#!/bin/bash
# Correctness gate: no perf win is allowed to cost functionality.
#
#   1. .auto/check_search.py -- the search fast paths against the plain loop.
#      Caches that go stale still return AN answer, so no scenario test can see
#      them; this compares fast and slow directly, for every typed prefix.
#   2. .auto/check_rename.py -- a rename keeps the listing's walk and only
#      re-renders its text. When that goes wrong the pane simply keeps showing
#      the old name, which nothing else notices. Checks a NARROW read (what
#      painting does) and a WIDE one (what building the search body does), and
#      the whole model against a rebuild.
#   3. .auto/check_edit.py -- the listing after an ITEM edit. bump_items(ea)
#      keeps the walk in front of the edit rather than discarding it; keeping
#      anything across a structural change is the risky half and it fails
#      silently, so the kept model is compared against a rebuild row for row.
#   4. tests/run.py -- the project's own front door: every suite, pure and IDA,
#      ~185s. It used to be only the scenario suite here, and that gap cost a
#      real regression: a faster worker connect left the loading overlay up a
#      moment longer relative to the index finishing, and project mode's first
#      keypress landed on the overlay. Only test_project_ui.py covers that, and
#      it was not being run.
#
# Only failures reach stdout: the agent sees the last 80 lines on failure, and a
# wall of "ok" would push the actual break out of view. A file that fails is
# re-run ALONE before it counts -- the IDA suites share a loaded box, and a
# worker that got CPU-starved mid-analysis reads as a failure but is not one
# (see the note in tests/run.py).
set -euo pipefail
cd "$(dirname "$0")/.."

PY="${IDATUI_PYTHON:-$HOME/ida-venv/bin/python}"

search=$("$PY" .auto/check_search.py targets/echo 2>&1) || {
    echo "--- search fast paths disagree with the plain loop ---"
    echo "$search" | tail -20
    exit 1
}
echo "$search" | tail -1

rn=$("$PY" .auto/check_rename.py targets/echo 2>&1) || {
    echo "--- the listing is wrong after a rename ---"
    echo "$rn" | tail -20
    exit 1
}
echo "$rn" | tail -1

ed=$("$PY" .auto/check_edit.py targets/echo 2>&1) || {
    echo "--- the listing is wrong after an item edit ---"
    echo "$ed" | tail -20
    exit 1
}
echo "$ed" | tail -1

out=$(python3 tests/run.py 2>&1) || true
echo "$out" | tail -2

# Files run.py marked bad, e.g. "  FAIL  scenarios   301 passed, 2 failed"
# run.py's summary says "N passed[, M failed]..."; no "failed" clause == green.
if ! echo "$out" | tail -3 | grep -q "failed"; then
    exit 0
fi

echo "--- first pass failures ---"
echo "$out" | grep -E "^  FAIL" | head -20
# run.py names them itself: "failing files: formats scenarios" (colourised).
files=$(echo "$out" | sed -e 's/\x1b\[[0-9;]*m//g' \
        | sed -n 's/^failing files: *//p' | tr '\n' ' ')
if [ -z "$files" ]; then
    echo "--- could not identify the failing file; full tail ---"
    echo "$out" | tail -30
    exit 1
fi
echo "--- retrying alone: $files ---"
retry=$(python3 tests/run.py $files 2>&1) || true
if echo "$retry" | tail -3 | grep -q "failed"; then
    echo "--- REAL regression ---"
    echo "$retry" | grep -E "^  FAIL" | head -30
    echo "$retry" | tail -3
    exit 1
fi
echo "flake: [$files] pass in isolation"