summaryrefslogtreecommitdiffstats
path: root/Lib/test/test_re.py
diff options
context:
space:
mode:
authorGuido van Rossum <guido@python.org>2007-02-09 05:37:30 (GMT)
committerGuido van Rossum <guido@python.org>2007-02-09 05:37:30 (GMT)
commitbe19ed77ddb047e02fe94d142181062af6d99dcc (patch)
tree70f214e06554046fcccbadeb78665f25e07ce965 /Lib/test/test_re.py
parent452bf519a70c3db0e7f0d2540b1bfb07d9085583 (diff)
downloadcpython-be19ed77ddb047e02fe94d142181062af6d99dcc.zip
cpython-be19ed77ddb047e02fe94d142181062af6d99dcc.tar.gz
cpython-be19ed77ddb047e02fe94d142181062af6d99dcc.tar.bz2
Fix most trivially-findable print statements.
There's one major and one minor category still unfixed: doctests are the major category (and I hope to be able to augment the refactoring tool to refactor bona fide doctests soon); other code generating print statements in strings is the minor category. (Oh, and I don't know if the compiler package works.)
Diffstat (limited to 'Lib/test/test_re.py')
-rw-r--r--Lib/test/test_re.py28
1 files changed, 14 insertions, 14 deletions
diff --git a/Lib/test/test_re.py b/Lib/test/test_re.py
index dafd82e..bb97433 100644
--- a/Lib/test/test_re.py
+++ b/Lib/test/test_re.py
@@ -603,7 +603,7 @@ class ReTests(unittest.TestCase):
def run_re_tests():
from test.re_tests import benchmarks, tests, SUCCEED, FAIL, SYNTAX_ERROR
if verbose:
- print 'Running re_tests test suite'
+ print('Running re_tests test suite')
else:
# To save time, only run the first and last 10 tests
#tests = tests[:10] + tests[-10:]
@@ -624,23 +624,23 @@ def run_re_tests():
except re.error:
if outcome == SYNTAX_ERROR: pass # Expected a syntax error
else:
- print '=== Syntax error:', t
+ print('=== Syntax error:', t)
except KeyboardInterrupt: raise KeyboardInterrupt
except:
- print '*** Unexpected error ***', t
+ print('*** Unexpected error ***', t)
if verbose:
traceback.print_exc(file=sys.stdout)
else:
try:
result = obj.search(s)
except re.error as msg:
- print '=== Unexpected exception', t, repr(msg)
+ print('=== Unexpected exception', t, repr(msg))
if outcome == SYNTAX_ERROR:
# This should have been a syntax error; forget it.
pass
elif outcome == FAIL:
if result is None: pass # No match, as expected
- else: print '=== Succeeded incorrectly', t
+ else: print('=== Succeeded incorrectly', t)
elif outcome == SUCCEED:
if result is not None:
# Matched, as expected, so now we compute the
@@ -668,17 +668,17 @@ def run_re_tests():
vardict[i] = gi
repl = eval(repl, vardict)
if repl != expected:
- print '=== grouping error', t,
- print repr(repl) + ' should be ' + repr(expected)
+ print('=== grouping error', t, end=' ')
+ print(repr(repl) + ' should be ' + repr(expected))
else:
- print '=== Failed incorrectly', t
+ print('=== Failed incorrectly', t)
# Try the match on a unicode string, and check that it
# still succeeds.
try:
result = obj.search(unicode(s, "latin-1"))
if result is None:
- print '=== Fails on unicode match', t
+ print('=== Fails on unicode match', t)
except NameError:
continue # 1.5.2
except TypeError:
@@ -689,7 +689,7 @@ def run_re_tests():
obj=re.compile(unicode(pattern, "latin-1"))
result = obj.search(s)
if result is None:
- print '=== Fails on unicode pattern match', t
+ print('=== Fails on unicode pattern match', t)
# Try the match with the search area limited to the extent
# of the match and see if it still succeeds. \B will
@@ -701,28 +701,28 @@ def run_re_tests():
obj = re.compile(pattern)
result = obj.search(s, result.start(0), result.end(0) + 1)
if result is None:
- print '=== Failed on range-limited match', t
+ print('=== Failed on range-limited match', t)
# Try the match with IGNORECASE enabled, and check that it
# still succeeds.
obj = re.compile(pattern, re.IGNORECASE)
result = obj.search(s)
if result is None:
- print '=== Fails on case-insensitive match', t
+ print('=== Fails on case-insensitive match', t)
# Try the match with LOCALE enabled, and check that it
# still succeeds.
obj = re.compile(pattern, re.LOCALE)
result = obj.search(s)
if result is None:
- print '=== Fails on locale-sensitive match', t
+ print('=== Fails on locale-sensitive match', t)
# Try the match with UNICODE locale enabled, and check
# that it still succeeds.
obj = re.compile(pattern, re.UNICODE)
result = obj.search(s)
if result is None:
- print '=== Fails on unicode-sensitive match', t
+ print('=== Fails on unicode-sensitive match', t)
def test_main():
run_unittest(ReTests)