Qiskit / qiskit-fermions / 29992301003

98%

push

github

fix(transpiler): build fresh output DAG in F2QSynthesis (#201)

F2QSynthesis produced its qubit output DAG by copying the fermionic
input DAG (`copy_empty_like`) and then deleting the mode wires via
`remove_qregs`/`remove_qubits`. That in-place surgery leaves the DAG's
internal wire graph structurally inconsistent, which trips a spurious
`not a DAG` panic in downstream analysis — concretely Qiskit's `Depth`
pass in the qubit stage of the JW preset.

The panic originates in Qiskit's rustworkx-backed DAG and is a
heisenbug: whether a given process enters the failing state depends on
memory layout (~70% of fresh processes here), but once it does, every
`depth()` call on the corrupted DAG panics. This is why it showed up as
a flaky CI failure. A standalone reducer (see the PR description)
demonstrates it with Qiskit alone.

Only circuits that synthesize to single-qubit gates on a subset of the
wires expose it: an identity `OrbitalRotation` lowers to just `X` gates
with idle qubits. Circuits whose 2-qubit gates touch every wire connect
all the wires and mask the inconsistency.

Build the output DAG from scratch instead — carry over name, global
phase and metadata, add the mapped qubit registers, and re-add any
input classical registers — so the wire graph stays consistent. The
synthesis plugins are unchanged: they still index the same layout
registers, whose bits are the canonical ones `add_qreg` holds.

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

6 of 7 new or added lines in 1 file covered. (85.71%)

9764 of 9996 relevant lines covered (97.68%)

329.87 hits per line