Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

Model result and observation constructors stamp `kind` into the Dataset they are handed

Aperta
  • Qualcosa non va?
#712 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
72/100
Tipo di issue
Bug
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
python
Ambito
data

Direzione di ricerca

Start with TimeSeries.init and _validate_dataset in src/modelskill/timeseries/_timeseries.py, then inspect the five constructor sites listed in the issue and the existing vertical re-wrapping tests. Run tests/observation/test_vertical_obs.py and tests/model/test_vertical.py before adding coverage for a source observation surviving a model-result read. Done means constructors no longer mutate handed-in datasets while existing behavior remains intact.

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

Descrizione

bug

TimeSeries.__init__ (src/modelskill/timeseries/_timeseries.py:171) keeps the dataset it is given, by reference:

self.data = data if self._is_input_validated(data) else _validate_dataset(data)

Each constructor then marks the primary variable's kind in place. On the validated path that variable belongs to the caller, so reading an observation's data turns the observation into a model result:

o = ms.PointObservation(df, x=1.0, y=2.0, item="WL")
o.data["WL"].attrs["kind"]     # 'observation'

mr = ms.PointModelResult(o.data)

o.data["WL"].attrs["kind"]     # 'model'   <- the observation changed
mr.data is o.data              # True

Sites

The same three lines are copy-pasted at five constructors:

  • src/modelskill/model/point.py:69
  • src/modelskill/model/track.py:71
  • src/modelskill/model/vertical.py:113
  • src/modelskill/obs.py:155
  • src/modelskill/model/network.py — fixed on network-phase-2 by copying first

Observation.__init__ (src/modelskill/obs.py:152-165) goes furthest. Beyond kind it rewrites the source's dataset-level attrs, its time variable, the plot color and weight:

data.attrs = {**data.attrs, **(attrs or {})}
data["time"] = self._parse_time(data.time)
data[data_var].attrs["color"] = color
super().__init__(data=data)
self.data.attrs["weight"] = weight

_validate_dataset (_timeseries.py:117-160) also mutates its argument and returns it, so the unvalidated path is only incidentally safe: it is safe because the parse functions built a fresh object, not because anything copied.

Why it has gone unnoticed

Re-wrapping a validated dataset is rare, and the two tests that do it re-wrap same-to-same, where the kind written matches the kind already there: tests/observation/test_vertical_obs.py:184 and tests/model/test_vertical.py:170. Nothing asserts that a source observation survives being read.

DummyModelResult is the one class already safe, and it is safe because it copies — da = observation.data[observation.name].copy() (src/modelskill/model/dummy.py:65).

comparison/_comparison.py:509 is a latent instance: PointModelResult(self.data[[str(key)]], name=str(key)) stamps a variable still shared with Comparer.data, because ds[[var]] shares Variable objects. It is benign only because the kind already reads model.

Suggested fix

A shallow ds.copy() before stamping is enough — xarray rebuilds each variable's attrs dict, so no array data is copied. deep=True is not needed. Note that ds[[var]], ds.sel(...) and ds.isel(...) do not work for this: they share the Variable objects and the write goes through to the parent.

Since the stamp is duplicated five times, one helper that copies and then marks the primary variable would remove the duplication and the aliasing together. _include_attributes and _include_coords (timeseries/_point.py:150, :168) already open with ds = ds.copy(), so the discipline exists in the codebase.

Lingua principale
Python
Stelle
56
Fork
9
Merge medio
1h 24m
PR unite (30g)
2

Preparare l'ambiente

Apri in Codespaces

Avvia il container di sviluppo del progetto nel browser, con il tuo account GitHub.

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 DHI/modelskill

Tutte le issue di DHI/modelskill

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.