Re: [tauruslib-devel] odo-compatible URIs (Was: Re: URIs)

[V. A. S. <so...@es...>]

On 27/06/2016 10:30, Carlos Pascual wrote:
> Hi Armando,
>
> I hope you do not mind me posting my response to the taurus mailing list
> since I think this can be an interesting topic for discussion with all
> the developers of Taurus. See my reply below.
>
> On Wed 22 June 2016 15:07:51 V. Armando Solé wrote:
>> Carlos,
>>
>> I have just seen your slides. Concerning URIs, can you try to be
>> compatible with pydata URIs behore Taurus4 is already deployed? We
>> intend to follow their way in order not to reinvent things.
>>
>> http://odo.pydata.org/en/latest/uri.html
>>
>> Best regards,
>>
>> Armando
>
> I did not know odo. Thanks for pointing to it.
>
> I will look into it later in more detail, but on a first look, it seems
> that there are no collisions with the *existing* URIs in Taurus. So it
> definitely seems a good idea to try to follow their conventions for the
> new schemes.
>
> BUT I am worried because while in Taurus TEP3 [1] we explicitly made
> changes to enforce  RFC3986 [2], odo / PyData [3] seems not tho comply
> with it:
>
> - odo uses "://" for specifying protocols. RFC3986 states that the
> protocol (scheme name) is separated by ":" and the "//" is actually a
> prefix for the authority part, not a separator. In practice this implies
> that if no authority is given, there must not be "//". So the "URI"
> `hdfstore://myfile.h5::/data/path` is not RFC3986-compliant (I am
> assuming here that "myfile.h5" is part of the URI path, not the
> "authority").

What do you want me to say? I do not know who cooked RFC3986, but people 
are so used to "://" following http that it seems more natural to 
specify a protocol considering the use we intend to give to the URIs:

We are not considering to use odo at this point. We were simply looking 
for a way to adapt our scripts using files as input to the use of 
particular datasets into HDF5 files. So, for the time being we'll be 
satisfied by myfile.h5::/path/to/dataset and we are looking for a way to 
specify a particular frame of a multiframe dataset. I would not be 
surprised to end up choosing something between 
myfile.h5::/path/to/dataset#10 (html like) or 
myfile.h5::/path/to/dataset::10 (odo like)

Armando