Count reads dropped by --dedup as duplicates instead of "passed filter" (#638, #528)

[cindykrafft]

Fixes #638. Also fixes #528.

Cause. With --dedup, the decision to drop a duplicate (dedupOut) only gates the output; the filter category is recorded before it is consulted (config->addFilterResult(result, 1) in seprocessor.cpp, addFilterResult(max(result1, result2), 2) in peprocessor.cpp) with result == PASS_FILTER. So every read removed by --dedup is still counted as passed: "reads passed filter" on stderr and filtering_result.passed_filter_reads in the JSON/HTML disagree with summary.after_filtering.total_reads and with the output file by exactly the number of removed duplicates, and there is no category that reports those reads. The merged branch in --merge mode never checks dedupOut at all, so --merge --dedup removes nothing.

Fix. A new filter category FAIL_DUPLICATE (following FAIL_ADAPTER_DIMER): a read or pair that would pass but is dropped by --dedup is recorded under it in all four places a result is recorded (SE; PE merged, --include_unmerged, plain). Reads failing another filter keep that reason. It is reported as reads failed due to duplication: on stderr, "duplicated_reads" in the JSON filtering_result and reads duplicated: in the HTML summary, only when --dedup is on, so reports of runs without --dedup are unchanged. One output change: because the merged branch now records FAIL_DUPLICATE and writes only on PASS_FILTER, --merge --dedup drops duplicate merged pairs as documented — happy to split that into a separate PR if you prefer.

Test. scripts/test_issue_638_dedup_filter_counts.sh, in the style of test_issue_697_stdout_merge.sh: an SE run with every read twice asserts passed_filter_reads == after_filtering.total_reads == reads written and duplicated_reads == input − written; an interleaved --merge --dedup run asserts that no more than the distinct pairs are written and that the two counters sum to the input. It fails on master (FAIL: SE --dedup wrote 500 reads but reported passed_filter_reads=1000 duplicated_reads=-1 after_filtering.total_reads=500) and passes with this change.

Run. make -j8 (no warnings), ./fastp test (ALL PASSED), scripts/test_issue_697_stdout_merge.sh (PASS) and the new script (PASS), on Ubuntu with libisal 2.31.0, libdeflate 1.19, libhwy 1.0.7. Synthetic check: 4,000 SE reads (2,000 distinct) with --dedup now report passed 2000 / duplicated 2000 / after_filtering 2000 / written 2000 instead of 4000 / – / 2000 / 2000.

Found while working through open issues in Mytochondria, a volunteer project that verifies fixes for the software behind published results (methods and harnesses: https://github.com/cindykrafft/mytochondria/tree/main/audits/fastp)

---

Generated by Claude Code