1
0
mirror of https://github.com/samba-team/samba.git synced 2025-01-12 09:18:10 +03:00

Import ComfyChair framework from trunk.

(This used to be commit a7fb1fcc72)
This commit is contained in:
Martin Pool 2003-03-12 03:06:42 +00:00
parent ffd5eae60e
commit 57b68013e1

266
source3/stf/comfychair.py Normal file
View File

@ -0,0 +1,266 @@
#! /usr/bin/env python
# Copyright (C) 2002, 2003 by Martin Pool <mbp@samba.org>
#
# This program is free software; you can redistribute it and/or
# modify it under the terms of the GNU General Public License as
# published by the Free Software Foundation; either version 2 of the
# License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
# USA
"""comfychair: a Python-based instrument of software torture.
Copyright (C) 2002, 2003 by Martin Pool <mbp@samba.org>
This is a test framework designed for testing programs written in
Python, or (through a fork/exec interface) any other language. It is
similar in design to the very nice 'svntest' system used by
Subversion, but has no Subversion-specific features.
It is somewhat similar to PyUnit, except:
- it allows capture of detailed log messages from a test, to be
optionally displayed if the test fails.
- it allows execution of a specified subset of tests
- it avoids Java idioms that are not so useful in Python
WRITING TESTS:
Each test case is a callable object, typically a function. Its
documentation string describes the test, and the first line of the
docstring should be a brief name.
The test should return 0 for pass, or non-zero for failure.
Alternatively they may raise an exception.
Tests may import this "comfychair" module to get some useful
utilities, but that is not strictly required.
"""
# TODO: Put everything into a temporary directory?
# TODO: Have a means for tests to customize the display of their
# failure messages. In particular, if a shell command failed, then
# give its stderr.
import sys, re
class TestCase:
"""A base class for tests. This class defines required functions which
can optionally be overridden by subclasses. It also provides some
utility functions for"""
def __init__(self):
self.test_log = ""
self.background_pids = []
def setUp(self):
"""Set up test fixture."""
pass
def tearDown(self):
"""Tear down test fixture."""
pass
def runTest(self):
"""Run the test."""
pass
def fail(self, reason = ""):
"""Say the test failed."""
raise AssertionError(reason)
def assert_(self, expr, reason = ""):
if not expr:
raise AssertionError(reason)
def assert_re_match(self, pattern, s):
"""Assert that a string matches a particular pattern
Inputs:
pattern string: regular expression
s string: to be matched
Raises:
AssertionError if not matched
"""
if not re.match(pattern, s):
raise AssertionError("string %s does not match regexp %s" % (`s`, `pattern`))
def assert_regexp(self, pattern, s):
"""Assert that a string *contains* a particular pattern
Inputs:
pattern string: regular expression
s string: to be searched
Raises:
AssertionError if not matched
"""
if not re.search(pattern, s):
raise AssertionError("string %s does not contain regexp %s" % (`s`, `pattern`))
def assert_no_file(self, filename):
import os.path
assert not os.path.exists(filename), ("file exists but should not: %s" % filename)
def runCmdNoWait(self, cmd):
import os
name = cmd[0]
self.test_log = self.test_log + "Run in background:\n" + `cmd` + "\n"
pid = os.spawnvp(os.P_NOWAIT, name, cmd)
self.test_log = self.test_log + "pid: %d\n" % pid
return pid
def runCmd(self, cmd, expectedResult = 0):
"""Run a command, fail if the command returns an unexpected exit
code. Return the output produced."""
rc, output = self.runCmdUnchecked(cmd)
if rc != expectedResult:
raise AssertionError("command returned %d; expected %s: \"%s\"" %
(rc, expectedResult, cmd))
return output
def runCmdUnchecked(self, cmd, skip_on_noexec = 0):
"""Invoke a command; return (exitcode, stdout)"""
import os, popen2
pobj = popen2.Popen4(cmd)
output = pobj.fromchild.read()
waitstatus = pobj.wait()
assert not os.WIFSIGNALED(waitstatus), \
("%s terminated with signal %d", cmd, os.WTERMSIG(waitstatus))
rc = os.WEXITSTATUS(waitstatus)
self.test_log = self.test_log + ("""Run command: %s
Wait status: %#x
Output:
%s""" % (cmd, waitstatus, output))
if skip_on_noexec and rc == 127:
# Either we could not execute the command or the command
# returned exit code 127. According to system(3) we can't
# tell the difference.
raise NotRunError, "could not execute %s" % cmd
return rc, output
def explainFailure(self, exc_info = None):
import traceback
# Move along, nothing to see here
if not exc_info and self.test_log == "":
return
print "-----------------------------------------------------------------"
if exc_info:
traceback.print_exc(file=sys.stdout)
print self.test_log
print "-----------------------------------------------------------------"
def require(self, predicate, message):
"""Check a predicate for running this test.
If the predicate value is not true, the test is skipped with a message explaining
why."""
if not predicate:
raise NotRunError, message
def require_root(self):
"""Skip this test unless run by root."""
import os
self.require(os.getuid() == 0,
"must be root to run this test")
def log(self, msg):
"""Log a message to the test log. This message is displayed if
the test fails, or when the runtests function is invoked with
the verbose option."""
self.test_log = self.test_log + msg + "\n"
class NotRunError(Exception):
def __init__(self, value = None):
self.value = value
def test_name(test):
"""Return a human-readable name for a test.
Inputs:
test some kind of callable test object
Returns:
name string: a short printable name
"""
try:
return test.__name__
except:
return `test`
def runtests(test_list, verbose = 0):
"""Run a series of tests.
Eventually, this routine will also examine sys.argv[] to handle
extra options.
Inputs:
test_list sequence of callable test objects
Returns:
unix return code: 0 for success, 1 for failures, 2 for test failure
"""
import traceback
ret = 0
for test in test_list:
print "%-60s" % test_name(test),
# flush now so that long running tests are easier to follow
sys.stdout.flush()
try:
try: # run test and show result
obj = test()
if hasattr(obj, "setUp"):
obj.setUp()
obj.runTest()
print "OK"
except KeyboardInterrupt:
print "INTERRUPT"
obj.explainFailure(sys.exc_info())
ret = 2
break
except NotRunError, msg:
print "NOTRUN, %s" % msg.value
except:
print "FAIL"
obj.explainFailure(sys.exc_info())
ret = 1
finally:
try:
if hasattr(obj, "tearDown"):
obj.tearDown()
except KeyboardInterrupt:
print "interrupted during tearDown"
obj.explainFailure(sys.exc_info())
ret = 2
break
except:
print "error during tearDown"
obj.explainFailure(sys.exc_info())
ret = 1
# Display log file if we're verbose
if ret == 0 and verbose:
obj.explainFailure()
return ret
if __name__ == '__main__':
print __doc__