Switched to session as parameter for the fs

[hechth]

Closes #9

[d-w-moore]

I think if using timestamp( ), we have to be careful since iRODS modify times are UTC.
Observe that running on my computer (EST timezone) I get

>>> datetime.datetime(1970,1,1,0,0).timestamp()
18000.0

But:

>>> utc=datetime.timezone(datetime.timedelta(0))
>>> datetime.datetime(1970,1,1,0,0).replace(tzinfo=utc).timestamp()
0.0

[hechth]

Ah so then it makes sense to always convert them to UTC.

[d-w-moore]

Convert in a matter of speaking yes. Although most people would take "conversion" as a translation from one timezone to another; and we're instead making a naive datetime object into a localized one of the proper timezone, ie UTC.

[d-w-moore]

If these work, I think it's a solid change - with the idea that once the session is given to the pyfilesystem object, it is owned by that object and shouldn't be used outside.

[hechth]

The main point is that otherwise returning files as is expected by the open and openbin functions becomes very difficult as the session is closed eventually before the file.

Right - the main idea being that our fs(iRODSFS) object should hold a reference to the session instance, thus preventing it from going out of scope for the lifetime of the pyfilesystem connection. It's a side benefit, I think, by virtue of your changes here, that the iRODS session may be created in any way the user desires. A side-effect is that the iRODSFS object will also need to be kept around for the open file descriptor's benefit, but this might be considered de rigeur anyway. To help enforce the idea, we might document that, as well as maybe give the iRODSFS an _irods_session_key method so that users can easily get that value for URL-style access. Or, if I'm off-base, please tell me.