fix(sinks): slice list and tuple custom_data values per row

[shaun0927]

Before submitting

• Self-reviewed the code
• Updated documentation, follow Google-style
• Added docs entry for autogeneration (if new functions/classes)
• Added/updated tests
• All tests pass locally

Description

Extend the per-row slicing already done for numpy.ndarray values in CSVSink and JSONSink so it also covers plain Python list / tuple values whose length matches the number of detections. Shorter or differently-sized sequences keep the current broadcast behavior, so existing complex-data tests continue to pass.

Type of Change

• 🐛 Bug fix (non-breaking change which fixes an issue)

Motivation and Context

#2199 fixed the "entire array on every row" bug for numpy.ndarray values in custom_data, but the patch only added a numpy.ndarray branch to _slice_value; any non-ndarray value still returned unchanged. When a user passes a plain Python list — e.g., per-detection string ids — every row receives the full list instead of its per-row element:

import numpy as np
import supervision as sv

det = sv.Detections(
    xyxy=np.array([[0, 0, 10, 10], [20, 20, 40, 40], [50, 50, 70, 70]]),
    class_id=np.array([1, 2, 3]),
    confidence=np.array([0.9, 0.8, 0.7]),
)
with sv.CSVSink("out.csv") as sink:
    sink.append(det, custom_data={"ids": ["a", "b", "c"]})

Before this PR the ids column contains "['a', 'b', 'c']" on every row; with the fix each row gets "a", "b", "c" respectively. The same regression applies to JSONSink.

The complex_data test cases in tests/detection/test_csv.py ({"tags": ["urgent", "review"]} broadcast across a single detection) still pass because the length check guarantees the existing broadcast behavior whenever the sequence length does not match the detection count.

Changes Made

• src/supervision/detection/tools/csv_sink.py — _slice_value now also indexes list / tuple values whose length equals the number of detections; other non-array values are still broadcast.
• src/supervision/detection/tools/json_sink.py — same change, keeping parity between the two sinks.
• tests/detection/test_csv.py — new test_csv_sink parameter covering a 3-row, 2-row, and 1-row sink over list and tuple custom data.
• tests/detection/test_json.py — equivalent new parameter for test_json_sink.

Testing

• I have tested this code locally
• I have added unit tests that prove my fix is effective or that my feature works
• All new and existing tests pass

Google Colab (optional)

Screenshots/Videos (optional)

Additional Notes

Backward compatibility: scalar values ({"frame_number": 42}) and sequences whose length does not match the number of detections ({"tags": ["urgent", "review"]} against a single detection) still write unchanged to every row, matching the behavior exercised by the existing complex_data tests in both sinks.

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 77%. Comparing base (5d165e0) to head (a3564a1).
⚠️ Report is 1 commits behind head on develop.

Additional details and impacted files

@@           Coverage Diff           @@
##           develop   #2216   +/-   ##
=======================================
  Coverage       77%     77%           
=======================================
  Files           66      66           
  Lines         8189    8195    +6     
=======================================
+ Hits          6303    6309    +6     
  Misses        1886    1886           

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

Fixes per-detection row serialization in CSVSink and JSONSink so that custom_data values provided as plain Python list/tuple are sliced per row (when their length matches the number of detections), matching the existing behavior for numpy.ndarray.

Changes:

• Extend _slice_value in both sinks to slice list/tuple values when len(value) == number_of_detections.
• Thread n = len(detections.xyxy) through parsing loops to support length-based slicing.
• Add/extend unit tests to cover list/tuple per-row slicing across multi-append scenarios.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
src/supervision/detection/tools/csv_sink.py	Slice list/tuple custom_data per detection row when length matches detection count.
src/supervision/detection/tools/json_sink.py	Mirror the same list/tuple slicing behavior in JSON serialization.
tests/detection/test_csv.py	Add coverage validating list/tuple custom_data is sliced per row (and scalar still broadcasts).
tests/detection/test_json.py	Add coverage validating list/tuple custom_data is sliced per row in JSON output.