Background

The dist/s_stat validation script complains about unused statistics fields. It extracts the field list from the generated struct in src/include/stat.h and searches for literal references to each field in the sources, skipping stat.c (which lists all fields by definition).

Problem

Two gaps let unused statistics fields through unnoticed:

• The extraction pattern ^ int64_t ([a-z_]*);$ does not accept digits, so every histogram bucket field (lt2, lt5, lt10, gt1000, …) is never extracted into the check list at all.
• The histogram increment macros reference fields by token pasting (stat##_lt10), which literal matching cannot see. Fields populated only through those macros have to be excused manually in the UNUSED_STAT_FIELDS exclusion list, and a bucket declared in stat_data.py but never written by a macro (the subject of WT-17991) goes undetected.

Proposed change

• Accept digits in the field-name extraction ([a-z0-9_]).
• Treat a field as used if it ends with one of the ##-pasted suffixes found in stat.h and its base name appears literally in the sources.
• Drop the perf_hist_total entries from the exclusion list — the suffix logic now detects their use on its own merits.
• Add the numeric compress_read_ratio_hist_ / compress_write_ratio_hist_ fields to the exclusion list — they are referenced literally, but only inside stat.h itself, which cannot join the search corpus without matching every struct declaration.

Verification

• Against an unfixed develop tree, the script exits 1 and flags exactly the ten never-populated bucket fields of WT-17991: {{perf_hist_ {bmread,bmwrite,fsread,fswrite}

  latency

  {lt2,lt5}}} and {{live_restore_hist_source_read_latency_{lt2,lt5}

  }}.

• With the WT-17991 macro fix applied, the script exits 0.

Definition of done

• dist/s_stat change as described; cd dist && ./s_stat passes on the fixed tree and fails with a useful message on unfixed develop.