Hi @brasie,

Thanks for reporting this. Pydantic now uses speedate for date/time related parsing.

The Pydantic v1 -> v2 migration guide provides this table with information about how Pydantic converts data during validation.

The table also illustrates that when you pass a str in for a timedelta field, it's assumed to be in ISO8601 format. Taking the str of a datetime.timedelta doesn't necessarily conform to that, as seen in your example.

If you still want to be able to validate data in this format, I'd recommend a custom field validator. Here's an example I threw together quite quickly, there might be some oversights here re behavior with days / without days included in the timedelta:

import datetime
import json

from pydantic import BaseModel, field_validator


class Foo(BaseModel):
    td: datetime.timedelta

    @field_validator('td', mode='before')
    def td_to_timedelta(cls, v):
        if isinstance(v, datetime.timedelta):
            return v
        elif isinstance(v, str):
            dt = datetime.datetime.strptime(v, '%d days, %H:%M:%S') if 'days' in v else datetime.datetime.strptime(v, '%H:%M:%S')
            return datetime.timedelta(days=dt.day, hours=dt.hour, minutes=dt.minute, seconds=dt.second)
        else:
            return v


# Works
Foo.model_validate({'td': str(datetime.timedelta(hours=10))})
Foo.model_validate({'td': str(datetime.timedelta(days=3, hours=10))})
Foo.model_validate({'td': str(datetime.timedelta(days=3, hours=10, minutes=5, seconds=10))})
Foo.model_validate_json(json.dumps({'td': str(datetime.timedelta(days=3, hours=9, minutes=5, seconds=10))}))

An even cleaner approach would be to just pad the hours if a single digit it used.

Closing for now, as this is more of a question than a bug. Thanks!

I see, I suppose; Thanks for the explanation.

It's just, you're already parsing more than ISO8601, assuming my understanding is correct that ISO8601 does not include a datetime.timedelta.__str__-like [D days,] HH:MM:SS format. Pydantic V1 also supported reading a timedelta.__str__-like format, but unlike Pydantic V1, V2 no longer supports all ways that timedelta.__str__ outputs.

Hence why it seemed bug-like.

But if it is expected behavior on your side, then perhaps this is more of a feature request for speedate.

I think it makes sense to support all timedelta str representations in speedate.

In my case v1 serialised datetime.timedelta to:

"session_expiry": 10800,

and v2 serialises it to:

"session_expiry": "PT10800S",

I fixed it by:

@field_serializer("session_expiry", mode="plain")
def session_expiry_to_seconds(self, session_expiry: datetime.timedelta) -> int:
    return session_expiry.seconds

Edit:
There is a bug in my code above, check the comment bellow. @igoose1, thanks for spotting it!

@rafrafek, thanks! I would notice that your solution can give an unexpected answer as timedelta.seconds are always less than 1 day and don't include microseconds. You probably want to use .total_seconds() instead.

Example:

>>> td = datetime.timedelta(days=4, seconds=20, microseconds=69)
>>> td.seconds
20
>>> td.total_seconds()
345620.000069

So this code can be re-written in this to avoid issues:

@field_serializer("session_expiry", mode="plain")
def session_expiry_to_seconds(self, session_expiry: datetime.timedelta) -> float:  # now it returns a float
    return session_expiry.total_seconds()  # now it returns all seconds