CBIIT / ccdi-dcc-federation-service / 29341458116

88%

main: 87%

push

github

Replace count reconciliation check with a cardinality-free invariant

The check asserted total == values_sum + missing. That identity only holds for
single-value fields: it is false by design whenever one (sample_id, study_id)
pair can carry several values -- a list property such as anatomical_sites, or a
1:N related node, since a sample has many sequencing_files, pathology_files and
diagnoses. It held for only 9 of the 15 countable fields.

The consequence was a diagnostic that cried wolf: library_selection_method,
library_strategy, library_source_material and specimen_molecular_analyte_type
each logged "Total count mismatch for combined query (should not happen)" at
WARNING on every request, so a genuine mismatch would have looked exactly like
steady state. multi_value_fields also listed two fields that took the combined
branch and could never reach it.

Assert instead the only relationship that is true by construction: every pair is
either missing or carries at least one value. This needs no per-field
cardinality list to keep in sync with the data model.

    0 <= missing <= total
    values_sum >= total - missing
    values_sum == 0  iff  missing == total

Holds for all 15 fields on live data and silences the 4 spurious warnings, while
still catching a missing-count that undercounts pairs. Also wired into the two
diagnosis count methods, which previously had no check at all.

Diagnostic only: no count is adjusted and no response changes.

21 of 22 new or added lines in 1 file covered. (95.45%)

17014 of 19392 relevant lines covered (87.74%)

1.31 hits per line