#!/usr/bin/env bash
# Usage: TIER=Q|F bench/steps/oracle.sh ENGINE [NSUB]
# Checks the first NSUB (default 200) trace substrings as and/or/phrase/not queries: the engine's
# match set and count(*) must equal the oracle's (tsvector, case folding, stemming off).
# An engine outside Postgres (Elasticsearch, steps/xlib.sh) is built in its own server, asked the match requests and the count requests
# over HTTP, and its answers are compared in the engine's database (oracle_diff).
# Env: ORACLE_ACCENTS=fold|keep (default fold, keep for the engines indexing raw 'simple': gin, rum, pg_textsearch)
#      ORACLE_STEM=off|<snowball language> (default off)
#      ORACLE_TOKENS=split|pg (default split: words are alphanumeric runs like sift's, pg for the engines using the PG parser:
#      gin, rum, pg_textsearch)
#      MODES="and or phrase not"
#      REBUILD=1 rebuilds the o_ENGINE database. Exit status 1 on any mismatch.
. "$(dirname "$0")/xlib.sh"

load_engine "${1:?engine}"
NSUB=${2:-200}
engine_available || { log "$ENGINE: not installed, skipping"; exit 0; }
mkdir -p "$OUT"
MODES=${MODES:-and or phrase not}
if [[ -n $EXTERNAL ]]; then
  RUN=$OUT/$ENGINE; mkdir -p "$RUN"
  clone_db "$(engine_db "$ENGINE")" >/dev/null
  trap 'xe "$ENGINE" x_stop 2>/dev/null || true' EXIT
  f=$(x_export_docs)
  x_start
  locked x_build "$f"
else
  check_db
fi
q -v accents="${ORACLE_ACCENTS:-fold}" -v stem="${ORACLE_STEM:-off}" -v tokens="${ORACLE_TOKENS:-split}" -f "$BENCH/sql/oracle.sql"
REPORT=$OUT/oracle-$ENGINE.txt

if [[ -n $EXTERNAL ]]; then
  x_oracle_answers "$NSUB" "$MODES"
  RESULTS_SQL=$(cat <<SQL
CREATE TEMP TABLE got_ids (id int, ids bigint[]);
CREATE TEMP TABLE got_counts (id int, n bigint);
\\copy got_ids FROM '$RUN/oracle.ids'
\\copy got_counts FROM '$RUN/oracle.counts'
CREATE TEMP TABLE res AS
SELECT q.id, q.mode, q.terms, r.*
FROM oracle_queries q JOIN got_ids g USING (id) JOIN got_counts c USING (id) CROSS JOIN LATERAL oracle_diff(q.mode, q.terms, g.ids, c.n) r;
SQL
)
else
  RESULTS_SQL=$(cat <<'SQL'
CREATE TEMP TABLE res AS
SELECT q.id, q.mode, q.terms, r.*
FROM oracle_queries q CROSS JOIN LATERAL oracle_check(q.mode, q.terms) r
WHERE q.sub <= :nsub AND q.mode = ANY (string_to_array(:'modes', ' '));
SQL
)
fi

{
  echo "$RESULTS_SQL"
  cat <<'SQL'

\echo == summary (accents/stem: see oracle_tsv comment) ==
SELECT obj_description('oracle_tsv'::regclass, 'pg_class') AS oracle_cfg, mode,
       count(n_engine) AS checked, count(*) - count(n_engine) AS unsupported,
       count(*) FILTER (WHERE missing + extra > 0 OR n_count <> n_oracle) AS mismatched,
       sum(n_oracle) AS oracle_rows
FROM res GROUP BY mode ORDER BY mode;

\echo == mismatches ==
SELECT r.id, r.mode, array_to_string(r.terms, ' ') AS query, r.n_oracle, r.n_engine, r.n_count,
       r.missing, r.missing_ids, r.extra, r.extra_ids,
       left(regexp_replace(d.body, '\s+', ' ', 'g'), 160) AS first_diff_doc
FROM res r
LEFT JOIN docs d ON d.id = coalesce(r.missing_ids[1], r.extra_ids[1])
WHERE r.missing + r.extra > 0 OR r.n_count <> r.n_oracle
ORDER BY r.id LIMIT 50;

SELECT count(*) = 0 AS ok FROM res WHERE missing + extra > 0 OR n_count <> n_oracle \gset
\if :ok
\else
\echo FAIL
\endif
SQL
} | psql -X -v ON_ERROR_STOP=1 -q -v nsub="$NSUB" -v modes="$MODES" | tee "$REPORT"
! grep -q '^FAIL$' "$REPORT"