plug-in development discussion
*Subject:* Re: [Nuke-dev] Op destructor does not get called?
Also note, Anders, that nothing is left hanging. Even though your
destructors are not called, all heap-allocated space associated with a
process is automatically freed when the process terminates. So, when
you exit Nuke, you get your memory back.
Steve
Sent from my iPad
On Jul 9, 2012, at 5:55 AM, Anders Langlands
<[email protected] <mailto:[email protected]>> wrote:
I'm finding that Nuke will sometimes destroy my Op, and sometimes
not. If I create half a dozen plugin nodes then delete them, maybe
2 will call MyOp::~MyOp(). Then when I quit Nuke, those
destructors still won't be called.
In my particular plugin I allocate ~25MB of interprocess shared
memory per Op, so leaving this hanging around is a bit of a deal.
Is this something to do with the undo functionality or something
else that I can turn off? Is there some other hook I can use that
*is* guaranteed to be called when an Op is no longer needed?
Cheers,
Anders
-----------------------
Anders Langlands
x8382/+447789206593
_______________________________________________
Nuke-dev mailing list
[email protected]
<mailto:[email protected]>,
http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
_______________________________________________
Nuke-dev mailing list
[email protected], http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
_______________________________________________
Nuke-dev mailing list
[email protected], http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
