From 759b30a9ac28d3114c2b19aff5141a79ec2a1c9f Mon Sep 17 00:00:00 2001
From: Wilhelm Schuermann <wimschuermann@googlemail.com>
Date: Mon, 26 Oct 2015 15:16:33 +0100
Subject: [PATCH] Performance: Add crude comparison script

---
 performance/compare_runs.py | 86 +++++++++++++++++++++++++++++++++++++
 1 file changed, 86 insertions(+)
 create mode 100755 performance/compare_runs.py

diff --git a/performance/compare_runs.py b/performance/compare_runs.py
new file mode 100755
index 000000000..cf4779342
--- /dev/null
+++ b/performance/compare_runs.py
@@ -0,0 +1,86 @@
+#!/usr/bin/env python3
+#
+# Compare performance of two Taskwarrior versions.
+#
+# Run without arguments for usage information.
+#
+# "Perf task" lines in debug output differing between the two versions is
+# poorly handled, if at all.  Best used for comparing performance between two
+# similar Taskwarrior versions only differing in performance related code.
+#
+# "total" will probably not be the sum of the shown individual parts, but
+# slightly larger since not all minimum values will have been present in the
+# same run.
+#
+
+import collections
+import re
+import sys
+
+# Adjust if more performance tests are added
+COMMANDS = "next list all add export import".split()
+
+TaskPerf = collections.namedtuple("TaskPerf", "version commit at timing")
+
+
+def parse_perf(input):
+    tests = {}
+    for command in COMMANDS:
+        tests[command] = []
+
+        # Parse concatenated run_perf output
+        for i in re.findall("^  - task %s\.\.\.\n"
+                            "Perf task ([^ ]+) ([^ ]+) ([^ ]+) (.+)$"
+                            % command, input, re.MULTILINE):
+            info = i[0:3] + ({k:v for k, v in (i.split(":") for i in i[-1].split())},)
+            pt = TaskPerf(*info)
+            tests[command].append(pt)
+    return tests
+
+
+def get_best(tests):
+    best = {}
+    for command in tests:
+        best[command] = {}
+        for k in tests[command][0].timing:
+            best[command][k] = min(k, *[t.timing[k] for t in tests[command]])
+    return best
+
+
+if len(sys.argv) != 3:
+    print("Usage:")
+    print(" $ %s file1 file2" % sys.argv[0])
+    print("Where file1, file2 are generated as such:")
+    print(" $ for i in `seq 20`; do ./run_perf >> filename 2>&1; done")
+    sys.exit(1)
+
+with open(sys.argv[1], "r") as fh:
+    tests_prev = parse_perf(fh.read())
+    best_prev = get_best(tests_prev)
+with open(sys.argv[2], "r") as fh:
+    tests_cur = parse_perf(fh.read())
+    best_cur = get_best(tests_cur)
+
+print("Previous: %s (%s)" % (tests_prev[COMMANDS[0]][0].version, tests_prev[COMMANDS[0]][0].commit))
+print("Current:  %s (%s)" % (tests_cur[COMMANDS[0]][0].version, tests_cur[COMMANDS[0]][0].commit))
+
+for test in COMMANDS:
+    print("# %s:" % test)
+
+    out = ["" for i in range(5)]
+    for k in sorted(best_prev[test].keys()):
+        diff = str(int(best_cur[test][k]) - int(best_prev[test][k]))
+
+        if float(best_prev[test][k]) > 0:
+            percentage = str(int((float(diff) / float(best_prev[test][k]) * 100))) + "%"
+        else:
+            percentage = "0%"
+
+        pad = max(map(len, (k, best_prev[test][k], best_cur[test][k], diff, percentage)))
+        out[0] += " %s" % k.rjust(pad)
+        out[1] += " %s" % best_prev[test][k].rjust(pad)
+        out[2] += " %s" % best_cur[test][k].rjust(pad)
+        out[3] += " %s" % diff.rjust(pad)
+        out[4] += " %s" % percentage.rjust(pad)
+    for line in out:
+        print(line)