tals — Time-Aware List Slicer

tals makes time-based slicing as natural as ordinary indexing. It extends Python's [start:end] syntax with datetime and calendar-period bounds, and works on any list of objects or dicts — it only needs to k now which attribute or key holds the timestamp.

---

Filtering a list of objects by time period is something every Python project does — and it never gets less tedious:

# Without tals
now = datetime(2026, 3, 10)
start = now.replace(day=1, hour=0, minute=0, second=0, microsecond=0)
end = (start + relativedelta(months=1))
february = [e for e in entries if start <= e.created_at < end]

tals collapses this to a single expressive call:

# With tals
slice_objects(entries, "-1M:0M", "created_at", now)

It extends Python's familiar [start:end] syntax with datetime and calendar-period bounds. Any list, any object — it only needs to know which attribute or key holds the timestamp.

from tals import slice_objects

slice_objects(entries, "-7d:",    "created_at", now)  # last 7 days
slice_objects(entries, "0M:+1M", "created_at", now)  # this calendar month
slice_objects(entries, "-1W:0W", "created_at", now)  # last full week
slice_objects(entries, "0Y:",    "created_at", now)  # since Jan 1

Installation

pip install tals

Overview

Three new bound types extend standard integer indexing:

• Time-delta bounds — relative to a reference datetime: -7d, +2h, -30m
• Calendar-period bounds — snapped to period boundaries: 0M, -1W, 0Y
• Literal timestamp bounds — an absolute ISO-8601 point in time: 2026-01-01, 2026-01-01T14:30:00+02:00

Bounds can be freely mixed: a temporal start can pair with an integer end, or vice versa. The library never calls datetime.now() — you supply the reference, so behaviour is always deterministic and testable.

API

def slice_objects(
    objects: list[Any],
    position: str,
    timestamp_key: str,
    reference_dt: datetime,
    week_start: int = 0,
    presorted: bool = False,
    inclusive_end: bool = False,
) -> Any | list[Any]:

Parameter	Description
objects	List of objects or dicts in any order
position	Slice expression string (see syntax below)
timestamp_key	Attribute name (objects) or key (dicts) holding the datetime value
reference_dt	Anchor for all relative expressions — never inferred from the system clock
week_start	First day of the week: 0 = Monday (default), 6 = Sunday
presorted	Skip sorting if the list is already in ascending timestamp order
inclusive_end	Make the end bound inclusive (default False)

Return value: a single-index expression returns one object (or None if out of bounds); a slice expression returns a list (possibly empty).

The list is stable-sorted ascending by timestamp_key before slicing. All indices operate on this sorted list.

Syntax

[index]
[start:end]
[start:]
[:end]
[:]

Bound types

Form	Description	Example
0, 1, -1	Integer index — same semantics as Python	[-1] → last object
-7d, +2h, -30m	Time-delta — relative to reference_dt	[-7d:] → last 7 days
0M, -1M, +1M	Calendar month — start of the Nth month	[0M:+1M] → this month exactly
0W, -1W, +1W	Calendar week — start of the Nth week	[-1W:0W] → last week exactly
0Y, -1Y, +1Y	Calendar year — Jan 1 of the Nth year	[0Y:+1Y] → this year exactly
2026-01-01, 2026-01-01T14:30, 2026-01-01T14:30:00+02:00	Literal timestamp — any ISO-8601 date or datetime, accepted by datetime.fromisoformat	[2026-01-01:2026-02-01] → January exactly

Period 0 is the period containing reference_dt; -1 is the previous period; +1 is the next.

Calendar and literal-timestamp bounds are not valid as a single index — they only make sense in a slice.

A literal timestamp with no offset (2026-01-01T14:30) is naive; if reference_dt is timezone-aware, it inherits reference_dt's timezone, the same way calendar boundaries are computed in reference_dt's timezone. A literal timestamp with an explicit offset or Z is aware and used as given — it does not need to match reference_dt's offset, since Python compares differing aware timezones correctly.

Inclusivity

By default, bounds follow Python's convention: start is inclusive, end is exclusive.

Expression	Meaning
[-7d:]	timestamp >= reference_dt − 7 days
[:-7d]	timestamp < reference_dt − 7 days
[-1M:0M]	timestamp >= start of last month and < start of this month

Pass inclusive_end=True to include the end boundary:

Expression	Default (inclusive_end=False)	With inclusive_end=True
[-1M:0M]	up to but not including Mar 1	up to and including Mar 1
[-7d:-1d]	up to but not including the -1d mark	up to and including the -1d mark
[0:2]	indices 0 and 1	indices 0, 1, and 2
[:-1]	all but the last	all items

Mixed bounds

A slice can mix bound types. Temporal bounds are resolved to datetime thresholds first, the list is filtered, then integer bounds are applied to the filtered result.

# March entries, excluding the last one
slice_objects(entries, "0M:-1", "created_at", reference_dt)
# → filter to [Mar 1, Apr 1), then apply [:-1]

# From 7 days ago, keep only the first three
slice_objects(entries, "-7d:3", "created_at", reference_dt)
# → filter to [now - 7d, …), then apply [:3]

# Between two known absolute timestamps
slice_objects(entries, "2026-01-01:2026-02-01", "created_at", reference_dt)
# → filter to [Jan 1, Feb 1)

Examples

Given four objects with a start attribute and reference_dt = 2026-03-10:

Object	start
A	2026-01-10
B	2026-02-05
C	2026-03-01
D	2026-03-10

slice_objects(objs, "[-1]",      "start", ref)  # → D
slice_objects(objs, "[0]",       "start", ref)  # → A
slice_objects(objs, "[:]",       "start", ref)  # → [A, B, C, D]
slice_objects(objs, "[:-1]",     "start", ref)  # → [A, B, C]
slice_objects(objs, "[0M:]",     "start", ref)  # → [C, D]          March onward
slice_objects(objs, "[0M:+1M]",  "start", ref)  # → [C, D]          March exactly
slice_objects(objs, "[-1M:0M]",  "start", ref)  # → [B]             February exactly
slice_objects(objs, "[-1M:]",    "start", ref)  # → [B, C, D]       Feb 1 onward
slice_objects(objs, "[0Y:+1Y]",  "start", ref)  # → [A, B, C, D]    2026 exactly
slice_objects(objs, "[0M:-1]",   "start", ref)  # → [C]             March, drop last
slice_objects(objs, "[2026-02-01:2026-03-05]", "start", ref)  # → [B, C]  Feb 1 – Mar 5

Inclusive end

Add inclusive_end=True when items exactly on the boundary should be included. A common case is querying a closed interval between two known timestamps:

# Events from Feb 5 through Mar 1 inclusive — useful when Mar 1 is a known event
slice_objects(objs, "-1M:0M", "start", ref)                       # → [B]       Mar 1 excluded
slice_objects(objs, "-1M:0M", "start", ref, inclusive_end=True)   # → [B, C]    Mar 1 included

It applies equally to integer ends, where it behaves like Ruby-style range slicing:

slice_objects(objs, "0:2",  "start", ref)                        # → [A, B]     index 2 excluded
slice_objects(objs, "0:2",  "start", ref, inclusive_end=True)    # → [A, B, C]  index 2 included

slice_objects(objs, ":-1",  "start", ref)                        # → [A, B, C]  last excluded
slice_objects(objs, ":-1",  "start", ref, inclusive_end=True)    # → [A, B, C, D]  last included

Timezone handling

reference_dt and all object timestamps must be either all timezone-aware or all timezone-naive — mixing the two raises TypeError. Objects with different (but both aware) timezones are compared correctly by Python and are fully supported.

Calendar boundaries are computed in the timezone of reference_dt, so 0M on a UTC+02:00 reference resolves to midnight of the 1st in that timezone.

Out of scope

tals is a pure slicing primitive. The following are the caller's responsibility:

• Pre-filtering — pass only the subset of objects that should be considered
• Field extraction — read attributes from the returned objects
• Fallback values — convert None or [] to domain-specific defaults

License

MIT