Qiskit / qiskit-fermions / 34236009275

98%

push

github

Test-suite sweep: duplication, runtime, gaps (#331)

* fix(operators)!: report FCIDump parse failures as catchable errors

`FCIDump::from_file` reported every parse failure by panicking. Crossing PyO3
that surfaced as `pyo3_runtime.PanicException`, which derives from
`BaseException` rather than `Exception`, so a caller guarding the parse with
`except Exception` never caught it and a Rust panic trace was printed to
stderr. A missing file is the single likeliest runtime error for this API, so
it was effectively unhandleable. On the C side the panic unwound across the FFI
boundary, which is undefined behaviour, and the function had no way to report
the failure at all because it returned the pointer directly.

The core now returns an `FCIDumpError` and each binding maps it onto its own
idiom: `OSError` for an unreadable path and `ValueError` for a format
violation in Python, a `QfExitCode` plus an out-parameter in C (matching the
convention the mapper functions already use).

The five Rust tests that pinned these paths asserted a panic message
substring; they now match on the error variant instead, which keeps them from
passing on an unrelated failure that happens to share wording. The gap that
hid this defect was on the Python side, which had no error-path test at all,
so the five cases are added there too.

Also resolves the two `TODO: actually assert all results!` markers in
`tests/c/test_fcidump.c`. Rather than transcribing the coefficients the Rust
tests already pin, the C tests assert the structural invariants the binding is
responsible for conveying: term and group counts, hermiticity, particle-number
conservation and max rank. `test_filtering.c` and `test_grouping.c` now check
the exit code too, so a bad fixture path fails cleanly instead of segfaulting.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* refactor(cext): route integral pointers through slice_from_ptr

The four triangular-integral constructors were the only place in the C API... (continued)

49 of 54 new or added lines in 5 files covered. (90.74%)

11454 of 11664 relevant lines covered (98.2%)

565116.11 hits per line