EMT::Ph3::AvVoltageSourceInverterDQ segfaults with withTrafo=false: mVirtualNodes[3] dereferenced when only 3 are allocated

[ysrji]

Summary

EMT::Ph3::AvVoltageSourceInverterDQ constructed with withTrafo = false segfaults during
initialisation. On that branch the constructor allocates three virtual nodes, but four sites
dereference mVirtualNodes[3] unconditionally.

Tested at 1b063c54298a46645c04c5ca156ad8178e37f486 (main, 2026-07-21), built from source with
Python bindings, EMT Ph3 / MNA, dt = 100 us.

Reproducer

examples/Notebooks/Circuits/EMT_Slack_PiLine_VSI_with_PF_Init.ipynb runs clean as committed. Change
only the with_trafo argument:

pv = dpsimpy.emt.ph3.AvVoltageSourceInverterDQ(
    "pv", "pv", dpsimpy.LogLevel.debug, with_trafo=False   # was True
)

and remove the now-inapplicable set_transformer_parameters(...) call. The process dies with SIGSEGV
after MnaSolverFactory ... creating KLUAdapter solver implementation, i.e. inside
initializeFromNodesAndTerminals, before the first step. Nothing else in the notebook is touched.

Cause

dpsim-models/src/EMT/EMT_Ph3_AvVoltageSourceInverterDQ.cpp, constructor:

if (withTrafo) {
  setVirtualNodeNumber(4);
  ...
} else {
  setVirtualNodeNumber(3);
}

With the transformer, index 3 is the filter-interface node. Without it, index 3 is one past the end.
These four dereferences are not guarded by mWithConnectionTransformer:

• L277 Real theta = std::arg(mVirtualNodes[3]->initialSingleVoltage());
• L291 matrixStateInit(0, 0) = std::arg(mVirtualNodes[3]->initialSingleVoltage());
• L292 matrixOutputInit(0, 0) = std::arg(mVirtualNodes[3]->initialSingleVoltage());
• L413 vcdq = parkTransformPowerInvariant(theta, **mVirtualNodes[3]->mVoltage);

The first three are in initializeFromNodesAndTerminals (where the crash lands); L413 is in the
per-step control update, so it would fire every step if initialisation were survived.

Note the surrounding code already handles the no-transformer case correctly — e.g. L263-266 picks
the terminal node instead of mVirtualNodes[3] for mSubResistorC, and L217-236 branches properly
on the initial-voltage calculation. It appears only these four references were missed.

Suggested fix

In each of the four places, use the node that is electrically equivalent when there is no transformer,
which is the one mSubResistorC is already connected to on that branch — mTerminals[0]->node().
Something of the shape:

auto filterIfNode = mWithConnectionTransformer
                        ? mVirtualNodes[3]
                        : mTerminals[0]->node();

then use filterIfNode at all four sites. That keeps the with-transformer path bit-identical.

Why it matters

For a low-voltage island there is no step-up transformer to model, so withTrafo = false is the
physically correct construction, and it is reachable from the Python bindings with no warning. A
guard, or an explicit "not supported" throw, would at least turn a segfault into something
diagnosable.

Happy to open a PR if the suggested shape looks right.

[leonardocarreras]

Hi @ysrji ! Thanks for looking into this. I can reproduce it on current master. Interesting one: the Python bindings default with_trafo to false, so it's the path you get without passing the flag. Our notebooks all set it to True, so that is a good guess on why we never run into it.

A PR would be great if you're up for it. EMT_Ph3_VSIVoltageControlVCO might be a useful reference, since it has the same node-count split and wraps the whole control-init block in an if/else.

EMT::Ph3 on its own is already nice if you want to do it. If you like, SP and DP look similar, and we can take those separately on other PRs once this one works.