SubgraphFusion crashes with IndexError: list index out of range error on certain graphs.

[pratyai]

Describe the bug

This came up as I was working with #1673 (i.e. fixing some sub-optimal behaviour of MapFusion causes an unrelated test to break through SubgraphFusion.

While I'd like to fix it myself already, I cannot tell from the subgraph fusion code what it was trying to do here. So, here is an issue for longer-term tracking. In general, the subgraph fusion code looks quite untested and unmaintained. If that's really the case, is it worth to keep SubgraphFusion in the default optimization routine?

To Reproduce

Use this SDFG: https://polybox.ethz.ch/index.php/s/pMlKv8jkJ3B1SdR/download?path=%2F&files=nbody-subgraphfusion-bug.sdfg

With the this code:

import dace
from dace import SDFG
from dace.transformation.auto.auto_optimize import auto_optimize

g = SDFG.from_file('nbody-subgraphfusion-bug.sdfg')
g.validate()
auto_optimize(g, device=dace.dtypes.DeviceType.CPU, validate_all=True)

Expected behavior

It wouldn't crash, and be successful with the fusion.