[SciPy-dev] timeseries tdates.py

[Pierre GM]

On Monday 12 February 2007 14:28:41 Christopher Clarke wrote:
> Hi
> I noted that you guys want to remove the dependency on  mx.DateTime
> Any idea of where this is heading???
> I was thinking dateutil, rrule and relativedate my self

Part of the code of the timeseries package that  deals with date manipulation 
is in C to speed things up. Dateutil is nothing but a reimplementation of 
mxDateTime in pure python (if I understand correctly), which is not what we 
want. For now, it's easier for Matt and I to use mxDateTime C objects 
directly, and getting rid of the Python layer.

There are some attempts to port tdates to C via Pyrex (at least the Dates 
part, not the DateArray yet), using datetime object as the base, but that's 
gonna have to wait for a couple of weeks. We need to make sure we're not 
crashing performances by too many calls to Python at the C level. (and I need 
to polish a couple of things w/ maskedarray & tseries).

> As i am actually working on something similar
You can use datetime objects w/ the SVN timeseries package anyway. Note that 
you don't need dateutil (except for its parsing capacities). The rrules are 
completely bypassed, as we're using a quite different approach.