#!/usr/bin/env bash

# Our custom diff tool which normalizes result and expected files before
# doing the actual diff. Rules for normalization are in normalize.sed. See
# the comments there for more information.
#
# Note that src/test/regress/Makefile adds this directory to $PATH so
# pg_regress uses this diff tool instead of the system diff tool.
set -eu -o pipefail

file1="${@:(-2):1}"
file2="${@:(-1):1}"
# pg_regress passes the expected file as the first argument ($file1 above),
# and results file as the second argument ($file2 above). We take the base
# filename of the expected file as the test name. We can have multiple
# expected files for a single test with _0, _1, ... suffixes.
# So for the test name, we also strip the additional suffix.
test=$(basename "$file1" .out | sed -E "s/_[0-9]+$//")
args=${@:1:$#-2}
BASEDIR=$(dirname "$0")

# whereis searches for standard unix places before $PATH. So select the first
# entry as the original diff tool.
DIFF=$(whereis diff | sed "s/diff://g" | awk '{print $1}')
if [ -z "$DIFF" ]
then
    DIFF=/usr/bin/diff
fi

if test -z "${VANILLATEST:-}"
then
	touch "$file1"  # when adding a new test the expected file does not exist
	normalize_file="$BASEDIR/normalize.sed"
	# when running tests on an existing cluster some changes need to be done on
	# normalize.sed file. So a new file is used.
	if [[ -f "$BASEDIR/normalize_modified.sed" ]]
	then
		normalize_file="$BASEDIR/normalize_modified.sed"
	fi
	sed -Ef "$normalize_file" < "$file1" > "$file1.modified"
	sed -Ef "$normalize_file" < "$file2" > "$file2.modified"
	"$DIFF" -w $args "$file1.modified" "$file2.modified" | LC_CTYPE=C.UTF-8 diff-filter "$BASEDIR/normalize.sed"
	exit ${PIPESTATUS[0]}
else
	exec "$DIFF" -w $args "$file1" "$file2"
fi