`.sel` with a bare number silently selects by a different scale on a coordinate in degrees or percent

Aperta
  • Qualcosa non va?
#400 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
56/100
Tipo di issue
Bug
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
python
Ambito
database

Direzione di ricerca

Start by reading PintIndex.sel and conversion.convert_indexer_units, then trace how array_convert_units handles bare numbers for index coordinates. Compare selection on scaled units such as degree and percent with radian and dimensionless, and review the existing indexer conversion tests. Done means the behavior is explicitly resolved and covered by tests for nearest selection and slices.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Hello. Thanks for all your work on pint-xarray, it's awesome.

To be up front, this issue was identified using AI (Claude Opus 5.5) - I was only really looking at quantified data, not coords, but this came up during testing. This doesn't affect me directly but seems like a case for a warning & a docs update (which I'm happy to do if you agree).

Everything below was written & tested using AI. Cheers!

What happened

I understand that a bare number on a quantified index counts as dimensionless, and that a Quantity is the intended label.

The problem is that on a coordinate whose unit is compatible with dimensionless but scaled, such as degree, that rule gives no error or warning, and code that worked before .pint.quantify() silently selects different labels.

import pint_xarray  # noqa: F401
import xarray as xr

da = xr.DataArray(
    [1.0, 2.0, 3.0],
    dims="lat",
    coords={"lat": ("lat", [10.0, 20.0, 30.0], {"units": "degree"})},
)
quantified = da.pint.quantify()

print(da.sel(lat=20.0, method="nearest").lat.item())  # 20.0
print(quantified.sel(lat=20.0, method="nearest").lat.item())  # 30.0

print(da.sel(lat=slice(10.0, 20.0)).sizes["lat"])  # 2
print(quantified.sel(lat=slice(10.0, 20.0)).sizes["lat"])  # 0

What I expected

An error, as on a coordinate in metres, where the same mistake is caught:

xr.DataArray(
    [1.0, 2.0],
    dims="z",
    coords={"z": ("z", [10.0, 20.0], {"units": "m"})},
).pint.quantify().sel(z=20.0)
# PintExceptionGroup: Cannot convert from 'dimensionless' (dimensionless) to 'meter' ([length])

Cause

PintIndex.sel passes its labels to conversion.convert_indexer_units, which ends in array_convert_units.
That treats a value that is not a quantity as dimensionless, as its docstring says and test_convert_indexer_units expects, and pint converts dimensionless to degrees by treating the number as radians.

So 20.0 becomes about 1146 degrees before the lookup:

  • method="nearest" then finds the largest label, 30.
  • The slice becomes roughly 573 to 1146 degrees, which is empty.
  • An exact .sel(lat=20.0) raises KeyError, so that case is at least loud.

.sel with a real quantity is correct, as in quantified.sel(lat=20.0 * ureg.degree, method="nearest").

Which units are affected

Run on the same three labels, 10, 20 and 30:

Coordinate unit .sel(c=20.0, method="nearest") .sel(c=slice(10.0, 20.0)) size
degree 30 0
arcminute 30 0
percent 30 0
radian 20 2
dimensionless 20 2
Unquantified, for comparison 20 2

The accessor gives a different answer from the index: quantified.pint.sel(lat=20.0, method="nearest") returns the 20 degree row, since it has no indexer units to convert to and strips the coordinate's own.

Latitude and longitude are the common case, since degrees_north is compatible with dimensionless under cf-xarray's registry, and .sel(lat=..., method="nearest") is routine.

Question

Is this the behaviour you intend?
If it is, could the docs say so, since docs/creation.rst still says xarray does not support units in indexes, and nothing I could find says that labels need to be quantities?

If not, one option is to raise for a bare number on an index with a scaled unit, matching what already happens with metres.
Selection on dimensionless and radian could keep working, since the number means the same either way.

Versions

Reproduced in a clean environment with only pint-xarray and xarray installed:

Package Version
Python 3.12
pint-xarray 0.6.1
pint 0.26.2
xarray 2026.9.0
numpy 2.5.4

convert_indexer_units and array_convert_units are unchanged on main.

The closest existing issues I found are #303, about method= being rejected, and #276, about indexing with quantities, which both concern a different failure.

Lingua principale
Python
Stelle
120
Fork
17
Nessuna PR unita negli ultimi 30g

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di xarray-contrib/pint-xarray

Tutte le issue di xarray-contrib/pint-xarray

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.