Change perf test to be time based

This is similar to how go's benchmarks run.  The benchmark
is run until a particular duration goal is hit rather than number
of iterations.  That way you can know how long test runs are going
to take based on the number of tests running.

Right now, each phase of the benchmark (lex,parse,search,combined)
runs for 0.5 seconds.

Author: jamesls

perf/perftest.py

@@ -11,20 +11,17 @@
 import os
 import json
 import sys
+import timeit
 
-try:
-    _clock = time.process_time
-except AttributeError:
-    # Python 2.x does not have time.process_time
-    _clock = time.clock
+_clock = timeit.default_timer
 
 
 from jmespath.parser import Parser
 from jmespath.lexer import Lexer
 
 
 DIRECTORY = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'cases')
-DEFAULT_NUM_LOOP = 100
+APPROX_RUN_TIME = 0.5
 
 
 def run_tests(tests):
@@ -47,50 +44,63 @@ def run_tests(tests):
 
 
 def _lex_time(expression, clock=_clock):
-    best = float('inf')
     lex = Lexer()
-    for i in range(DEFAULT_NUM_LOOP):
+    duration = 0
+    i = 0
+    while True:
+        i += 1
         start = clock()
         list(lex.tokenize(expression))
         end = clock()
         total = end - start
-        if total < best:
-            best = total
-    return best
+        duration += total
+        if duration >= APPROX_RUN_TIME:
+            break
+    return duration / i
 
 
 def _search_time(expression, given, clock=_clock):
     p = Parser()
     parsed = p.parse(expression)
-    best = float('inf')
-    for i in range(DEFAULT_NUM_LOOP):
+    duration =  0
+    i = 0
+    while True:
+        i += 1
         start = clock()
         parsed.search(given)
         end = clock()
         total = end - start
-        if total < best:
-            best = total
-    return best
+        duration += total
+        if duration >= APPROX_RUN_TIME:
+            break
+    return duration / i
 
 
 def _parse_time(expression, clock=_clock):
     best = float('inf')
     p = Parser()
-    for i in range(DEFAULT_NUM_LOOP):
+    duration = 0
+    i = 0
+    while True:
+        i += 1
         p.purge()
         start = clock()
         p.parse(expression)
         end = clock()
         total = end - start
-        if total < best:
-            best = total
-    return best
+        duration += total
+        if duration >= APPROX_RUN_TIME:
+            break
+    return duration / i
 
 
 def _combined_time(expression, given, result, clock=_clock):
     best = float('inf')
     p = Parser()
-    for i in range(DEFAULT_NUM_LOOP):
+    duration = 0
+    i = 0
+    while True:
+        i += 1
         p.purge()
         start = clock()
         r = p.parse(expression).search(given)
@@ -99,9 +109,10 @@ def _combined_time(expression, given, result, clock=_clock):
         if r != result:
             raise RuntimeError("Unexpected result, received: %s, "
                                "expected: %s" % (r, result))
-        if total < best:
-            best = total
-    return best
+        duration += total
+        if duration >= APPROX_RUN_TIME:
+            break
+    return duration / i
 
 
 def load_tests(filename):