Re: [Discuss-gnuradio] Stream to PDU - Guaranteed to not lump tags?

[Dave NotTelling]

Michael,

     Thanks for the response!  Apologies for the unclear question.  Here's another shot:

Can one call to the work function yield multiple offsets from the `get_tags_in_range(0, nitems_read(0), nitems_read(0) + noutput_items)` call?  With multiple offsets then that means there are multiple sets of PDUs (assuming that the stream was created from pdu_to_tagged_stream).  Let's say for fun that there are 3 PDUs like the following (CAR, CDR):

• ( {'pdu_len': 10, 'offset': 0, 'frame' : 1}, (0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00) )
  
• ( {'pdu_len': 10, 'offset': 10, 'frame' : 2}, (0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00) )
  
• ( {'pdu_len': 10, 'offset': 11, 'frame' : 3}, (0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00) )
  

What I'm curious about is whether or not a single call to work() will end up with tags for more than one of the PDUs above.  For instance, if the buffer for the block is large enough, would 20 samples get lumped in to a single call to work() resulting in say offset 0 and offset 10 both showing up in 'get_tags_in_range(0, nitems_read(0), nitems_read(0) + noutput_items)'?  This of course assumes that pdu_to_tagged_stream and tagged_stream_to_pdu both are set to use 'pdu_len' as the length tag.

     I seem to remember having that very issue once.  I got multiple offset values in a work() function in the past.  Pretty sure I did at least O.o  Really, just curious if there is a guarantee somewhere deeper in the block code that ensures that can't happen. 

     As I write this I realize that this is a pretty easy thing to test out.  If it's still unclear, then I'll just do that and post results :)

Thanks!

On Wed, Jan 10, 2018 at 3:32 PM, Michael Dickens <address@hidden> wrote:

Hi Dave - The tagged stream -> PDU block will look at for the provided tag string at the current 0 offset & if found then that's what the PDU data size (in items) will be [this is actually handled in the "tagged_stream_block" parent class]. If there are interim tags (within the output PDU size in items), then they are just added as meta-data to the PDU. Each PDU is created independent of the other PDUs, and just 1 created per call to "work". Not sure if this is what you were looking to have answered; if not, please clarity. Cheers! - MLD

On Wed, Jan 10, 2018, at 11:30 AM, Dave NotTelling wrote:
> The wording of the title likely needs work, but the basic idea is this:
>
>  * Suppose that I have a ZMQ message source that has arbitrarily sized vectors of some consistent type
>  * I convert that over to a tagged stream, do some operations on it, then convert it back to a PDU
>  * Assume for a second that several of the offsets in the tagged stream are very small (only say 10 samples each) and back to back
>  * Is there a guarantee that each of the 10 sample offsets will result in a single PDU?
>
>Looking at the source of tagged_stream_to_pdu it seems that there is no check that multiple tag offsets arrived in a single call to work.  Or is that something that cannot happen?  Having a contract that no more than one offset can arrive at a single call to work would be really nice, but I imagine that doing so could cause some pretty serious performance issues.