|
Boost-Commit : |
From: jurko.gospodnetic_at_[hidden]
Date: 2008-01-07 13:15:13
Author: jurko
Date: 2008-01-07 13:15:13 EST (Mon, 07 Jan 2008)
New Revision: 42567
URL: http://svn.boost.org/trac/boost/changeset/42567
Log:
Code cleanup. Converted to using True/False instead of 1/0. Changed to obey documented coding conventions regarding whitespace and function parameters. Renamed private member functions to use the __ name prefix.
Text files modified:
trunk/tools/build/v2/test/BoostBuild.py | 100 ++++++++++++++++++++-------------------
1 files changed, 51 insertions(+), 49 deletions(-)
Modified: trunk/tools/build/v2/test/BoostBuild.py
==============================================================================
--- trunk/tools/build/v2/test/BoostBuild.py (original)
+++ trunk/tools/build/v2/test/BoostBuild.py 2008-01-07 13:15:13 EST (Mon, 07 Jan 2008)
@@ -66,11 +66,11 @@
# Detect the host OS.
-windows = 0
+windows = False
if os.environ.get('OS','').lower().startswith('windows') or \
os.__dict__.has_key('uname') and \
os.uname()[0].lower().startswith('cygwin'):
- windows = 1
+ windows = True
suffixes = {}
@@ -78,7 +78,8 @@
# Prepare the map of suffixes
def prepare_suffix_map(toolset):
- global windows, suffixes
+ global windows
+ global suffixes
suffixes = {'.exe': '', '.dll': '.so', '.lib': '.a', '.obj': '.o'}
suffixes['.implib'] = '.no_implib_files_on_this_platform'
if windows:
@@ -91,7 +92,7 @@
suffixes['.dll'] = '.dylib'
-def re_remove(sequence,regex):
+def re_remove(sequence, regex):
me = re.compile(regex)
result = filter( lambda x: me.match(x), sequence )
if 0 == len(result):
@@ -100,7 +101,7 @@
sequence.remove(r)
-def glob_remove(sequence,pattern):
+def glob_remove(sequence, pattern):
result = fnmatch.filter(sequence,pattern)
if 0 == len(result):
raise ValueError()
@@ -110,10 +111,10 @@
# Configuration stating whether Boost Build is expected to automatically prepend
# prefixes to built library targets.
-lib_prefix = 1
-dll_prefix = 1
+lib_prefix = True
+dll_prefix = True
if windows:
- dll_prefix = 0
+ dll_prefix = False
#
@@ -121,7 +122,7 @@
# Should be moved to TestCmd.py?
#
if os.name == 'posix':
- def _failed(self, status = 0):
+ def _failed(self, status=0):
if self.status is None:
return None
if os.WIFSIGNALED(status):
@@ -133,12 +134,11 @@
else:
return -1
elif os.name == 'nt':
- def _failed(self, status = 0):
+ def _failed(self, status=0):
return not self.status is None and self.status != status
def _status(self):
return self.status
-
class Tester(TestCmd.TestCmd):
"""Class for testing Boost.Build.
@@ -242,8 +242,8 @@
self
, program=program_list
, match=match
- , workdir = workdir
- , inpath = inpath_bjam
+ , workdir=workdir
+ , inpath=inpath_bjam
, **keywords)
os.chdir(self.workdir)
@@ -265,7 +265,7 @@
# Seems like it's not possible to remove the current a directory.
d = os.getcwd()
os.chdir(os.path.dirname(self.workdir))
- shutil.rmtree(self.workdir, ignore_errors=0)
+ shutil.rmtree(self.workdir, ignore_errors=False)
if not os.path.isabs(tree_location):
tree_location = os.path.join(self.original_workdir, tree_location)
@@ -337,7 +337,7 @@
n = self.glob_file(string.replace(name, "$toolset", self.toolset+"*"))
if n:
if os.path.isdir(n):
- shutil.rmtree(n, ignore_errors=0)
+ shutil.rmtree(n, ignore_errors=False)
else:
os.unlink(n)
@@ -347,7 +347,8 @@
os.chdir(self.workdir)
def expand_toolset(self, name):
- """Expands $toolset in the given file to tested toolset"""
+ """Expands $toolset in the given file to tested toolset.
+ """
content = self.read(name)
content = string.replace(content, "$toolset", self.toolset)
self.write(name, content)
@@ -359,7 +360,7 @@
#
# FIXME: Large portion copied from TestSCons.py, should be moved?
#
- def run_build_system(self, extra_args='', subdir='', stdout=None, stderr='',
+ def run_build_system(self, extra_args="", subdir="", stdout=None, stderr="",
status=0, match=None, pass_toolset=None, use_test_config=None,
ignore_toolset_requirements=None, **kw):
@@ -420,7 +421,7 @@
if stderr:
annotation("STDERR", stderr)
self.maybe_do_diff(self.stdout(), stdout)
- self.fail_test(1, dump_stdio = 0)
+ self.fail_test(1, dump_stdio=False)
# Intel tends to produce some message to stderr which makes tests fail.
intel_workaround = re.compile("^xi(link|lib): executing.*\n", re.M)
@@ -432,7 +433,7 @@
annotation("Actual STDERR", self.stderr())
annotation("STDOUT", self.stdout())
self.maybe_do_diff(actual_stderr, stderr)
- self.fail_test(1, dump_stdio = 0)
+ self.fail_test(1, dump_stdio=False)
self.tree = tree.build_tree(self.workdir)
self.difference = tree.trees_difference(self.previous_tree, self.tree)
@@ -454,7 +455,7 @@
result = result[0]
return result
- def read(self, name, binary = 0):
+ def read(self, name, binary=False):
try:
if self.toolset:
name = string.replace(name, "$toolset", self.toolset+"*")
@@ -478,7 +479,7 @@
else:
return result
- def fail_test(self, condition, dump_stdio = 1, *args):
+ def fail_test(self, condition, dump_stdio=True, *args):
if not condition:
return
@@ -496,7 +497,7 @@
print
path = os.path.join(self.original_workdir, "failed_test")
if os.path.isdir(path):
- shutil.rmtree(path, ignore_errors=0)
+ shutil.rmtree(path, ignore_errors=False)
elif os.path.exists(path):
raise "Path " + path + " already exists and is not a directory";
shutil.copytree(self.workdir, path)
@@ -615,14 +616,14 @@
self.unexpected_difference.pprint()
self.fail_test(1)
- def _expect_line(self, content, expected, expected_to_exist):
+ def __expect_line(self, content, expected, expected_to_exist):
expected = expected.strip()
lines = content.splitlines()
- found = 0
+ found = False
for line in lines:
line = line.strip()
if fnmatch.fnmatch(line, expected):
- found = 1
+ found = True
break
if expected_to_exist and not found:
@@ -637,13 +638,13 @@
self.fail_test(1)
def expect_output_line(self, line, expected_to_exist=True):
- self._expect_line(self.stdout(), line, expected_to_exist)
+ self.__expect_line(self.stdout(), line, expected_to_exist)
def expect_content_line(self, name, line, expected_to_exist=True):
- content = self._read_file(name)
- self._expect_line(content, line, expected_to_exist)
+ content = self.__read_file(name)
+ self.__expect_line(content, line, expected_to_exist)
- def _read_file(self, name, exact=0):
+ def __read_file(self, name, exact=False):
name = self.adjust_names(name)[0]
result = ""
try:
@@ -656,11 +657,11 @@
self.fail_test(1)
return result
- def expect_content(self, name, content, exact=0):
- actual = self._read_file(name, exact)
+ def expect_content(self, name, content, exact=False):
+ actual = self.__read_file(name, exact)
content = string.replace(content, "$toolset", self.toolset+"*")
- matched = 0
+ matched = False
if exact:
matched = fnmatch.fnmatch(actual,content)
else:
@@ -700,25 +701,26 @@
# Helpers.
def mul(self, *arguments):
if len(arguments) == 0:
- return None
- else:
- here = arguments[0]
- if type(here) == type(''):
- here = [here]
-
- if len(arguments) > 1:
- there = apply(self.mul, arguments[1:])
- result = []
- for i in here:
- for j in there:
- result.append(i + j)
- return result
- else:
- return here
+ return None
+
+ here = arguments[0]
+ if type(here) == type(''):
+ here = [here]
+
+ if len(arguments) > 1:
+ there = apply(self.mul, arguments[1:])
+ result = []
+ for i in here:
+ for j in there:
+ result.append(i + j)
+ return result
+
+ return here
# Internal methods.
def ignore_elements(self, list, wildcard):
- """Removes in-place, element of 'list' that match the given wildcard."""
+ """Removes in-place, element of 'list' that match the given wildcard.
+ """
list[:] = filter(lambda x, w=wildcard: not fnmatch.fnmatch(x, w), list)
def adjust_lib_name(self, name):
@@ -763,7 +765,7 @@
# Adjusts suffixes on all names.
def adjust_names(self, names):
if type(names) == types.StringType:
- names = [names]
+ names = [names]
r = map(self.adjust_lib_name, names)
r = map(self.adjust_suffix, r)
r = map(lambda x, t=self.toolset: string.replace(x, "$toolset", t+"*"), r)
Boost-Commit list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk