nautechsystems/nautilus_trader · error · TypeError

value must be datetime-like

Error message

value must be datetime-like

What it means

The pandas-less fallback of dt_to_unix_nanos only understands int (raw ns), ISO strings, and datetime instances; anything else — float, date, numpy datetime64, pd.Timestamp-without-pandas — reaches the terminal 'raise TypeError("value must be datetime-like")'. With pandas installed the same inputs are instead funneled through pd.Timestamp(value), which accepts a wider set.

Source

Thrown at python/nautilus_trader/core/datetime.py:91

def dt_to_unix_nanos(value: Any) -> int:
    """
    Return the UNIX timestamp in nanoseconds for the given datetime-like value.
    """
    if value is None:
        raise ValueError("value must not be None")

    try:
        import pandas as pd
    except ImportError:
        if isinstance(value, int):
            return value
        if isinstance(value, str):
            if _has_more_than_microsecond_precision(value):
                raise ValueError("pandas is required for nanosecond-precision datetimes") from None
            value = datetime.fromisoformat(value)
        if isinstance(value, datetime):
            return _datetime_to_unix_nanos(value)
        raise TypeError("value must be datetime-like") from None

    if isinstance(value, pd.Timestamp):
        return int(value.value)

    return int(pd.Timestamp(value).value)


def _has_more_than_microsecond_precision(value: str) -> bool:
    _, separator, remainder = value.partition(".")
    if not separator:
        return False

    digits = 0

    for char in remainder:
        if not char.isdigit():
            break
        digits += 1

View on GitHub (pinned to a4b06ed870)

Solutions

1. Convert before calling: dt_to_unix_nanos(datetime(2024, 1, 1, tzinfo=timezone.utc)) or pass int ns
2. Coerce floats to int ns explicitly (int(1_700_000_000.5 * 1e9)) if that is the true unit
3. Install pandas to broaden accepted input types via pd.Timestamp

Example fix

# before (no pandas installed)
dt_to_unix_nanos(1_700_000_000.5)  # TypeError: value must be datetime-like

# after
dt_to_unix_nanos(int(1_700_000_000.5 * 1e9))
# or
dt_to_unix_nanos(datetime.fromtimestamp(1_700_000_000.5, tz=timezone.utc))

from datetime import datetime, date

if not isinstance(value, (int, str, datetime)):
    if isinstance(value, date):
        value = datetime(value.year, value.month, value.day)
    elif isinstance(value, float):
        value = int(value)
    else:
        raise TypeError(f'Unsupported timestamp type: {type(value).__name__}')
ts = dt_to_unix_nanos(value)

from datetime import datetime

def is_datetime_like(value: object) -> bool:
    return isinstance(value, (int, str, datetime))

try:
    ts = dt_to_unix_nanos(value)
except TypeError as e:
    if 'datetime-like' in str(e):
        ts = int(pd_or_manual_conversion(value))  # coerce explicitly, then retry
    else:
        raise

When it happens

Trigger: dt_to_unix_nanos(1_700_000_000.5) or dt_to_unix_nanos(date(2024, 1, 1)) or a numpy scalar, in an environment where pandas import fails. The int branch returns early, the str branch parses ISO, the datetime branch converts — everything else falls through to the TypeError.

Common situations: Lightweight deployments without pandas receiving mixed-type timestamp columns (floats from CSV parses, numpy scalars, date objects); code that worked under pandas accepting datetime64/date then run pandas-free.

Related errors

• pandas is required for nanosecond-precision datetimes
• Chart renderer must be callable, was {type(renderer)}
• value must not be None
• No tearsheet chart registered under '{chart_name}'.{hint} Re
• Grid has {rows * cols} cells but {len(charts)} charts were c

AI-assisted analysis of nautechsystems/nautilus_trader@a4b06ed870 (2026-08-16). Data as JSON: /api/errors/98741bdf77040214. Report an issue: GitHub.