Re: [Discuss-gnuradio] File formats, some header data per frame

[Eric Blossom]

On Fri, Apr 01, 2005 at 02:06:53PM -0500, James Cooley wrote:
> Hi all,
> 
> There was some talk a few weeks back on creating a file format in which 
> there's a header storing info like the sample rate, capture freq, 
> decimation, etc.
> 
> I would like to create such a file, but would like to store some small 
> amount of data once per work cycle. The idea is to constantly record a 
> changing freq and other data and be able to play it back later.
> 
> I followed the model of file_sink/file_source and designed format like 
> this:
> 
> <int> length of header stuff (could be 0)
>  <int> header ints
> .... The sample data
> 
> <int> length of header stuff (could be 0)
>  <int> header ints
> .... The sample data
> 
> <int> length of header stuff (could be 0)
>  <int> header ints
> .... The sample data
> 
> So far this doesn't work because of my assumption of the sample data 
> size. Am I assuming correctly that sample data (n_input_items) items is 
> a fixed size?

No.  Not a safe assumption.  The only thing your work routine can
count on is that it will be called with noutput_items = k * output_multiple()
for some integer k.

> How is that determined in the flow graph? 

It's an implementation detail that is subject to change depending on
the implementation of the scheduler.

> Will it be the same when I go back to read in the file? (I think
> not, this is why my code fails, it assumes that it will find my
> header info (size, header) at the start of each work cycle when
> reading in the file).

Unlikely ;-)

I think you can get what you want by including the length of each
chunk of data in the file.

> <int> length of header stuff (could be 0)
>  <int> header ints
>  <int> number of samples in this chunk
> .... The sample data

> <int> length of header stuff (could be 0)
>  <int> header ints
>  <int> number of samples in this chunk
> .... The sample data

Eric