mpsutil.common.graph topological is stable

[abstraktor]

sorting is maintaining order of equal items (a.k.a. "a stable sorting algorithm")

this would solve Issue 1 of #2139

[coolya]

Before we merge this, apart from fixing the obvious build problem, I will need collect some real world performance numbers on how this impacts performance, especially for interpreters.

[abstraktor]

sorting the interpreters seems to be the only usage of common.graph from mpsutils. I asked around and didn't hear about any other project using it.

Regarding the perormance of sorting the interpreters: Our largest interpreter is the javainterpreter with around 10 interpreter nodes. So we're talking about performance of sorting 10 items, which shouldn't be of too high impact, I guess. Or are there places where the interpreter is called in such a frequent fashion that this turns relevant?

[coolya]

The java interpreter is a toy example. That's why we need metrics from real world and yes there are places especially when collecting coverage information where hundreds of interpreters are called rapidly.

[abstraktor]

Oh I see. I didn't know that.

[coolya]

While the sorted version seems slower, most interpreter usecases will anyway use InterpreterEvaluationHelper which caches the interpreter for a given category hence the sorted evaluators are also cached. From an interpreter point of view I see now reason not to merge this change since the performance impact is not really relevant.

[alexanderpann]

I am closing this PR because the changes were merged but not through this PR.