Maybe use ISO8601 as input format?

[jgsogo]

(Thinking about machines and importing data)

There is an standard about how to express durations, ISO8601 [1], it would be interesting to allow this format as input, although maintain as output both, ISO8601 and the more readable one (current).

[1] http://en.wikipedia.org/wiki/ISO_8601#Durations

[johnpaulett]

I was unaware of the ISO 8601 Duration specification. A patch that added support for IS0 8601 and related tests would be very welcomed. Probably involves altering str_to_timedelta.

For backwards compatibility, we would want to still support the existing duration syntax.

[jgsogo]

It seems that isodate [1] or dateutil [2] should do the trick, working with strptime would be so hazardous. I'll try to work on it this week

[1] https://pypi.python.org/pypi/isodate [2] http://labix.org/python-dateutil

[codingjoe]

I looked into it, as I desperately need an implementation of ISO 8601 durations. Sadly you decided to work with pythons datetime.timedelta which is a poor implementation of a duration.

The only way I see to make this happening is, changing your to_python to an object that inherits from timedelta but implements a isoformat() as datetime does it self. This would be downwards compatible, but has one big disadvantage.

timedelta is not a duration. A duration can't be negative by definition, a timedelta can. The cleaner approach would be implementing a real duration and using an UNSIGNED BIGINT your database.

I would be will to implement this if you'd be cool with that little change.

[johnpaulett]

@codingjoe -- I'm trying to understand your use case. Are you looking for:

1. A means to pass in a IS0 8601 Duration into a DurationField (which the original issue proposes)?
2. A way to get a timedelta formatted into an ISO 8601 Duration string?
3. Altering the way this package stores the data in the database?
4. Something else?

I would completely welcome a patch for the first item. I think the leading "P" of an ISO8601 duration should make detecting it straightforward in str_to_timedelta.

The second item may be currently accomplishable with the isodate library (duration_isoformat can take a timedelta).

I won't argue that perhaps a better name for this library currently is IntervalField (essentially it provides PostgreSQL's INTERVAL across various database backends) or TimedeltaField. However, I believe using form & model validators you could enforce only positive durations using this library currently.

Definitely some opportunity for improvement, but I'm just trying to grasp exactly what you are looking for.

[codingjoe]

@johnpaulett sorry for my confusing post.

Here's what I'm up to: I currently writing a django implementation of mircodata objects as defined on http://schema.org/

That requires a ISO duration. I know that integers are more practical for machine use, but they suck as a service to service interface. I find it a little sad, that durations aren't implemented in python itself.

I didn't know that this library intends to implement a PostgreSQL Interval. PostgreSQL has a confusing naming there as well. In ISO an interval is DateTime+Duration.

I think I'm going to write a patch for 2. if that's cool with you. It looks like 1. is already implemented in str_to_timedelta

[johnpaulett]

To be clear, 1 is not available in str_to_timedelta currently, but I would welcome a patch for it.

Regarding 2, would have to see code a bit more, not 100% sure if this is something this library should provide or if a more widely applicable item for all timedeltas in Python.

[codingjoe]

It's not my number one priority. It's more a long shot, of a library want to develop to maybe use it later commercially. In other words it's gonna take me a while. But I'll keep you posted on developments and i'll supply patches if need be.

[jnelson]

@johnpaulett Do you have a preference between aniso8601 and isodate as supporting lib for a PR on this issue?

[johnpaulett]

@jnelson I would steer away from including a GPL-licensed dependency (aniso8601) for this BSD-licensed library.

[jnelson]

Sorry I didn't catch that. Looks like isodate is also BSD, so that's easy. Thanks!