An absent variable is indistinguishable from a zero variable

[FBumann]

Issue Description

Structural operations — .shift(), .where(), mask= — leave absent slots: coordinates where the variable does not exist. linopy does not mark these as absent once the variable enters an expression, so after multiplication an absent slot is indistinguishable from a genuine zero-valued term, and fillna() cannot reach it.

Reproducible Example

import pandas as pd
import linopy

m = linopy.Model()
x = m.add_variables(coords=[pd.RangeIndex(5, name="time")])

xs = x.shift(time=1) # time=0 is now absent
print((xs * 3).isnull().values[0]) # False — the absent slot looks present

After xs * 3 the slot at time=0 reports isnull() == False — a present, 0-valued term — so it cannot be told apart from a real zero term.

Expected Behavior

An absent slot should stay absent through multiplication (isnull() == True), so absent and zero remain distinguishable.

Confirmed on linopy 0.7.0. Part of the arithmetic-convention work (#591).

[FabianHofmann]

I am wondering how we want to deal with it (you probably thought about this). given that #713 will make nan's raise, in v1 the operation above will too. just to confirm: this is what we want, right? ie. no nan-propagation whatsoever

[FBumann]

@FabianHofmann nan in user data is different than an absent slot in a variabe or expression.
nan means "absent". But providing user data (dataframes, dataarrays,...) with nan raises. So "absence" can only be introduced through operations on linopy objects like shift, reindex, ..., but not on parameters (data).

[FabianHofmann]

@FabianHofmann nan in user data is different than an absent slot in a variabe or expression. nan means "absent". But providing user data (dataframes, dataarrays,...) with nan raises. So "absence" can only be introduced through operations on linopy objects like shift, reindex, ..., but not on parameters (data).

fully agree, missed the distinction, that means the above should not raise but confirm the "null" data. So in the end there will be

1. implicit absent data from operations above stored as nan's -> should return true for isnull
2. explicit masked data from defining and passing masks -> should not return true for isnull (?)

[FBumann]

1. yes
2. Also return null

Both mean absence of a slot. isnull tests the absene of a slot

[FabianHofmann]

1. yes
2. Also return null

Both mean absence of a slot. isnull tests the absene of a slot

I like, important that we have it in the convention

[FBumann]

1. yes
2. Also return null

Both mean absence of a slot. isnull tests the absene of a slot

I like, important that we have it in the convention

Its in there