discussion: remove id exposure from interface

[KaiSzuttor]

No description provided.

[KaiSzuttor]

from the user perspective the system state should not change if two particles are exchanged. the particle id is an implementation detail. this would also remove the odd slice index to id coupling which does not make sense to me

[jngrad]

import numpy as np
import espressomd
system = espressomd.System(box_l=3*[10])
for i in range(3):
    system.part.add(pos=np.random.random(3) * system.box_l)

system.part[1].remove()
system.part[:].pos  # returns 2 elements
system.part[1].pos  # RuntimeError: Particle node for id 1 not found

[KaiSzuttor]

import numpy as np
import espressomd
system = espressomd.System(box_l=3*[10])
for i in range(3):
    system.part.add(pos=np.random.random(3) * system.box_l)

system.part[1].remove()
system.part[:].pos  # returns 2 elements
system.part[1].pos  # RuntimeError: Particle node for id 1 not found

if somebody knows python, this is a very surprising behavior since there is no slice with gaps in the python world. i'd really prefer not to hold onto this slicing syntax.

[fweik]

While I agree with you in principal, I think from a practical point of view I think this is unwise: It will break most user scripts, without there being an obvious reason for the break. What is your cost/benefit analysis for this? I also think that there are more
serious problems in the particle interface that should be addressed first, e.g. that the code is not reusable for particles that are not in the cell system, and the implementation of the slices seems rather incidental (pythonic?) to me.

(This think this also smells of the Potemkin Pattern)

[KaiSzuttor]

I obviouly did not do a complete analysis, that's why I wanted to have a lively discussion in this issue.

Generally I think that chamging this slicing interface could lead to more readable simulation scripts. I'm thinking of giving groups of filtered particles a name for further reuse, e.g..
But yes, breaking interface changes should not be done, if we do bot have a major benefit overall.

[RudolfWeeber]

First step is to change the text and examples in `doc/sphinx/part.rst` to make it clear that ids don’t hve to be specified and that the references returned by `system.part.add()` should be used

[fweik]

On a general note, I have come to think that the Potemkin Pattern is a much bigger design issue than we used to think. The problem with emulating a different API in the interface than in the core is that this always encodes semantic information into the python interface (as opposed to merely syntactic information), and as such is always problematic. If you think about this, the purpose of such a construction is actually coupling...

[pkreissl]

As already mentioned in PR #4058, there are for now still some places, especially in the espressomd.observables, where cannot be omitted, such as ParticlePositions.

[KaiSzuttor]

of course we would need to have typebased observables instead of id based

[jngrad]

After #4402, bond lists and particle-based observables are the last consumers of the particle ids in the python interface.

[RudolfWeeber]

By now, to my understanding, only the particle observables are left. This can be addressed on the Python level.

#4954