Re: Socket functions from octave

[John Swensen]

Tom Holroyd (NIH/NIMH) [E] wrote:
> John Swensen wrote:
>>  * Only allows the sending of uint8 arrays or strings (This could be
>> changed to detect the size and perform endian swapping to network byte
>> order)
>>  * Only receives data into a uint8 RowVector
>
> Wouldn't it be better to implement as low-level an interface as
> possible, and then let the formatting of data be done by other libraries?
>
> Ideally, it would be nice to have an interface that allowed arbitrary
> Octave objects to be sent transparently.  This could be written on top
> of a low-level socket layer that took care of the packaging and so
> on.  There are actually several libraries/protocols that do similar
> things; libaxl packages content in xml and transfers it, but that's
> meant for web services. We want something similar but for things like
> HDF5 content.
>
> So a lower level socket interface should allow reading and writing of
> bytes, basically, and that's it.  A higher level API could then use
> that to package Octave objects and send them?
>
That is kindof what I was thinking.  That is why I only supported uint8
arrays and strings, since both are in essence byte arrays.  Then let the
user or wrappers on top of these simple byte-array function munge it
into whatever format is needed.  I will finish implimenting the
low-level socket function, then we can decide what wrappers (e.g. for
HDF5 content) you want written on top of the low-level functions.

Thanks for the input.

John Swensen