-----BEGIN PGP SIGNED MESSAGE-----
On 02.02.2014 03:30, Activecat wrote:
Thanks. This solves the problem completely.
In summary, to stop the flowgraph when the source is empty, we need
to: 1). The source block’s work() method returns “WORK_DONE” (
which is -1 ) 2). The sink block’s work() method returns
Well, 2. is not really true.
To explain the mechanism (to understand better, maybe read about sync,
fixed ratio and general_work blocks in the wiki):
In normal operation, when a block downstream is finished with one
iteration of work, the scheduler knows how much items on its input
buffer have been consumed, determining the free space for the upstream
work(), since that work’s output is the same as the formers input
buffer, to avoid unnecessary copying.
So, if an upstream block says “hey, I’m done forever” (WORK_DONE),
then the scheduler knows that there is nothing new to do for the
It therefore continues to process the samples that are still in
buffers, until there are no blocks that could produce any output since
there is not enough input anymore.
So, in summary:
- the downstream blocks are used as long as there is still work to do.
Which of course will end fastest when they consume all their input
Additional Questions: 1). In general, shouldn’t a sink block’s
work() method always return 0 instead of “noutput_items”? (Sink
block doesn’t produce any output, isn’t the number of output
produced is zero?)
You should define your sink’s output IO signature to have an itemsize
of zero on all zero output streams. Since sinks are usually sync
blocks, there are as many output as input samples; if the former are
of size 0, then there are no buffers to shuffle around, though you can
still tell the runtime that you produced X samples, consuming X.
alternatively, return 0 and call “consume_each(X)” before.
2). In practical cases, under what circumstances we should let the
source block’s work() method return 0?
Never, unless you really can’t produce nothing. If your block is sync,
your flowgraph will start to stall.
When it returns 0, the sceduler will still continue to call the
source block’s work() method repeatly.
That sounds only right to me if downstream blocks don’t consume all
available buffer at once. The flowgraph will usually stop if a block
consumes 0 input items, since the runtime then assumes the block can’t
3). Under what scenario we will need to use the stop() method ? (In
python script we see stop() method quit often, but it seems never
needed in c++ code)
Wrong perspective: In Flowgraph code you often see start() stop()
since it is used to manage flowgraphs from a higher view.
inside work functions, you should never call stop(). Just say you’re
WORK_DONE, since you can’t say anything about the work in progress at
the rest of the flowgraph. Let the runtime manage shutting down the FG.
If your FG doesn’t contain message passing blocks (not even inside
hier blocks), the python main program will stop when it’s done, I’m
fairly certain. Can you share more about your flowgraph?
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
-----END PGP SIGNATURE-----