Summary
Several robustness gaps in src/easyreflectometry/data/measurement.py and orso_utils.py:
- 5+-column files crash.
_load_txt (measurement.py:96-100) unpacks exactly 4 names in the num_columns >= 4 branch; a 5-column file (q, R, sR, sQ, lambda — common) raises "too many values to unpack" instead of taking the first four.
sc.concatenate is the old scipp API. merge_datagroups calls sc.concatenate([a, b]) (measurement.py:131, 137); current scipp uses sc.concat(list, dim). No test covers the merge path, so this is likely broken at runtime.- Silent format fallback.
load (measurement.py:24-27) catches (IndexError, ValueError) from the ORSO loader and falls back to the bare-text loader — a malformed .ort loses its header (and resolution-column semantics, metadata) without any warning. Also load_orso_data (orso_utils.py:230) assumes 4 columns, so a legal 3-column .ort (sQz optional in the ORSO spec) takes this silent fallback path.
- Error-column semantics unvalidated.
_load_txt treats column 3 as sigma and squares it (measurement.py:108) — correct for the common convention but undocumented; a file carrying variance is silently mis-scaled by sigma^4.
- ORSO SLD units read heuristically.
_get_sld_values (orso_utils.py:166-204) assumes file SLDs are absolute A^-2 and multiplies by 1e6, ignoring the unit field orsopy provides; the >1e-2 magnitude warning is a guard, but reading the declared unit would be correct rather than heuristic.
Suggested fix
Take the first four columns explicitly (data[:, :4].T), move to sc.concat with a test, warn on ORSO→txt fallback, document/validate the error-column convention, and honor orsopy's unit metadata.
Found during deep code review (DEEP_ANALYSIS.md §4.9).
Summary
Several robustness gaps in
src/easyreflectometry/data/measurement.pyandorso_utils.py:_load_txt(measurement.py:96-100) unpacks exactly 4 names in thenum_columns >= 4branch; a 5-column file (q, R, sR, sQ, lambda — common) raises "too many values to unpack" instead of taking the first four.sc.concatenateis the old scipp API.merge_datagroupscallssc.concatenate([a, b])(measurement.py:131, 137); current scipp usessc.concat(list, dim). No test covers the merge path, so this is likely broken at runtime.load(measurement.py:24-27) catches(IndexError, ValueError)from the ORSO loader and falls back to the bare-text loader — a malformed.ortloses its header (and resolution-column semantics, metadata) without any warning. Alsoload_orso_data(orso_utils.py:230) assumes 4 columns, so a legal 3-column.ort(sQz optional in the ORSO spec) takes this silent fallback path._load_txttreats column 3 as sigma and squares it (measurement.py:108) — correct for the common convention but undocumented; a file carrying variance is silently mis-scaled by sigma^4._get_sld_values(orso_utils.py:166-204) assumes file SLDs are absolute A^-2 and multiplies by 1e6, ignoring the unit field orsopy provides; the >1e-2 magnitude warning is a guard, but reading the declared unit would be correct rather than heuristic.Suggested fix
Take the first four columns explicitly (
data[:, :4].T), move tosc.concatwith a test, warn on ORSO→txt fallback, document/validate the error-column convention, and honor orsopy's unit metadata.Found during deep code review (DEEP_ANALYSIS.md §4.9).