Issue #13165: stringbench is now available in the Tools/stringbench folder.

It used to live in its own SVN project.

4 files changed, 1560 insertions, 0 deletions

diff --git a/Misc/NEWS b/Misc/NEWS
index 3585764..47628d0 100644
--- a/Misc/NEWS
+++ b/Misc/NEWS
@@ -57,6 +57,12 @@ Tests
 - Issue #14355: Regrtest now supports the standard unittest test loading, and
   will use it if a test file contains no `test_main` method.
 
+Tools / Demos
+-------------
+
+- Issue #13165: stringbench is now available in the Tools/stringbench folder.
+  It used to live in its own SVN project.
+
 
 What's New in Python 3.3.0 Alpha 2?
 ===================================
diff --git a/Tools/README b/Tools/README
index c1f89ba..44050c6 100644
--- a/Tools/README
+++ b/Tools/README
@@ -32,6 +32,9 @@ scripts         A number of useful single-file programs, e.g. tabnanny.py
                 tabs and spaces, and 2to3, which converts Python 2 code
                 to Python 3 code.
 
+stringbench     A suite of micro-benchmarks for various operations on
+                strings (both 8-bit and unicode).
+
 test2to3        A demonstration of how to use 2to3 transparently in setup.py.
 
 unicode         Tools for generating unicodedata and codecs from unicode.org
diff --git a/Tools/stringbench/README b/Tools/stringbench/README
new file mode 100644
index 0000000..a271f12
--- /dev/null
+++ b/Tools/stringbench/README
@@ -0,0 +1,68 @@
+stringbench is a set of performance tests comparing byte string
+operations with unicode operations.  The two string implementations
+are loosely based on each other and sometimes the algorithm for one is
+faster than the other.
+
+These test set was started at the Need For Speed sprint in Reykjavik
+to identify which string methods could be sped up quickly and to
+identify obvious places for improvement.
+
+Here is an example of a benchmark
+
+
+@bench('"Andrew".startswith("A")', 'startswith single character', 1000)
+def startswith_single(STR):
+    s1 = STR("Andrew")
+    s2 = STR("A")
+    s1_startswith = s1.startswith
+    for x in _RANGE_1000:
+        s1_startswith(s2)
+
+The bench decorator takes three parameters.  The first is a short
+description of how the code works.  In most cases this is Python code
+snippet.  It is not the code which is actually run because the real
+code is hand-optimized to focus on the method being tested.
+
+The second parameter is a group title.  All benchmarks with the same
+group title are listed together.  This lets you compare different
+implementations of the same algorithm, such as "t in s"
+vs. "s.find(t)".
+
+The last is a count.  Each benchmark loops over the algorithm either
+100 or 1000 times, depending on the algorithm performance.  The output
+time is the time per benchmark call so the reader needs a way to know
+how to scale the performance.
+
+These parameters become function attributes.
+
+
+Here is an example of the output
+
+
+========== count newlines
+38.54   41.60   92.7    ...text.with.2000.newlines.count("\n") (*100)
+========== early match, single character
+1.14    1.18    96.8    ("A"*1000).find("A") (*1000)
+0.44    0.41    105.6   "A" in "A"*1000 (*1000)
+1.15    1.17    98.1    ("A"*1000).index("A") (*1000)
+
+The first column is the run time in milliseconds for byte strings.
+The second is the run time for unicode strings.  The third is a
+percentage; byte time / unicode time.  It's the percentage by which
+unicode is faster than byte strings.
+
+The last column contains the code snippet and the repeat count for the
+internal benchmark loop.
+
+The times are computed with 'timeit.py' which repeats the test more
+and more times until the total time takes over 0.2 seconds, returning
+the best time for a single iteration.
+
+The final line of the output is the cumulative time for byte and
+unicode strings, and the overall performance of unicode relative to
+bytes.  For example
+
+4079.83 5432.25 75.1    TOTAL
+
+However, this has no meaning as it evenly weights every test.
+
diff --git a/Tools/stringbench/stringbench.py b/Tools/stringbench/stringbench.py
new file mode 100755
index 0000000..0098bb1
--- /dev/null
+++ b/Tools/stringbench/stringbench.py
@@ -0,0 +1,1483 @@
+
+# Various microbenchmarks comparing unicode and byte string performance
+# Please keep this file both 2.x and 3.x compatible!
+
+import timeit
+import itertools
+import operator
+import re
+import sys
+import datetime
+import optparse
+
+VERSION = '2.0'
+
+def p(*args):
+    sys.stdout.write(' '.join(str(s) for s in args) + '\n')
+
+if sys.version_info >= (3,):
+    BYTES = bytes_from_str = lambda x: x.encode('ascii')
+    UNICODE = unicode_from_str = lambda x: x
+else:
+    BYTES = bytes_from_str = lambda x: x
+    UNICODE = unicode_from_str = lambda x: x.decode('ascii')
+
+class UnsupportedType(TypeError):
+    pass
+
+
+p('stringbench v%s' % VERSION)
+p(sys.version)
+p(datetime.datetime.now())
+
+REPEAT = 1
+REPEAT = 3
+#REPEAT = 7
+
+if __name__ != "__main__":
+    raise SystemExit("Must run as main program")
+
+parser = optparse.OptionParser()
+parser.add_option("-R", "--skip-re", dest="skip_re",
+                  action="store_true",
+                  help="skip regular expression tests")
+parser.add_option("-8", "--8-bit", dest="bytes_only",
+                  action="store_true",
+                  help="only do 8-bit string benchmarks")
+parser.add_option("-u", "--unicode", dest="unicode_only",
+                  action="store_true",
+                  help="only do Unicode string benchmarks")
+
+
+_RANGE_1000 = list(range(1000))
+_RANGE_100 = list(range(100))
+_RANGE_10 = list(range(10))
+
+dups = {}
+def bench(s, group, repeat_count):
+    def blah(f):
+        if f.__name__ in dups:
+            raise AssertionError("Multiple functions with same name: %r" %
+                                 (f.__name__,))
+        dups[f.__name__] = 1
+        f.comment = s
+        f.is_bench = True
+        f.group = group
+        f.repeat_count = repeat_count
+        return f
+    return blah
+
+def uses_re(f):
+    f.uses_re = True
+
+####### 'in' comparisons
+
+@bench('"A" in "A"*1000', "early match, single character", 1000)
+def in_test_quick_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("A")
+    for x in _RANGE_1000:
+        s2 in s1
+
+@bench('"B" in "A"*1000', "no match, single character", 1000)
+def in_test_no_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("B")
+    for x in _RANGE_1000:
+        s2 in s1
+
+
+@bench('"AB" in "AB"*1000', "early match, two characters", 1000)
+def in_test_quick_match_two_characters(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("AB")
+    for x in _RANGE_1000:
+        s2 in s1
+
+@bench('"BC" in "AB"*1000', "no match, two characters", 1000)
+def in_test_no_match_two_character(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("BC")
+    for x in _RANGE_1000:
+        s2 in s1
+
+@bench('"BC" in ("AB"*300+"C")', "late match, two characters", 1000)
+def in_test_slow_match_two_characters(STR):
+    s1 = STR("AB" * 300+"C")
+    s2 = STR("BC")
+    for x in _RANGE_1000:
+        s2 in s1
+
+@bench('s="ABC"*33; (s+"E") in ((s+"D")*300+s+"E")',
+       "late match, 100 characters", 100)
+def in_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = (m+d)*300 + m+e
+    s2 = m+e
+    for x in _RANGE_100:
+        s2 in s1
+
+# Try with regex
+@uses_re
+@bench('s="ABC"*33; re.compile(s+"D").search((s+"D")*300+s+"E")',
+       "late match, 100 characters", 100)
+def re_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = (m+d)*300 + m+e
+    s2 = m+e
+    pat = re.compile(s2)
+    search = pat.search
+    for x in _RANGE_100:
+        search(s1)
+
+
+#### same tests as 'in' but use 'find'
+
+@bench('("A"*1000).find("A")', "early match, single character", 1000)
+def find_test_quick_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("A")
+    s1_find = s1.find
+    for x in _RANGE_1000:
+        s1_find(s2)
+
+@bench('("A"*1000).find("B")', "no match, single character", 1000)
+def find_test_no_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("B")
+    s1_find = s1.find
+    for x in _RANGE_1000:
+        s1_find(s2)
+
+
+@bench('("AB"*1000).find("AB")', "early match, two characters", 1000)
+def find_test_quick_match_two_characters(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("AB")
+    s1_find = s1.find
+    for x in _RANGE_1000:
+        s1_find(s2)
+
+@bench('("AB"*1000).find("BC")', "no match, two characters", 1000)
+def find_test_no_match_two_character(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("BC")
+    s1_find = s1.find
+    for x in _RANGE_1000:
+        s1_find(s2)
+
+@bench('("AB"*1000).find("CA")', "no match, two characters", 1000)
+def find_test_no_match_two_character_bis(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("CA")
+    s1_find = s1.find
+    for x in _RANGE_1000:
+        s1_find(s2)
+
+@bench('("AB"*300+"C").find("BC")', "late match, two characters", 1000)
+def find_test_slow_match_two_characters(STR):
+    s1 = STR("AB" * 300+"C")
+    s2 = STR("BC")
+    s1_find = s1.find
+    for x in _RANGE_1000:
+        s1_find(s2)
+
+@bench('("AB"*300+"CA").find("CA")', "late match, two characters", 1000)
+def find_test_slow_match_two_characters_bis(STR):
+    s1 = STR("AB" * 300+"CA")
+    s2 = STR("CA")
+    s1_find = s1.find
+    for x in _RANGE_1000:
+        s1_find(s2)
+
+@bench('s="ABC"*33; ((s+"D")*500+s+"E").find(s+"E")',
+       "late match, 100 characters", 100)
+def find_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = (m+d)*500 + m+e
+    s2 = m+e
+    s1_find = s1.find
+    for x in _RANGE_100:
+        s1_find(s2)
+
+@bench('s="ABC"*33; ((s+"D")*500+"E"+s).find("E"+s)',
+       "late match, 100 characters", 100)
+def find_test_slow_match_100_characters_bis(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = (m+d)*500 + e+m
+    s2 = e+m
+    s1_find = s1.find
+    for x in _RANGE_100:
+        s1_find(s2)
+
+
+#### Same tests for 'rfind'
+
+@bench('("A"*1000).rfind("A")', "early match, single character", 1000)
+def rfind_test_quick_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("A")
+    s1_rfind = s1.rfind
+    for x in _RANGE_1000:
+        s1_rfind(s2)
+
+@bench('("A"*1000).rfind("B")', "no match, single character", 1000)
+def rfind_test_no_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("B")
+    s1_rfind = s1.rfind
+    for x in _RANGE_1000:
+        s1_rfind(s2)
+
+
+@bench('("AB"*1000).rfind("AB")', "early match, two characters", 1000)
+def rfind_test_quick_match_two_characters(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("AB")
+    s1_rfind = s1.rfind
+    for x in _RANGE_1000:
+        s1_rfind(s2)
+
+@bench('("AB"*1000).rfind("BC")', "no match, two characters", 1000)
+def rfind_test_no_match_two_character(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("BC")
+    s1_rfind = s1.rfind
+    for x in _RANGE_1000:
+        s1_rfind(s2)
+
+@bench('("AB"*1000).rfind("CA")', "no match, two characters", 1000)
+def rfind_test_no_match_two_character_bis(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("CA")
+    s1_rfind = s1.rfind
+    for x in _RANGE_1000:
+        s1_rfind(s2)
+
+@bench('("C"+"AB"*300).rfind("CA")', "late match, two characters", 1000)
+def rfind_test_slow_match_two_characters(STR):
+    s1 = STR("C" + "AB" * 300)
+    s2 = STR("CA")
+    s1_rfind = s1.rfind
+    for x in _RANGE_1000:
+        s1_rfind(s2)
+
+@bench('("BC"+"AB"*300).rfind("BC")', "late match, two characters", 1000)
+def rfind_test_slow_match_two_characters_bis(STR):
+    s1 = STR("BC" + "AB" * 300)
+    s2 = STR("BC")
+    s1_rfind = s1.rfind
+    for x in _RANGE_1000:
+        s1_rfind(s2)
+
+@bench('s="ABC"*33; ("E"+s+("D"+s)*500).rfind("E"+s)',
+       "late match, 100 characters", 100)
+def rfind_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = e+m + (d+m)*500
+    s2 = e+m
+    s1_rfind = s1.rfind
+    for x in _RANGE_100:
+        s1_rfind(s2)
+
+@bench('s="ABC"*33; (s+"E"+("D"+s)*500).rfind(s+"E")',
+       "late match, 100 characters", 100)
+def rfind_test_slow_match_100_characters_bis(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = m+e + (d+m)*500
+    s2 = m+e
+    s1_rfind = s1.rfind
+    for x in _RANGE_100:
+        s1_rfind(s2)
+
+
+#### Now with index.
+# Skip the ones which fail because that would include exception overhead.
+
+@bench('("A"*1000).index("A")', "early match, single character", 1000)
+def index_test_quick_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("A")
+    s1_index = s1.index
+    for x in _RANGE_1000:
+        s1_index(s2)
+
+@bench('("AB"*1000).index("AB")', "early match, two characters", 1000)
+def index_test_quick_match_two_characters(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("AB")
+    s1_index = s1.index
+    for x in _RANGE_1000:
+        s1_index(s2)
+
+@bench('("AB"*300+"C").index("BC")', "late match, two characters", 1000)
+def index_test_slow_match_two_characters(STR):
+    s1 = STR("AB" * 300+"C")
+    s2 = STR("BC")
+    s1_index = s1.index
+    for x in _RANGE_1000:
+        s1_index(s2)
+
+@bench('s="ABC"*33; ((s+"D")*500+s+"E").index(s+"E")',
+       "late match, 100 characters", 100)
+def index_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = (m+d)*500 + m+e
+    s2 = m+e
+    s1_index = s1.index
+    for x in _RANGE_100:
+        s1_index(s2)
+
+
+#### Same for rindex
+
+@bench('("A"*1000).rindex("A")', "early match, single character", 1000)
+def rindex_test_quick_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("A")
+    s1_rindex = s1.rindex
+    for x in _RANGE_1000:
+        s1_rindex(s2)
+
+@bench('("AB"*1000).rindex("AB")', "early match, two characters", 1000)
+def rindex_test_quick_match_two_characters(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("AB")
+    s1_rindex = s1.rindex
+    for x in _RANGE_1000:
+        s1_rindex(s2)
+
+@bench('("C"+"AB"*300).rindex("CA")', "late match, two characters", 1000)
+def rindex_test_slow_match_two_characters(STR):
+    s1 = STR("C" + "AB" * 300)
+    s2 = STR("CA")
+    s1_rindex = s1.rindex
+    for x in _RANGE_1000:
+        s1_rindex(s2)
+
+@bench('s="ABC"*33; ("E"+s+("D"+s)*500).rindex("E"+s)',
+       "late match, 100 characters", 100)
+def rindex_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = e + m + (d+m)*500
+    s2 = e + m
+    s1_rindex = s1.rindex
+    for x in _RANGE_100:
+        s1_rindex(s2)
+
+
+#### Same for partition
+
+@bench('("A"*1000).partition("A")', "early match, single character", 1000)
+def partition_test_quick_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("A")
+    s1_partition = s1.partition
+    for x in _RANGE_1000:
+        s1_partition(s2)
+
+@bench('("A"*1000).partition("B")', "no match, single character", 1000)
+def partition_test_no_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("B")
+    s1_partition = s1.partition
+    for x in _RANGE_1000:
+        s1_partition(s2)
+
+
+@bench('("AB"*1000).partition("AB")', "early match, two characters", 1000)
+def partition_test_quick_match_two_characters(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("AB")
+    s1_partition = s1.partition
+    for x in _RANGE_1000:
+        s1_partition(s2)
+
+@bench('("AB"*1000).partition("BC")', "no match, two characters", 1000)
+def partition_test_no_match_two_character(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("BC")
+    s1_partition = s1.partition
+    for x in _RANGE_1000:
+        s1_partition(s2)
+
+@bench('("AB"*300+"C").partition("BC")', "late match, two characters", 1000)
+def partition_test_slow_match_two_characters(STR):
+    s1 = STR("AB" * 300+"C")
+    s2 = STR("BC")
+    s1_partition = s1.partition
+    for x in _RANGE_1000:
+        s1_partition(s2)
+
+@bench('s="ABC"*33; ((s+"D")*500+s+"E").partition(s+"E")',
+       "late match, 100 characters", 100)
+def partition_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = (m+d)*500 + m+e
+    s2 = m+e
+    s1_partition = s1.partition
+    for x in _RANGE_100:
+        s1_partition(s2)
+
+
+#### Same for rpartition
+
+@bench('("A"*1000).rpartition("A")', "early match, single character", 1000)
+def rpartition_test_quick_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("A")
+    s1_rpartition = s1.rpartition
+    for x in _RANGE_1000:
+        s1_rpartition(s2)
+
+@bench('("A"*1000).rpartition("B")', "no match, single character", 1000)
+def rpartition_test_no_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("B")
+    s1_rpartition = s1.rpartition
+    for x in _RANGE_1000:
+        s1_rpartition(s2)
+
+
+@bench('("AB"*1000).rpartition("AB")', "early match, two characters", 1000)
+def rpartition_test_quick_match_two_characters(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("AB")
+    s1_rpartition = s1.rpartition
+    for x in _RANGE_1000:
+        s1_rpartition(s2)
+
+@bench('("AB"*1000).rpartition("BC")', "no match, two characters", 1000)
+def rpartition_test_no_match_two_character(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("BC")
+    s1_rpartition = s1.rpartition
+    for x in _RANGE_1000:
+        s1_rpartition(s2)
+
+@bench('("C"+"AB"*300).rpartition("CA")', "late match, two characters", 1000)
+def rpartition_test_slow_match_two_characters(STR):
+    s1 = STR("C" + "AB" * 300)
+    s2 = STR("CA")
+    s1_rpartition = s1.rpartition
+    for x in _RANGE_1000:
+        s1_rpartition(s2)
+
+@bench('s="ABC"*33; ("E"+s+("D"+s)*500).rpartition("E"+s)',
+       "late match, 100 characters", 100)
+def rpartition_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = e + m + (d+m)*500
+    s2 = e + m
+    s1_rpartition = s1.rpartition
+    for x in _RANGE_100:
+        s1_rpartition(s2)
+
+
+#### Same for split(s, 1)
+
+@bench('("A"*1000).split("A", 1)', "early match, single character", 1000)
+def split_test_quick_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("A")
+    s1_split = s1.split
+    for x in _RANGE_1000:
+        s1_split(s2, 1)
+
+@bench('("A"*1000).split("B", 1)', "no match, single character", 1000)
+def split_test_no_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("B")
+    s1_split = s1.split
+    for x in _RANGE_1000:
+        s1_split(s2, 1)
+
+
+@bench('("AB"*1000).split("AB", 1)', "early match, two characters", 1000)
+def split_test_quick_match_two_characters(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("AB")
+    s1_split = s1.split
+    for x in _RANGE_1000:
+        s1_split(s2, 1)
+
+@bench('("AB"*1000).split("BC", 1)', "no match, two characters", 1000)
+def split_test_no_match_two_character(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("BC")
+    s1_split = s1.split
+    for x in _RANGE_1000:
+        s1_split(s2, 1)
+
+@bench('("AB"*300+"C").split("BC", 1)', "late match, two characters", 1000)
+def split_test_slow_match_two_characters(STR):
+    s1 = STR("AB" * 300+"C")
+    s2 = STR("BC")
+    s1_split = s1.split
+    for x in _RANGE_1000:
+        s1_split(s2, 1)
+
+@bench('s="ABC"*33; ((s+"D")*500+s+"E").split(s+"E", 1)',
+       "late match, 100 characters", 100)
+def split_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = (m+d)*500 + m+e
+    s2 = m+e
+    s1_split = s1.split
+    for x in _RANGE_100:
+        s1_split(s2, 1)
+
+
+#### Same for rsplit(s, 1)
+
+@bench('("A"*1000).rsplit("A", 1)', "early match, single character", 1000)
+def rsplit_test_quick_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("A")
+    s1_rsplit = s1.rsplit
+    for x in _RANGE_1000:
+        s1_rsplit(s2, 1)
+
+@bench('("A"*1000).rsplit("B", 1)', "no match, single character", 1000)
+def rsplit_test_no_match_single_character(STR):
+    s1 = STR("A" * 1000)
+    s2 = STR("B")
+    s1_rsplit = s1.rsplit
+    for x in _RANGE_1000:
+        s1_rsplit(s2, 1)
+
+
+@bench('("AB"*1000).rsplit("AB", 1)', "early match, two characters", 1000)
+def rsplit_test_quick_match_two_characters(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("AB")
+    s1_rsplit = s1.rsplit
+    for x in _RANGE_1000:
+        s1_rsplit(s2, 1)
+
+@bench('("AB"*1000).rsplit("BC", 1)', "no match, two characters", 1000)
+def rsplit_test_no_match_two_character(STR):
+    s1 = STR("AB" * 1000)
+    s2 = STR("BC")
+    s1_rsplit = s1.rsplit
+    for x in _RANGE_1000:
+        s1_rsplit(s2, 1)
+
+@bench('("C"+"AB"*300).rsplit("CA", 1)', "late match, two characters", 1000)
+def rsplit_test_slow_match_two_characters(STR):
+    s1 = STR("C" + "AB" * 300)
+    s2 = STR("CA")
+    s1_rsplit = s1.rsplit
+    for x in _RANGE_1000:
+        s1_rsplit(s2, 1)
+
+@bench('s="ABC"*33; ("E"+s+("D"+s)*500).rsplit("E"+s, 1)',
+       "late match, 100 characters", 100)
+def rsplit_test_slow_match_100_characters(STR):
+    m = STR("ABC"*33)
+    d = STR("D")
+    e = STR("E")
+    s1 = e + m + (d+m)*500
+    s2 = e + m
+    s1_rsplit = s1.rsplit
+    for x in _RANGE_100:
+        s1_rsplit(s2, 1)
+
+
+#### Benchmark the operator-based methods
+
+@bench('"A"*10', "repeat 1 character 10 times", 1000)
+def repeat_single_10_times(STR):
+    s = STR("A")
+    for x in _RANGE_1000:
+        s * 10
+
+@bench('"A"*1000', "repeat 1 character 1000 times", 1000)
+def repeat_single_1000_times(STR):
+    s = STR("A")
+    for x in _RANGE_1000:
+        s * 1000
+
+@bench('"ABCDE"*10', "repeat 5 characters 10 times", 1000)
+def repeat_5_10_times(STR):
+    s = STR("ABCDE")
+    for x in _RANGE_1000:
+        s * 10
+
+@bench('"ABCDE"*1000', "repeat 5 characters 1000 times", 1000)
+def repeat_5_1000_times(STR):
+    s = STR("ABCDE")
+    for x in _RANGE_1000:
+        s * 1000
+
+# + for concat
+
+@bench('"Andrew"+"Dalke"', "concat two strings", 1000)
+def concat_two_strings(STR):
+    s1 = STR("Andrew")
+    s2 = STR("Dalke")
+    for x in _RANGE_1000:
+        s1+s2
+
+@bench('s1+s2+s3+s4+...+s20', "concat 20 strings of words length 4 to 15",
+       1000)
+def concat_many_strings(STR):
+    s1=STR('TIXSGYNREDCVBHJ')
+    s2=STR('PUMTLXBZVDO')
+    s3=STR('FVZNJ')
+    s4=STR('OGDXUW')
+    s5=STR('WEIMRNCOYVGHKB')
+    s6=STR('FCQTNMXPUZH')
+    s7=STR('TICZJYRLBNVUEAK')
+    s8=STR('REYB')
+    s9=STR('PWUOQ')
+    s10=STR('EQHCMKBS')
+    s11=STR('AEVDFOH')
+    s12=STR('IFHVD')
+    s13=STR('JGTCNLXWOHQ')
+    s14=STR('ITSKEPYLROZAWXF')
+    s15=STR('THEK')
+    s16=STR('GHPZFBUYCKMNJIT')
+    s17=STR('JMUZ')
+    s18=STR('WLZQMTB')
+    s19=STR('KPADCBW')
+    s20=STR('TNJHZQAGBU')
+    for x in _RANGE_1000:
+        (s1 + s2+ s3+ s4+ s5+ s6+ s7+ s8+ s9+s10+
+         s11+s12+s13+s14+s15+s16+s17+s18+s19+s20)
+
+
+#### Benchmark join
+
+def get_bytes_yielding_seq(STR, arg):
+    if STR is BYTES and sys.version_info >= (3,):
+        raise UnsupportedType
+    return STR(arg)
+
+@bench('"A".join("")',
+       "join empty string, with 1 character sep", 100)
+def join_empty_single(STR):
+    sep = STR("A")
+    s2 = get_bytes_yielding_seq(STR, "")
+    sep_join = sep.join
+    for x in _RANGE_100:
+        sep_join(s2)
+
+@bench('"ABCDE".join("")',
+       "join empty string, with 5 character sep", 100)
+def join_empty_5(STR):
+    sep = STR("ABCDE")
+    s2 = get_bytes_yielding_seq(STR, "")
+    sep_join = sep.join
+    for x in _RANGE_100:
+        sep_join(s2)
+
+@bench('"A".join("ABC..Z")',
+       "join string with 26 characters, with 1 character sep", 1000)
+def join_alphabet_single(STR):
+    sep = STR("A")
+    s2 = get_bytes_yielding_seq(STR, "ABCDEFGHIJKLMnOPQRSTUVWXYZ")
+    sep_join = sep.join
+    for x in _RANGE_1000:
+        sep_join(s2)
+
+@bench('"ABCDE".join("ABC..Z")',
+       "join string with 26 characters, with 5 character sep", 1000)
+def join_alphabet_5(STR):
+    sep = STR("ABCDE")
+    s2 = get_bytes_yielding_seq(STR, "ABCDEFGHIJKLMnOPQRSTUVWXYZ")
+    sep_join = sep.join
+    for x in _RANGE_1000:
+        sep_join(s2)
+
+@bench('"A".join(list("ABC..Z"))',
+       "join list of 26 characters, with 1 character sep", 1000)
+def join_alphabet_list_single(STR):
+    sep = STR("A")
+    s2 = [STR(x) for x in "ABCDEFGHIJKLMnOPQRSTUVWXYZ"]
+    sep_join = sep.join
+    for x in _RANGE_1000:
+        sep_join(s2)
+
+@bench('"ABCDE".join(list("ABC..Z"))',
+       "join list of 26 characters, with 5 character sep", 1000)
+def join_alphabet_list_five(STR):
+    sep = STR("ABCDE")
+    s2 = [STR(x) for x in "ABCDEFGHIJKLMnOPQRSTUVWXYZ"]
+    sep_join = sep.join
+    for x in _RANGE_1000:
+        sep_join(s2)
+
+@bench('"A".join(["Bob"]*100))',
+       "join list of 100 words, with 1 character sep", 1000)
+def join_100_words_single(STR):
+    sep = STR("A")
+    s2 = [STR("Bob")]*100
+    sep_join = sep.join
+    for x in _RANGE_1000:
+        sep_join(s2)
+
+@bench('"ABCDE".join(["Bob"]*100))',
+       "join list of 100 words, with 5 character sep", 1000)
+def join_100_words_5(STR):
+    sep = STR("ABCDE")
+    s2 = [STR("Bob")]*100
+    sep_join = sep.join
+    for x in _RANGE_1000:
+        sep_join(s2)
+
+#### split tests
+
+@bench('("Here are some words. "*2).split()', "split whitespace (small)", 1000)
+def whitespace_split(STR):
+    s = STR("Here are some words. "*2)
+    s_split = s.split
+    for x in _RANGE_1000:
+        s_split()
+
+@bench('("Here are some words. "*2).rsplit()', "split whitespace (small)", 1000)
+def whitespace_rsplit(STR):
+    s = STR("Here are some words. "*2)
+    s_rsplit = s.rsplit
+    for x in _RANGE_1000:
+        s_rsplit()
+
+@bench('("Here are some words. "*2).split(None, 1)',
+       "split 1 whitespace", 1000)
+def whitespace_split_1(STR):
+    s = STR("Here are some words. "*2)
+    s_split = s.split
+    N = None
+    for x in _RANGE_1000:
+        s_split(N, 1)
+
+@bench('("Here are some words. "*2).rsplit(None, 1)',
+       "split 1 whitespace", 1000)
+def whitespace_rsplit_1(STR):
+    s = STR("Here are some words. "*2)
+    s_rsplit = s.rsplit
+    N = None
+    for x in _RANGE_1000:
+        s_rsplit(N, 1)
+
+@bench('("Here are some words. "*2).partition(" ")',
+       "split 1 whitespace", 1000)
+def whitespace_partition(STR):
+    sep = STR(" ")
+    s = STR("Here are some words. "*2)
+    s_partition = s.partition
+    for x in _RANGE_1000:
+        s_partition(sep)
+
+@bench('("Here are some words. "*2).rpartition(" ")',
+       "split 1 whitespace", 1000)
+def whitespace_rpartition(STR):
+    sep = STR(" ")
+    s = STR("Here are some words. "*2)
+    s_rpartition = s.rpartition
+    for x in _RANGE_1000:
+        s_rpartition(sep)
+
+human_text = """\
+Python is a dynamic object-oriented programming language that can be
+used for many kinds of software development. It offers strong support
+for integration with other languages and tools, comes with extensive
+standard libraries, and can be learned in a few days. Many Python
+programmers report substantial productivity gains and feel the language
+encourages the development of higher quality, more maintainable code.
+
+Python runs on Windows, Linux/Unix, Mac OS X, OS/2, Amiga, Palm
+Handhelds, and Nokia mobile phones. Python has also been ported to the
+Java and .NET virtual machines.
+
+Python is distributed under an OSI-approved open source license that
+makes it free to use, even for commercial products.
+"""*25
+human_text_bytes = bytes_from_str(human_text)
+human_text_unicode = unicode_from_str(human_text)
+def _get_human_text(STR):
+    if STR is UNICODE:
+        return human_text_unicode
+    if STR is BYTES:
+        return human_text_bytes
+    raise AssertionError
+
+@bench('human_text.split()', "split whitespace (huge)", 10)
+def whitespace_split_huge(STR):
+    s = _get_human_text(STR)
+    s_split = s.split
+    for x in _RANGE_10:
+        s_split()
+
+@bench('human_text.rsplit()', "split whitespace (huge)", 10)
+def whitespace_rsplit_huge(STR):
+    s = _get_human_text(STR)
+    s_rsplit = s.rsplit
+    for x in _RANGE_10:
+        s_rsplit()
+
+
+
+@bench('"this\\nis\\na\\ntest\\n".split("\\n")', "split newlines", 1000)
+def newlines_split(STR):
+    s = STR("this\nis\na\ntest\n")
+    s_split = s.split
+    nl = STR("\n")
+    for x in _RANGE_1000:
+        s_split(nl)
+
+
+@bench('"this\\nis\\na\\ntest\\n".rsplit("\\n")', "split newlines", 1000)
+def newlines_rsplit(STR):
+    s = STR("this\nis\na\ntest\n")
+    s_rsplit = s.rsplit
+    nl = STR("\n")
+    for x in _RANGE_1000:
+        s_rsplit(nl)
+
+@bench('"this\\nis\\na\\ntest\\n".splitlines()', "split newlines", 1000)
+def newlines_splitlines(STR):
+    s = STR("this\nis\na\ntest\n")
+    s_splitlines = s.splitlines
+    for x in _RANGE_1000:
+        s_splitlines()
+
+## split text with 2000 newlines
+
+def _make_2000_lines():
+    import random
+    r = random.Random(100)
+    chars = list(map(chr, range(32, 128)))
+    i = 0
+    while i < len(chars):
+        chars[i] = " "
+        i += r.randrange(9)
+    s = "".join(chars)
+    s = s*4
+    words = []
+    for i in range(2000):
+        start = r.randrange(96)
+        n = r.randint(5, 65)
+        words.append(s[start:start+n])
+    return "\n".join(words)+"\n"
+
+_text_with_2000_lines = _make_2000_lines()
+_text_with_2000_lines_bytes = bytes_from_str(_text_with_2000_lines)
+_text_with_2000_lines_unicode = unicode_from_str(_text_with_2000_lines)
+def _get_2000_lines(STR):
+    if STR is UNICODE:
+        return _text_with_2000_lines_unicode
+    if STR is BYTES:
+        return _text_with_2000_lines_bytes
+    raise AssertionError
+
+
+@bench('"...text...".split("\\n")', "split 2000 newlines", 10)
+def newlines_split_2000(STR):
+    s = _get_2000_lines(STR)
+    s_split = s.split
+    nl = STR("\n")
+    for x in _RANGE_10:
+        s_split(nl)
+
+@bench('"...text...".rsplit("\\n")', "split 2000 newlines", 10)
+def newlines_rsplit_2000(STR):
+    s = _get_2000_lines(STR)
+    s_rsplit = s.rsplit
+    nl = STR("\n")
+    for x in _RANGE_10:
+        s_rsplit(nl)
+
+@bench('"...text...".splitlines()', "split 2000 newlines", 10)
+def newlines_splitlines_2000(STR):
+    s = _get_2000_lines(STR)
+    s_splitlines = s.splitlines
+    for x in _RANGE_10:
+        s_splitlines()
+
+
+## split text on "--" characters
+@bench(
+    '"this--is--a--test--of--the--emergency--broadcast--system".split("--")',
+    "split on multicharacter separator (small)", 1000)
+def split_multichar_sep_small(STR):
+    s = STR("this--is--a--test--of--the--emergency--broadcast--system")
+    s_split = s.split
+    pat = STR("--")
+    for x in _RANGE_1000:
+        s_split(pat)
+@bench(
+    '"this--is--a--test--of--the--emergency--broadcast--system".rsplit("--")',
+    "split on multicharacter separator (small)", 1000)
+def rsplit_multichar_sep_small(STR):
+    s = STR("this--is--a--test--of--the--emergency--broadcast--system")
+    s_rsplit = s.rsplit
+    pat = STR("--")
+    for x in _RANGE_1000:
+        s_rsplit(pat)
+
+## split dna text on "ACTAT" characters
+@bench('dna.split("ACTAT")',
+       "split on multicharacter separator (dna)", 10)
+def split_multichar_sep_dna(STR):
+    s = _get_dna(STR)
+    s_split = s.split
+    pat = STR("ACTAT")
+    for x in _RANGE_10:
+        s_split(pat)
+
+@bench('dna.rsplit("ACTAT")',
+       "split on multicharacter separator (dna)", 10)
+def rsplit_multichar_sep_dna(STR):
+    s = _get_dna(STR)
+    s_rsplit = s.rsplit
+    pat = STR("ACTAT")
+    for x in _RANGE_10:
+        s_rsplit(pat)
+
+
+
+## split with limits
+
+GFF3_example = "\t".join([
+    "I", "Genomic_canonical", "region", "357208", "396183", ".", "+", ".",
+    "ID=Sequence:R119;note=Clone R119%3B Genbank AF063007;Name=R119"])
+
+@bench('GFF3_example.split("\\t")', "tab split", 1000)
+def tab_split_no_limit(STR):
+    sep = STR("\t")
+    s = STR(GFF3_example)
+    s_split = s.split
+    for x in _RANGE_1000:
+        s_split(sep)
+
+@bench('GFF3_example.split("\\t", 8)', "tab split", 1000)
+def tab_split_limit(STR):
+    sep = STR("\t")
+    s = STR(GFF3_example)
+    s_split = s.split
+    for x in _RANGE_1000:
+        s_split(sep, 8)
+
+@bench('GFF3_example.rsplit("\\t")', "tab split", 1000)
+def tab_rsplit_no_limit(STR):
+    sep = STR("\t")
+    s = STR(GFF3_example)
+    s_rsplit = s.rsplit
+    for x in _RANGE_1000:
+        s_rsplit(sep)
+
+@bench('GFF3_example.rsplit("\\t", 8)', "tab split", 1000)
+def tab_rsplit_limit(STR):
+    sep = STR("\t")
+    s = STR(GFF3_example)
+    s_rsplit = s.rsplit
+    for x in _RANGE_1000:
+        s_rsplit(sep, 8)
+
+#### Count characters
+
+@bench('...text.with.2000.newlines.count("\\n")',
+       "count newlines", 10)
+def count_newlines(STR):
+    s = _get_2000_lines(STR)
+    s_count = s.count
+    nl = STR("\n")
+    for x in _RANGE_10:
+        s_count(nl)
+
+# Orchid sequences concatenated, from Biopython
+_dna = """
+CGTAACAAGGTTTCCGTAGGTGAACCTGCGGAAGGATCATTGTTGAGATCACATAATAATTGATCGGGTT
+AATCTGGAGGATCTGTTTACTTTGGTCACCCATGAGCATTTGCTGTTGAAGTGACCTAGAATTGCCATCG
+AGCCTCCTTGGGAGCTTTCTTGTTGGCGAGATCTAAACCCTTGCCCGGCGCAGTTTTGCTCCAAGTCGTT
+TGACACATAATTGGTGAAGGGGGTGGCATCCTTCCCTGACCCTCCCCCAACTATTTTTTTAACAACTCTC
+AGCAACGGAGACTCAGTCTTCGGCAAATGCGATAAATGGTGTGAATTGCAGAATCCCGTGCACCATCGAG
+TCTTTGAACGCAAGTTGCGCCCGAGGCCATCAGGCCAAGGGCACGCCTGCCTGGGCATTGCGAGTCATAT
+CTCTCCCTTAACGAGGCTGTCCATACATACTGTTCAGCCGGTGCGGATGTGAGTTTGGCCCCTTGTTCTT
+TGGTACGGGGGGTCTAAGAGCTGCATGGGCTTTTGATGGTCCTAAATACGGCAAGAGGTGGACGAACTAT
+GCTACAACAAAATTGTTGTGCAGAGGCCCCGGGTTGTCGTATTAGATGGGCCACCGTAATCTGAAGACCC
+TTTTGAACCCCATTGGAGGCCCATCAACCCATGATCAGTTGATGGCCATTTGGTTGCGACCCCAGGTCAG
+GTGAGCAACAGCTGTCGTAACAAGGTTTCCGTAGGGTGAACTGCGGAAGGATCATTGTTGAGATCACATA
+ATAATTGATCGAGTTAATCTGGAGGATCTGTTTACTTGGGTCACCCATGGGCATTTGCTGTTGAAGTGAC
+CTAGATTTGCCATCGAGCCTCCTTGGGAGCATCCTTGTTGGCGATATCTAAACCCTCAATTTTTCCCCCA
+ATCAAATTACACAAAATTGGTGGAGGGGGTGGCATTCTTCCCTTACCCTCCCCCAAATATTTTTTTAACA
+ACTCTCAGCAACGGATATCTCAGCTCTTGCATCGATGAAGAACCCACCGAAATGCGATAAATGGTGTGAA
+TTGCAGAATCCCGTGAACCATCGAGTCTTTGAACGCAAGTTGCGCCCGAGGCCATCAGGCCAAGGGCACG
+CCTGCCTGGGCATTGCGAGTCATATCTCTCCCTTAACGAGGCTGTCCATACATACTGTTCAGCCGGTGCG
+GATGTGAGTTTGGCCCCTTGTTCTTTGGTACGGGGGGTCTAAGAGATGCATGGGCTTTTGATGGTCCTAA
+ATACGGCAAGAGGTGGACGAACTATGCTACAACAAAATTGTTGTGCAAAGGCCCCGGGTTGTCGTATAAG
+ATGGGCCACCGATATCTGAAGACCCTTTTGGACCCCATTGGAGCCCATCAACCCATGTCAGTTGATGGCC
+ATTCGTAACAAGGTTTCCGTAGGTGAACCTGCGGAAGGATCATTGTTGAGATCACATAATAATTGATCGA
+GTTAATCTGGAGGATCTGTTTACTTGGGTCACCCATGGGCATTTGCTGTTGAAGTGACCTAGATTTGCCA
+TCGAGCCTCCTTGGGAGCTTTCTTGTTGGCGATATCTAAACCCTTGCCCGGCAGAGTTTTGGGAATCCCG
+TGAACCATCGAGTCTTTGAACGCAAGTTGCGCCCGAGGCCATCAGGCCAAGGGCACGCCTGCCTGGGCAT
+TGCGAGTCATATCTCTCCCTTAACGAGGCTGTCCATACACACCTGTTCAGCCGGTGCGGATGTGAGTTTG
+GCCCCTTGTTCTTTGGTACGGGGGGTCTAAGAGCTGCATGGGCTTTTGATGGTCCTAAATACGGCAAGAG
+GTGGACGAACTATGCTACAACAAAATTGTTGTGCAAAGGCCCCGGGTTGTCGTATTAGATGGGCCACCAT
+AATCTGAAGACCCTTTTGAACCCCATTGGAGGCCCATCAACCCATGATCAGTTGATGGCCATTTGGTTGC
+GACCCAGTCAGGTGAGGGTAGGTGAACCTGCGGAAGGATCATTGTTGAGATCACATAATAATTGATCGAG
+TTAATCTGGAGGATCTGTTTACTTTGGTCACCCATGGGCATTTGCTGTTGAAGTGACCTAGATTTGCCAT
+CGAGCCTCCTTGGGAGCTTTCTTGTTGGCGAGATCTAAACCCTTGCCCGGCGGAGTTTGGCGCCAAGTCA
+TATGACACATAATTGGTGAAGGGGGTGGCATCCTGCCCTGACCCTCCCCAAATTATTTTTTTAACAACTC
+TCAGCAACGGATATCTCGGCTCTTGCATCGATGAAGAACGCAGCGAAATGCGATAAATGGTGTGAATTGC
+AGAATCCCGTGAACCATCGAGTCTTTGGAACGCAAGTTGCGCCCGAGGCCATCAGGCCAAGGGCACGCCT
+GCCTGGGCATTGGGAATCATATCTCTCCCCTAACGAGGCTATCCAAACATACTGTTCATCCGGTGCGGAT
+GTGAGTTTGGCCCCTTGTTCTTTGGTACCGGGGGTCTAAGAGCTGCATGGGCATTTGATGGTCCTCAAAA
+CGGCAAGAGGTGGACGAACTATGCCACAACAAAATTGTTGTCCCAAGGCCCCGGGTTGTCGTATTAGATG
+GGCCACCGTAACCTGAAGACCCTTTTGAACCCCATTGGAGGCCCATCAACCCATGATCAGTTGATGACCA
+TTTGTTGCGACCCCAGTCAGCTGAGCAACCCGCTGAGTGGAAGGTCATTGCCGATATCACATAATAATTG
+ATCGAGTTAATCTGGAGGATCTGTTTACTTGGTCACCCATGAGCATTTGCTGTTGAAGTGACCTAGATTT
+GCCATCGAGCCTCCTTGGGAGTTTTCTTGTTGGCGAGATCTAAACCCTTGCCCGGCGGAGTTGTGCGCCA
+AGTCATATGACACATAATTGGTGAAGGGGGTGGCATCCTGCCCTGACCCTCCCCAAATTATTTTTTTAAC
+AACTCTCAGCAACGGATATCTCGGCTCTTGCATCGATGAAGAACGCAGCGAAATGCGATAAATGGTGTGA
+ATTGCAGAATCCCGTGAACCATCGAGTCTTTGAACGCAAGTTGCGCCCGAGGCCATCAGGCCAAGGGCAC
+GCCTGCCTGGGCATTGCGAGTCATATCTCTCCCTTAACGAGGCTGTCCATACATACTGTTCATCCGGTGC
+GGATGTGAGTTTGGCCCCTTGTTCTTTGGTACGGGGGGTCTAAGAGCTGCATGGGCATTTGATGGTCCTC
+AAAACGGCAAGAGGTGGACGAACTATGCTACAACCAAATTGTTGTCCCAAGGCCCCGGGTTGTCGTATTA
+GATGGGCCACCGTAACCTGAAGACCCTTTTGAACCCCATTGGAGGCCCATCAACCCATGATCAGTTGATG
+ACCATGTGTTGCGACCCCAGTCAGCTGAGCAACGCGCTGAGCGTAACAAGGTTTCCGTAGGTGGACCTCC
+GGGAGGATCATTGTTGAGATCACATAATAATTGATCGAGGTAATCTGGAGGATCTGCATATTTTGGTCAC
+"""
+_dna = "".join(_dna.splitlines())
+_dna = _dna * 25
+_dna_bytes = bytes_from_str(_dna)
+_dna_unicode = unicode_from_str(_dna)
+
+def _get_dna(STR):
+    if STR is UNICODE:
+        return _dna_unicode
+    if STR is BYTES:
+        return _dna_bytes
+    raise AssertionError
+
+@bench('dna.count("AACT")', "count AACT substrings in DNA example", 10)
+def count_aact(STR):
+    seq = _get_dna(STR)
+    seq_count = seq.count
+    needle = STR("AACT")
+    for x in _RANGE_10:
+        seq_count(needle)
+
+##### startswith and endswith
+
+@bench('"Andrew".startswith("A")', 'startswith single character', 1000)
+def startswith_single(STR):
+    s1 = STR("Andrew")
+    s2 = STR("A")
+    s1_startswith = s1.startswith
+    for x in _RANGE_1000:
+        s1_startswith(s2)
+
+@bench('"Andrew".startswith("Andrew")', 'startswith multiple characters',
+       1000)
+def startswith_multiple(STR):
+    s1 = STR("Andrew")
+    s2 = STR("Andrew")
+    s1_startswith = s1.startswith
+    for x in _RANGE_1000:
+        s1_startswith(s2)
+
+@bench('"Andrew".startswith("Anders")',
+       'startswith multiple characters - not!', 1000)
+def startswith_multiple_not(STR):
+    s1 = STR("Andrew")
+    s2 = STR("Anders")
+    s1_startswith = s1.startswith
+    for x in _RANGE_1000:
+        s1_startswith(s2)
+
+
+# endswith
+
+@bench('"Andrew".endswith("w")', 'endswith single character', 1000)
+def endswith_single(STR):
+    s1 = STR("Andrew")
+    s2 = STR("w")
+    s1_endswith = s1.endswith
+    for x in _RANGE_1000:
+        s1_endswith(s2)
+
+@bench('"Andrew".endswith("Andrew")', 'endswith multiple characters', 1000)
+def endswith_multiple(STR):
+    s1 = STR("Andrew")
+    s2 = STR("Andrew")
+    s1_endswith = s1.endswith
+    for x in _RANGE_1000:
+        s1_endswith(s2)
+
+@bench('"Andrew".endswith("Anders")',
+       'endswith multiple characters - not!', 1000)
+def endswith_multiple_not(STR):
+    s1 = STR("Andrew")
+    s2 = STR("Anders")
+    s1_endswith = s1.endswith
+    for x in _RANGE_1000:
+        s1_endswith(s2)
+
+#### Strip
+
+@bench('"Hello!\\n".strip()', 'strip terminal newline', 1000)
+def terminal_newline_strip_right(STR):
+    s = STR("Hello!\n")
+    s_strip = s.strip
+    for x in _RANGE_1000:
+        s_strip()
+
+@bench('"Hello!\\n".rstrip()', 'strip terminal newline', 1000)
+def terminal_newline_rstrip(STR):
+    s = STR("Hello!\n")
+    s_rstrip = s.rstrip
+    for x in _RANGE_1000:
+        s_rstrip()
+
+@bench('"\\nHello!".strip()', 'strip terminal newline', 1000)
+def terminal_newline_strip_left(STR):
+    s = STR("\nHello!")
+    s_strip = s.strip
+    for x in _RANGE_1000:
+        s_strip()
+
+@bench('"\\nHello!\\n".strip()', 'strip terminal newline', 1000)
+def terminal_newline_strip_both(STR):
+    s = STR("\nHello!\n")
+    s_strip = s.strip
+    for x in _RANGE_1000:
+        s_strip()
+
+@bench('"\\nHello!".rstrip()', 'strip terminal newline', 1000)
+def terminal_newline_lstrip(STR):
+    s = STR("\nHello!")
+    s_lstrip = s.lstrip
+    for x in _RANGE_1000:
+        s_lstrip()
+
+@bench('s="Hello!\\n"; s[:-1] if s[-1]=="\\n" else s',
+       'strip terminal newline', 1000)
+def terminal_newline_if_else(STR):
+    s = STR("Hello!\n")
+    NL = STR("\n")
+    for x in _RANGE_1000:
+        s[:-1] if (s[-1] == NL) else s
+
+
+# Strip multiple spaces or tabs
+
+@bench('"Hello\\t   \\t".strip()', 'strip terminal spaces and tabs', 1000)
+def terminal_space_strip(STR):
+    s = STR("Hello\t   \t!")
+    s_strip = s.strip
+    for x in _RANGE_1000:
+        s_strip()
+
+@bench('"Hello\\t   \\t".rstrip()', 'strip terminal spaces and tabs', 1000)
+def terminal_space_rstrip(STR):
+    s = STR("Hello!\t   \t")
+    s_rstrip = s.rstrip
+    for x in _RANGE_1000:
+        s_rstrip()
+
+@bench('"\\t   \\tHello".rstrip()', 'strip terminal spaces and tabs', 1000)
+def terminal_space_lstrip(STR):
+    s = STR("\t   \tHello!")
+    s_lstrip = s.lstrip
+    for x in _RANGE_1000:
+        s_lstrip()
+
+
+#### replace
+@bench('"This is a test".replace(" ", "\\t")', 'replace single character',
+       1000)
+def replace_single_character(STR):
+    s = STR("This is a test!")
+    from_str = STR(" ")
+    to_str = STR("\t")
+    s_replace = s.replace
+    for x in _RANGE_1000:
+        s_replace(from_str, to_str)
+
+@uses_re
+@bench('re.sub(" ", "\\t", "This is a test"', 'replace single character',
+       1000)
+def replace_single_character_re(STR):
+    s = STR("This is a test!")
+    pat = re.compile(STR(" "))
+    to_str = STR("\t")
+    pat_sub = pat.sub
+    for x in _RANGE_1000:
+        pat_sub(to_str, s)
+
+@bench('"...text.with.2000.lines...replace("\\n", " ")',
+       'replace single character, big string', 10)
+def replace_single_character_big(STR):
+    s = _get_2000_lines(STR)
+    from_str = STR("\n")
+    to_str = STR(" ")
+    s_replace = s.replace
+    for x in _RANGE_10:
+        s_replace(from_str, to_str)
+
+@uses_re
+@bench('re.sub("\\n", " ", "...text.with.2000.lines...")',
+       'replace single character, big string', 10)
+def replace_single_character_big_re(STR):
+    s = _get_2000_lines(STR)
+    pat = re.compile(STR("\n"))
+    to_str = STR(" ")
+    pat_sub = pat.sub
+    for x in _RANGE_10:
+        pat_sub(to_str, s)
+
+
+@bench('dna.replace("ATC", "ATT")',
+       'replace multiple characters, dna', 10)
+def replace_multiple_characters_dna(STR):
+    seq = _get_dna(STR)
+    from_str = STR("ATC")
+    to_str = STR("ATT")
+    seq_replace = seq.replace
+    for x in _RANGE_10:
+        seq_replace(from_str, to_str)
+
+# This increases the character count
+@bench('"...text.with.2000.newlines...replace("\\n", "\\r\\n")',
+       'replace and expand multiple characters, big string', 10)
+def replace_multiple_character_big(STR):
+    s = _get_2000_lines(STR)
+    from_str = STR("\n")
+    to_str = STR("\r\n")
+    s_replace = s.replace
+    for x in _RANGE_10:
+        s_replace(from_str, to_str)
+
+
+# This decreases the character count
+@bench('"When shall we three meet again?".replace("ee", "")',
+       'replace/remove multiple characters', 1000)
+def replace_multiple_character_remove(STR):
+    s = STR("When shall we three meet again?")
+    from_str = STR("ee")
+    to_str = STR("")
+    s_replace = s.replace
+    for x in _RANGE_1000:
+        s_replace(from_str, to_str)
+
+
+big_s = "A" + ("Z"*128*1024)
+big_s_bytes = bytes_from_str(big_s)
+big_s_unicode = unicode_from_str(big_s)
+def _get_big_s(STR):
+    if STR is UNICODE: return big_s_unicode
+    if STR is BYTES: return big_s_bytes
+    raise AssertionError
+
+# The older replace implementation counted all matches in
+# the string even when it only neeed to make one replacement.
+@bench('("A" + ("Z"*128*1024)).replace("A", "BB", 1)',
+       'quick replace single character match', 10)
+def quick_replace_single_match(STR):
+    s = _get_big_s(STR)
+    from_str = STR("A")
+    to_str = STR("BB")
+    s_replace = s.replace
+    for x in _RANGE_10:
+        s_replace(from_str, to_str, 1)
+
+@bench('("A" + ("Z"*128*1024)).replace("AZZ", "BBZZ", 1)',
+       'quick replace multiple character match', 10)
+def quick_replace_multiple_match(STR):
+    s = _get_big_s(STR)
+    from_str = STR("AZZ")
+    to_str = STR("BBZZ")
+    s_replace = s.replace
+    for x in _RANGE_10:
+        s_replace(from_str, to_str, 1)
+
+
+####
+
+# CCP does a lot of this, for internationalisation of ingame messages.
+_format = "The %(thing)s is %(place)s the %(location)s."
+_format_dict = { "thing":"THING", "place":"PLACE", "location":"LOCATION", }
+_format_bytes = bytes_from_str(_format)
+_format_unicode = unicode_from_str(_format)
+_format_dict_bytes = dict((bytes_from_str(k), bytes_from_str(v)) for (k,v) in _format_dict.items())
+_format_dict_unicode = dict((unicode_from_str(k), unicode_from_str(v)) for (k,v) in _format_dict.items())
+
+def _get_format(STR):
+    if STR is UNICODE:
+        return _format_unicode
+    if STR is BYTES:
+        if sys.version_info >= (3,):
+            raise UnsupportedType
+        return _format_bytes
+    raise AssertionError
+
+def _get_format_dict(STR):
+    if STR is UNICODE:
+        return _format_dict_unicode
+    if STR is BYTES:
+        if sys.version_info >= (3,):
+            raise UnsupportedType
+        return _format_dict_bytes
+    raise AssertionError
+
+# Formatting.
+@bench('"The %(k1)s is %(k2)s the %(k3)s."%{"k1":"x","k2":"y","k3":"z",}',
+       'formatting a string type with a dict', 1000)
+def format_with_dict(STR):
+    s = _get_format(STR)
+    d = _get_format_dict(STR)
+    for x in _RANGE_1000:
+        s % d
+
+
+#### Upper- and lower- case conversion
+
+@bench('("Where in the world is Carmen San Deigo?"*10).lower()',
+       "case conversion -- rare", 1000)
+def lower_conversion_rare(STR):
+    s = STR("Where in the world is Carmen San Deigo?"*10)
+    s_lower = s.lower
+    for x in _RANGE_1000:
+        s_lower()
+
+@bench('("WHERE IN THE WORLD IS CARMEN SAN DEIGO?"*10).lower()',
+       "case conversion -- dense", 1000)
+def lower_conversion_dense(STR):
+    s = STR("WHERE IN THE WORLD IS CARMEN SAN DEIGO?"*10)
+    s_lower = s.lower
+    for x in _RANGE_1000:
+        s_lower()
+
+
+@bench('("wHERE IN THE WORLD IS cARMEN sAN dEIGO?"*10).upper()',
+       "case conversion -- rare", 1000)
+def upper_conversion_rare(STR):
+    s = STR("Where in the world is Carmen San Deigo?"*10)
+    s_upper = s.upper
+    for x in _RANGE_1000:
+        s_upper()
+
+@bench('("where in the world is carmen san deigo?"*10).upper()',
+       "case conversion -- dense", 1000)
+def upper_conversion_dense(STR):
+    s = STR("where in the world is carmen san deigo?"*10)
+    s_upper = s.upper
+    for x in _RANGE_1000:
+        s_upper()
+
+
+# end of benchmarks
+
+#################
+
+class BenchTimer(timeit.Timer):
+    def best(self, repeat=1):
+        for i in range(1, 10):
+            number = 10**i
+            x = self.timeit(number)
+            if x > 0.02:
+                break
+        times = [x]
+        for i in range(1, repeat):
+            times.append(self.timeit(number))
+        return min(times) / number
+
+def main():
+    (options, test_names) = parser.parse_args()
+    if options.bytes_only and options.unicode_only:
+        raise SystemExit("Only one of --8-bit and --unicode are allowed")
+
+    bench_functions = []
+    for (k,v) in globals().items():
+        if hasattr(v, "is_bench"):
+            if test_names:
+                for name in test_names:
+                    if name in v.group:
+                        break
+                else:
+                    # Not selected, ignore
+                    continue
+            if options.skip_re and hasattr(v, "uses_re"):
+                continue
+
+            bench_functions.append( (v.group, k, v) )
+    bench_functions.sort()
+
+    p("bytes\tunicode")
+    p("(in ms)\t(in ms)\t%\tcomment")
+
+    bytes_total = uni_total = 0.0
+
+    for title, group in itertools.groupby(bench_functions,
+                                      operator.itemgetter(0)):
+        # Flush buffer before each group
+        sys.stdout.flush()
+        p("="*10, title)
+        for (_, k, v) in group:
+            if hasattr(v, "is_bench"):
+                bytes_time = 0.0
+                bytes_time_s = " - "
+                if not options.unicode_only:
+                    try:
+                        bytes_time = BenchTimer("__main__.%s(__main__.BYTES)" % (k,),
+                                                "import __main__").best(REPEAT)
+                        bytes_time_s = "%.2f" % (1000 * bytes_time)
+                        bytes_total += bytes_time
+                    except UnsupportedType:
+                        bytes_time_s = "N/A"
+                uni_time = 0.0
+                uni_time_s = " - "
+                if not options.bytes_only:
+                    try:
+                        uni_time = BenchTimer("__main__.%s(__main__.UNICODE)" % (k,),
+                                              "import __main__").best(REPEAT)
+                        uni_time_s = "%.2f" % (1000 * uni_time)
+                        uni_total += uni_time
+                    except UnsupportedType:
+                        uni_time_s = "N/A"
+                try:
+                    average = bytes_time/uni_time
+                except (TypeError, ZeroDivisionError):
+                    average = 0.0
+                p("%s\t%s\t%.1f\t%s (*%d)" % (
+                    bytes_time_s, uni_time_s, 100.*average,
+                    v.comment, v.repeat_count))
+
+    if bytes_total == uni_total == 0.0:
+        p("That was zippy!")
+    else:
+        try:
+            ratio = bytes_total/uni_total
+        except ZeroDivisionError:
+            ratio = 0.0
+        p("%.2f\t%.2f\t%.1f\t%s" % (
+            1000*bytes_total, 1000*uni_total, 100.*ratio,
+            "TOTAL"))
+
+if __name__ == "__main__":
+    main()
+