def __setup_value(self):
if "path" in dir(self.__filename):
self.subobjects["filename"] = self.__filename
self.__filename = self.__filename.path
self.__filename = os.path.abspath(self.__filename)
fn = self.__filename
extract_mode = ""
if "tar.gz" in fn or "tgz" in fn:
extract_mode = "x"
if "tar.bz2" in fn or "bzip2" in fn:
extract_mode = "j"
with self.tmp_directory as d:
try:
os.mkdir(self.name)
except OSError:
# ignore errors if the directory should already exist for some reason
pass
with Directory(self.name) as d2:
dirname = os.path.abspath(".")
(out, ret) = shell("tar %szvf %s", extract_mode, fn)
if ret != 0:
raise RuntimeError("Extracting of %s failed" % fn)
cd = None
for line in out:
if (cd == None
or len(line) < len(cd)) and line.endswith("/"):
cd = line
if cd and all([x.startswith(cd) for x in out]):
dirname = cd
return Directory(os.path.abspath(dirname))
class SimpleExperiment(Experiment):
outputs = {"dir1": Directory("d1"),
"dir2": Directory("d2"),
"filtered": Directory(".", filename_filter="*.log*"),
}
def run(self):
a = self.o.dir1.new_file("barfoo")
a.value="abc"
a.flush()
os.mkdir(self.o.dir1.path + "/tmpdir")
with open(self.o.dir1.path + "/tmpdir/foo", "w+") as fd:
fd.write("Hallo")
self.o.dir2.mirror_directory(self.o.dir1.path,
lambda x: True)
a = self.filtered.new_file("foo.log")
a.value = "xx"
try:
a = self.filtered.new_file("bar.xxx")
raise Exception("Filter does not work")
except RuntimeError as e:
pass # Everything is good
b = self.filtered.new_file("barfoo.log.gz", compressed=True)
b.value = "xx"
assert type(a) == File
assert type(b) == GzipFile
class SimpleExperiment(Experiment):
outputs = {"dir1": Directory("d1"),
"dir2": Directory("d2")}
def run(self):
a = self.o.dir1.new_file("barfoo")
a.value="abc"
a.flush()
os.mkdir(self.o.dir1.path + "/tmpdir")
with open(self.o.dir1.path + "/tmpdir/foo", "w+") as fd:
fd.write("Hallo")
self.o.dir2.mirror_directory(self.o.dir1.path,
lambda x: True)
def __setup_value(self):
if "path" in dir(self.__clone_url):
self.subobjects["clone-url"] = self.__clone_url
self.__clone_url = self.__clone_url.path
logging.info("copying git archive %s", self.__clone_url)
with self.tmp_directory as d:
os.mkdir(self.name)
if self.__shallow:
cmd = "cd '%s' && git archive --format=tar --remote=%s %s | tar x"
args = (self.name, self.__clone_url, self.__ref)
else:
cmd = "git clone %s %s"
args = (self.__clone_url, self.name)
(lines, ret) = shell(cmd, *args)
if ret != 0:
print("\n".join(lines))
sys.exit(-1)
if not self.__shallow:
cmd = "cd %s && git gc && git fetch %s %s && git checkout FETCH_HEAD"
args = (self.name, self.__clone_url, self.__ref)
(lines, ret) = shell(cmd, *args)
if ret != 0:
print("\n".join(lines))
sys.exit(-1)
return Directory(os.path.abspath(self.name))
class FailImport(Experiment):
inputs = {
"trace": FailTrace("FailTrace"),
"fail-tool-dir": Directory("/proj/i4danceos/tools/fail"),
}
def run(self):
variant = "erika/error-hook"
for (label, importer, importer_args) in [\
("mem", "MemoryImporter", []),
("regs", "RegisterImporter", []),
("ip", "RegisterImporter", ["--no-gp", "--ip"]),
("flags", "RegisterImporter", ["--no-gp", "--flags"]),
]:
benchmark = label
logging.info("Importing coredos/%s", benchmark)
cmdline = "%(path)s/import-trace -v %(variant)s -b %(benchmark)s -i %(importer)s "\
+ "-t %(trace)s -e %(elf)s %(args)s"
shell(cmdline %\
{"path": self.fail_tool_dir.path,
"variant": variant,
"benchmark": benchmark,
"importer": importer,
"trace": os.path.join(self.trace.trace.path, "trace.pb"),
"elf": self.trace.elf.path,
"args": " ".join(importer_args)})
shell("%s/prune-trace -v %s -b %% -p basic --overwrite",
self.fail_tool_dir.path, variant)
class FailTrace(Experiment):
inputs = {
"erika": GitArchive("[email protected]:erika"),
"bochs-runner": Executable("/proj/i4danceos/tools/fail/bochs-experiment-runner.py"),
"erika-tracing": Executable("/proj/i4danceos/tools/fail/erika-tracing"),
}
outputs = {
"trace": Directory("trace"),
"elf": File("erika.elf"),
"iso": File("erika.iso"),
}
def run(self):
logging.info("Cloning ERIKA...")
with self.erika as erika_path:
shell("cd %s/examples/x86/coptermock-isorc; make", erika_path)
self.iso.copy_contents(os.path.join(erika_path, "examples/x86/coptermock-isorc/Debug/erika.iso"))
self.elf.copy_contents(os.path.join(erika_path, "examples/x86/coptermock-isorc/Debug/Debug/out.elf"))
shell(("cd %(resultdir)s; python %(bochs)s -F 50 -i %(iso)s -e %(elf)s -f %(fail)s"
+ " -m 8 -1 -- -Wf,--end-symbol=test_finish -Wf,--start-symbol=EE_oo_StartOS"
+ " -Wf,--trace-file=trace.pb -Wf,--save-symbol=EE_oo_StartOS") % {
"resultdir": self.trace.path,
"bochs": self.bochs_runner.path,
"iso": self.iso.path,
"elf": self.elf.path,
"fail": self.erika_tracing.path
}
)
class ExploreConfig(AttributeExperiment):
inputs = {
"config_hash": String("FIXME"),
"kconfig_hash": String("FIXME"),
"project_root": Directory("/tmp"),
"project_version": String("FIXME"),
"clean_command": String("make clean"),
"build_command": String("make"),
"attr_command": String("make attributes"),
}
class SimpleExperiment(Experiment):
inputs = {
"input_key": String("default key"),
"input_value": String("default value")
}
outputs = {
"output_file": File("output"),
"output_directory": Directory("output_directory")
}
def run(self):
# Combine the input parameters
content = self.inputs.input_key.value \
+ ": " + self.inputs.input_value.value
# write the result to the output file
self.outputs.output_file.value = content + "\n"
# New output directory
x = self.output_directory.new_directory("foo").new_file("lala")
def __setup_value(self):
if "path" in dir(self.__clone_url):
self.subobjects["clone-url"] = self.__clone_url
self.__clone_url = self.__clone_url.path
logging.info("copying git archive %s", self.__clone_url)
with self.tmp_directory as d:
os.mkdir(self.name)
if self.__shallow:
cmd = "cd '%s' && git archive --format=tar --remote=%s %s | tar x"
args = (self.name,
self.__clone_url,
self.__ref)
else:
cmd = "git clone %s %s"
args = (self.__clone_url, self.name)
(lines, ret) = shell(cmd, *args, stderr=sys.stderr)
if ret != 0:
print("\n".join(lines))
sys.exit(-1)
if not self.__shallow:
cmd = "cd %s && git gc && git fetch %s %s && git checkout FETCH_HEAD"
args = (self.name, self.__clone_url, self.__ref)
(lines, ret) = shell(cmd, *args, stderr=sys.stderr)
if ret != 0:
print("\n".join(lines))
sys.exit(-1)
# Fetch all visible branches and tags
for branch in self.__metadata.get("branches", {}):
cmd = "cd %s && git fetch %s refs/heads/%s && git update-ref refs/heads/%s FETCH_HEAD"
shell(cmd, self.name, self.__clone_url, branch, branch, stderr=sys.stderr)
for tag in self.__metadata.get("tags", {}):
cmd = "cd %s && git fetch %s refs/tags/%s && git update-ref refs/tags/%s FETCH_HEAD"
shell(cmd, self.name, self.__clone_url, tag, tag, stderr=sys.stderr)
return Directory(os.path.abspath(self.name))
def before_experiment_run(self, parameter_type):
# When experiment run as input, just run the normal input handlers
if parameter_type == "input":
Type.before_experiment_run(self, "input")
return
for (name, inp) in self.inputs.items():
if type(inp) == LambdaType:
continue
ret = inp.inp_extract_cmdline_parser(self.__opts, self.__args)
if ret:
(self.__opts, self.__args) = ret
# After all input parameters are parsed. Execute the
# calculated input parameters
for (name, inp) in self.inputs.items():
if type(inp) != LambdaType:
continue
inp = inp(self)
inp.name = name
self.subobjects[name] = inp
self.inputs[name] = inp
self.subobjects.update()
# Now set up the experiment tmp directory
self.tmp_directory = Directory(tempfile.mkdtemp())
self.subobjects["tmp_directory"] = self.tmp_directory
for obj in self.inputs.values():
obj.before_experiment_run("input")
self.__calculate_metadata()
for obj in self.outputs.values():
obj.before_experiment_run("output")
pass # Everything is good
b = self.filtered.new_file("barfoo.log.gz", compressed=True)
b.value = "xx"
assert type(a) == File
assert type(b) == GzipFile
if __name__ == "__main__":
import shutil, sys,os
experiment = SimpleExperiment()
dirname = experiment(sys.argv)
assert os.path.isdir(experiment.o.dir2.path + "/tmpdir")
assert os.path.exists(experiment.o.dir2.path + "/barfoo")
assert os.path.exists(experiment.o.dir2.path + "/tmpdir/foo")
N = Directory(experiment.path, "*.log*")
assert experiment.filtered.value == N.value
assert os.path.exists(experiment.path + "/foo.log")
contents = [x.value for x in N]
assert len(contents) == 2
assert contents[0] == contents[1], contents
if dirname:
shutil.rmtree(dirname)
print("success")
class HistoricalCompilationGlobalEvaluation(Experiment):
inputs = {
"clang_hash": GitArchive("/home/cip/2015/yb90ifym/clang-hash/"),
"project": GitArchive("/home/cip/2015/yb90ifym/clang-hash/hash-projects/lua"),
"commits": Integer(4744),
"jobs": Integer(1), # was 4
"dataset": Directory("/home/cip/2015/yb90ifym/clang-hash/experiments/HistoricalCompilation-4e7c977077afea3d2ad77aeefe3b472c"), # full lua
"hot_threshold_percentage": Integer(10), # minimal change percentage for commit to be classified as "hot"
}
outputs = {
"stats": File("summary.dict"),
"eval_data": File("eval.txt"),
"hot_commits_histo": File("global_hot_commits.pdf"),
}
def project_name(self):
return os.path.basename(self.metadata['project-clone-url'])
def run(self):
# Project name
logging.info("Cloning project... %s", self.project_name())
self.build_info = {"project-name": self.project_name(),
"commit-hash": self.metadata["project-hash"],
'builds': []}
with self.project as src_path:
time = 0
os.chdir(self.dataset.path)
# Read summary file from data collection run
commits = None
with open("summary.dict") as sf:
summary = eval(sf.read())
commits = summary['builds']
def read_chash_data(commit):
element_hashes = []
try:
with open(commit, 'r') as cf:
commit_data = eval(cf.read())
for ofile_data in commit_data:
element_hashes.extend(ofile_data['element-hashes'])
except:
pass
return element_hashes
stats = {
'data-empty': set(), # commits with empty info files, e.g. failed to be collected, (first n commits -> missing makefile o.a.)
'commits': {},
'elements': {}, # symbol -> how often did this symbol change
}
total_changed_globals = 0 # How often was any global changed throughout the history?
total_changed_records = 0 # How often was any record changed throughout the history?
total_changed_static_funcs = 0 # How often was any static function changed throughout the history?
total_changed_functions = 0 # without static functions
total_insdel_globals = 0 # How often was any global introduced/removed throughout the history?
total_insdel_records = 0 # How often was any record introduced/removed throughout the history?
total_insdel_static_funcs = 0 # How often was any static function introduced/removed throughout the history?
total_insdel_functions = 0 # without static functions
# in-degree: how many SLOs depend on E?
# out-degree: how many SLOs does E depend on?
in_degrees = {} # indegree -> nr of elements with that indegree
out_degrees = {} # outdegree -> nr of elements wirh that outdegree
max_in_degree = (None, 0) # (element, degree)
max_out_degree = (None, 0) # (element, degree)
prev_commit = None
prev_hashes = None
prev_used_definitions = None
prev_global_hashes = None
counter = 1
for info in commits:
print "\n%d/%d" % (counter, len(commits))
counter += 1
commit = info['commit']
parent = info['parent']
if not parent: # first commit has no parent
print "No parent"
continue
commit_data = read_chash_data(commit)
if not commit_data:
# If the data does not exist, note and skip
#print "Data empty"
stats['data-empty'].add(commit)
continue
local_hashes = {}
used_definitions = {}
# just 4 testing:
for element in commit_data:
name = element[0]
if name.startswith('static function:') or name.startswith('function:'):
name = element[0].split(':')[1]
local_hashes[name] = element[1]
try:
used_definitions[name] = set()
for used_def in element[2]:
if used_def.startswith('static function:') or used_def.startswith('function:'):
used_definitions[name].add(used_def.split(':')[1])
except:
pass
# prev:
#for element in commit_data:
# local_hashes[element[0]] = element[1]
# try:
# used_definitions[element[0]] = element[2]
# except:
# pass
parent_hashes = {}
parent_global_hashes = {}
parent_used_definitions = {}
if parent == prev_commit and prev_global_hashes and prev_used_definitions and prev_hashes:
#print "Reuse prev_commit"
parent_hashes = prev_hashes
parent_used_definitions = prev_used_definitions
parent_global_hashes = prev_global_hashes
else:
#print "Cannot reuse prev_commit"
parent_data = read_chash_data(parent)
# just 4 testing:
for element in parent_data:
name = element[0]
if name.startswith('static function:') or name.startswith('function:'):
name = element[0].split(':')[1]
parent_hashes[name] = element[1]
try:
parent_used_definitions[name] = set()
for used_def in element[2]:
if used_def.startswith('static function:') or used_def.startswith('function:'):
parent_used_definitions[name].add(used_def.split(':')[1])
except:
pass
# prev:
#for element in parent_data:
# parent_hashes[element[0]] = element[1]
# try:
# parent_used_definitions[element[0]] = element[2]
# except:
# pass
if not parent_hashes:
# If the data does not exist, note and skip
stats['data-empty'].add(commit)
# Save data for reuse
prev_commit = commit
prev_hashes = local_hashes
prev_used_definitions = used_definitions
continue
##########################
# GLOBAL HASH EVALUATION #
##########################
commit_stats = {
'element-count' : len(local_hashes),
'changed-elements' : [],
}
elements = set(local_hashes.keys())
parent_elements = set(parent_hashes.keys())
# calculate in- and out-degree
# reverse used_definitions
out_use_defs = { s:0 for s in used_definitions.keys() } # element -> nr of depending elements
for element in elements:
for el in used_definitions[element]:
try:
out_use_defs[el] += 1
except:
pass
for element in elements:
out_degree = len(used_definitions[element])
in_degree = out_use_defs[element]
if in_degree > max_in_degree[1]:
max_in_degree = (element, in_degree)
if out_degree > max_out_degree[1]:
max_out_degree = (element, out_degree)
if in_degree not in in_degrees:
in_degrees[in_degree] = 0
in_degrees[in_degree] += 1
if out_degree not in out_degrees:
out_degrees[out_degree] = 0
out_degrees[out_degree] += 1
commit_stats['changed-elements'] = elements ^ parent_elements # elements either added or removed
for element in commit_stats['changed-elements']:
if element.startswith('record:'): # do this here to get only insertions and deletions
total_insdel_records += 1
elif element.startswith('variable:') or element.startswith('static variable:'):
total_insdel_globals += 1
elif element.startswith('static function:'):
total_insdel_static_funcs += 1
else:
total_insdel_functions += 1
# Compare hashes
common_elements = elements & parent_elements
global_hashes = {}
for element in common_elements:
global_hash = get_global_hash(element, global_hashes, local_hashes, used_definitions)
parent_global_hash = get_global_hash(element, parent_global_hashes, parent_hashes, parent_used_definitions)
if global_hash != parent_global_hash:
commit_stats['changed-elements'].add(element)
if element.startswith('record:'): # do this here to ignore insertions and deletions
total_changed_records += 1
elif element.startswith('variable:') or element.startswith('static variable:'):
total_changed_globals += 1
elif element.startswith('static function:'):
total_changed_static_funcs += 1
else:
total_changed_functions += 1
commit_stats['changed-element-count'] = len(commit_stats['changed-elements']);
stats['commits'][commit] = commit_stats
# Count how often each element was changed over the whole history
for element in commit_stats['changed-elements']:
if element not in stats['elements']:
stats['elements'][element] = 0;
stats['elements'][element] += 1
# Save data for reuse
prev_commit = commit
prev_hashes = local_hashes
prev_used_definitions = used_definitions
prev_global_hashes = global_hashes
self.build_info['stats'] = stats
#in_degrees = {} # indegree -> nr of elements with that indegree
#out_degrees = {} # outdegree -> nr of elements wirh that outdegree
#max_in_degree = (None, 0) # (element, degree)
#max_out_degree = (None, 0) # (element, degree)
summed_in_degrees = sum([k*v for k,v in in_degrees.iteritems()])
nr_of_elements = sum(in_degrees.values())
avg_in_degree = summed_in_degrees/float(nr_of_elements)
avg_out_degree = sum([k*v for k,v in out_degrees.iteritems()])/float(sum(out_degrees.values()))
eval_info = {
'nr-of-commits' : len(commits),
'change-percentage' : {}, # change percentage -> nr of commits with change < percentage
'hot-commits': {},
'total-changed-globals': total_changed_globals,
'total-changed-records': total_changed_records,
'total-changed-static-funcs': total_changed_static_funcs,
'total-changed-functions': total_changed_functions,
'total-insdel-globals': total_insdel_globals,
'total-insdel-records': total_insdel_records,
'total-insdel-static-funcs': total_insdel_static_funcs,
'total-insdel-functions': total_insdel_functions,
'max_in_degree': max_in_degree,
'max_out_degree': max_out_degree,
'avg_in_degree': avg_in_degree,
'avg_out_degree': avg_out_degree,
}
# Get most changed elements
eval_info['most-changed-elements'] = {k:v for k,v in stats['elements'].iteritems() if v > 1000} # arbitrary value (about 20% of commits)
# Calc average nr and percentage of (changed) symbols per commit
summed_avg_change_percentage = 0
summed_changed_elements = 0
summed_total_elements = 0
commits = self.build_info['stats']['commits']
for commit in commits:
commit_stat = commits[commit]
change_percentage = len(commit_stat['changed-elements'])/float(commit_stat['element-count'])
summed_avg_change_percentage += change_percentage
summed_changed_elements += len(commit_stat['changed-elements'])
summed_total_elements += commit_stat['element-count']
percentage = int(round(change_percentage * 100))
if percentage not in eval_info['change-percentage']:
eval_info['change-percentage'][percentage] = 0
eval_info['change-percentage'][percentage] += 1
# Identify hot commits
#if percentage > self.hot_threshold_percentage.value:
#eval_info['hot-commits'][commit] = percentage
eval_info['avg-change-percentage'] = summed_avg_change_percentage / float(len(stats['commits']))
eval_info['avg-changed-elements'] = summed_changed_elements / eval_info['nr-of-commits']
eval_info['avg-total-elements'] = summed_total_elements / eval_info['nr-of-commits']
eval_info['nr-hot-commits'] = len(eval_info['hot-commits'])
with open(self.eval_data.path, "w+") as fd:
fd.write(repr(eval_info))
# Output the summary of this build into the statistics file.
with open(self.stats.path, "w+") as fd:
fd.write(repr(self.build_info))
def plot_hash_count_histogram(hash_values, filename):
dictionary = plt.figure()
fig, ax = plt.subplots()
plt.xlabel('Prozentanteil geaenderter Elemente')
plt.ylabel('Anzahl von Commits')
axes = plt.gca()
axes.set_xlim([-10,100])
axes.set_ylim([0,1600])
ax.bar(hash_values.keys(), hash_values.values(), align='center')
fig.savefig(filename)
# clean data for plotting
data = {k:v for k,v in eval_info['change-percentage'].iteritems() if k <= 100}
plot_hash_count_histogram(data, self.hot_commits_histo.path)
def variant_name(self):
return "%s-%s"%(self.project_name(), self.metadata['mode'])
def symlink_name(self):
return "%s-%s"%(self.title, self.variant_name())
class HistoricalCompilationCallGraphEvaluation(Experiment):
inputs = {
"clang_hash":
GitArchive("/home/cip/2015/yb90ifym/clang-hash/"),
"project":
GitArchive("/home/cip/2015/yb90ifym/clang-hash/hash-projects/lua"),
"commits":
Integer(4744),
"jobs":
Integer(1), # was 4
"dataset":
Directory(
"/home/cip/2015/yb90ifym/clang-hash/experiments/HistoricalCompilation-4e7c977077afea3d2ad77aeefe3b472c"
), # full lua
"hot_threshold_percentage":
Integer(
10
), # minimal change percentage for commit to be classified as "hot"
}
outputs = {
"stats": File("summary.dict"),
"eval_data": File("eval.txt"),
"hot_commits_histo": File("cg_hot_commits.pdf"),
}
def project_name(self):
return os.path.basename(self.metadata['project-clone-url'])
def run(self):
# Project name
logging.info("Cloning project... %s", self.project_name())
self.build_info = {
"project-name": self.project_name(),
"commit-hash": self.metadata["project-hash"],
'builds': []
}
with self.project as src_path:
time = 0
os.chdir(self.dataset.path)
# Read summary file from data collection run
commits = None
with open("summary.dict") as sf:
summary = eval(sf.read())
commits = summary['builds']
def read_chash_data(commit):
element_hashes = []
try:
with open(commit, 'r') as cf:
commit_data = eval(cf.read())
for ofile_data in commit_data:
element_hashes.extend(ofile_data['element-hashes'])
except:
pass
return element_hashes
stats = {
'data-empty': set(
), # commits with empty info files, e.g. failed to be collected, (first n commits -> missing makefile o.a.)
'commits': {},
'elements': {}, # symbol -> how often did this symbol change
}
total_changed_functions = 0 # How often was any function changed throughout the history?
total_insdel_functions = 0 # How often was any function introduced/removed throughout the history?
prev_commit = None
prev_functions = None
prev_used_definitions = None
counter = 1
for info in commits:
print "%d/%d" % (counter, len(commits))
counter += 1
commit = info['commit']
parent = info['parent']
if not parent: # first commit has no parent
print "No parent"
continue
commit_data = read_chash_data(commit)
if not commit_data:
# If the data does not exist, note and skip
#print "Data empty"
stats['data-empty'].add(commit)
continue
functions = set()
used_definitions = {}
for element in commit_data:
if element[0].startswith('static function:') or element[
0].startswith('function:'):
clean_name = element[0].split(':')[1]
functions.add(clean_name)
used_definitions[clean_name] = set()
for used_def in element[2]:
if used_def.startswith(
'static function:') or used_def.startswith(
'function:'):
used_definitions[clean_name].add(
used_def.split(':')[1])
parent_functions = {}
parent_used_definitions = {}
if parent == prev_commit and prev_functions and prev_used_definitions:
#print "Reuse prev_commit"
parent_functions = prev_functions
parent_used_definitions = prev_used_definitions
else:
#print "Cannot reuse prev_commit"
parent_data = read_chash_data(parent)
for element in parent_data:
if element[0].startswith(
'static function:') or element[0].startswith(
'function:'):
clean_name = element[0].split(':')[1]
parent_functions.insert(clean_name)
parent_used_definitions[clean_name] = set()
for used_def in element[2]:
if used_def.startswith(
'static function:'
) or used_def.startswith('function:'):
parent_used_definitions[clean_name].add(
used_def.split(':')[1])
if not parent_functions:
# If the data does not exist, note and skip
stats['data-empty'].add(commit)
# Save data for reuse
prev_commit = commit
prev_functions = functions
prev_used_definitions = used_definitions
continue
#########################
# CALL GRAPH EVALUATION #
#########################
commit_stats = {
'element-count': len(functions),
'changed-elements':
[], # contains changed + impacted functions
#'changed-not-impacted': set(), # contains directly changed functions only
}
elements = functions
parent_elements = parent_functions
commit_stats['changed-elements'] = set(
) #elements ^ parent_elements # elements either added or removed
total_insdel_functions += len(commit_stats['changed-elements'])
cwd = os.getcwd()
os.chdir(src_path)
changed_functions = get_changed_functions_from_commit(
src_path, commit)
os.chdir(cwd)
commit_stats['changed-not-impacted'] = changed_functions.copy()
# Get impacted functions
changed_functions |= get_impacted_funcs_fake_hash(
changed_functions, used_definitions)
commit_stats['changed-elements'] |= changed_functions
total_changed_functions += len(changed_functions)
commit_stats['changed-element-count'] = len(
commit_stats['changed-elements'])
stats['commits'][commit] = commit_stats
# Count how often each element was changed over the whole history
for element in commit_stats['changed-elements']:
if element not in stats['elements']:
stats['elements'][element] = 0
stats['elements'][element] += 1
# Save data for reuse
prev_commit = commit
prev_functions = functions
prev_used_definitions = used_definitions
self.build_info['stats'] = stats
eval_info = {
'nr-of-commits': len(commits),
'change-percentage':
{}, # change percentage -> nr of commits with change < percentage
'hot-commits': {},
'total-changed-functions': total_changed_functions,
'total-insdel-functions': total_insdel_functions,
}
# Get most changed elements
eval_info['most-changed-elements'] = {
k: v
for k, v in stats['elements'].iteritems() if v > 400
} # arbitrary value (about 10% of commits)
# Calc average nr and percentage of (changed) symbols per commit
summed_avg_change_percentage = 0
summed_changed_elements = 0
summed_total_elements = 0
commits = self.build_info['stats']['commits']
for commit in commits:
commit_stat = commits[commit]
change_percentage = len(commit_stat['changed-elements']) / float(
commit_stat['element-count'])
summed_avg_change_percentage += change_percentage
summed_changed_elements += len(commit_stat['changed-elements'])
summed_total_elements += commit_stat['element-count']
percentage = int(round(change_percentage * 100))
if percentage not in eval_info['change-percentage']:
eval_info['change-percentage'][percentage] = 0
eval_info['change-percentage'][percentage] += 1
# Identify hot commits
#if percentage > self.hot_threshold_percentage.value:
#eval_info['hot-commits'][commit] = percentage
eval_info[
'avg-change-percentage'] = summed_avg_change_percentage / float(
len(stats['commits']))
eval_info[
'avg-changed-elements'] = summed_changed_elements / eval_info[
'nr-of-commits']
eval_info['avg-total-elements'] = summed_total_elements / eval_info[
'nr-of-commits']
eval_info['nr-hot-commits'] = len(eval_info['hot-commits'])
with open(self.eval_data.path, "w+") as fd:
fd.write(repr(eval_info))
# Output the summary of this build into the statistics file.
with open(self.stats.path, "w+") as fd:
fd.write(repr(self.build_info))
def plot_hash_count_histogram(hash_values, filename):
dictionary = plt.figure()
fig, ax = plt.subplots()
plt.xlabel('Prozentanteil geaenderter Elemente')
plt.ylabel('Anzahl von Commits')
axes = plt.gca()
axes.set_xlim([-10, 100])
axes.set_ylim([0, 1600])
ax.bar(hash_values.keys(), hash_values.values(), align='center')
fig.savefig(filename)
# clean data for plotting
data = {
k: v
for k, v in eval_info['change-percentage'].iteritems() if k <= 100
}
plot_hash_count_histogram(data, self.hot_commits_histo.path)
def variant_name(self):
return "%s-%s" % (self.project_name(), self.metadata['mode'])
def symlink_name(self):
return "%s-%s" % (self.title, self.variant_name())
def __setup_tmp_directory(self):
"""Creat temporary directory and assign it to every input and
output directories tmp_directory slots"""
# Create temp directory
self.tmp_directory = Directory(tempfile.mkdtemp())
self.subobjects["tmp_directory"] = self.tmp_directory
class HistoricalCompilationEvaluation(Experiment):
inputs = {
"clang_hash":
GitArchive("/home/cip/2015/yb90ifym/clang-hash/"),
"project":
GitArchive("/home/cip/2015/yb90ifym/clang-hash/hash-projects/lua"),
"commits":
Integer(4744),
"jobs":
Integer(1), # was 4
"dataset":
Directory(
"/home/cip/2015/yb90ifym/clang-hash/experiments/HistoricalCompilation-4e7c977077afea3d2ad77aeefe3b472c"
), # full lua
"hot_threshold_percentage":
Integer(
50
), # minimal change percentage for commit to be classified as "hot"
}
outputs = {
"stats": File("summary.dict"),
"eval_data": File("eval.txt"),
"hot_commits_histo": File("local_hot_commits.pdf"),
"compare_approx_elem": File("local_compare_approx_elem.pdf"),
}
def project_name(self):
return os.path.basename(self.metadata['project-clone-url'])
def run(self):
# Project name
logging.info("Cloning project... %s", self.project_name())
self.build_info = {
"project-name": self.project_name(),
"commit-hash": self.metadata["project-hash"],
'builds': []
}
with self.project as src_path:
time = 0
os.chdir(self.dataset.path)
# Read summary file from data collection run
commits = None
with open("summary.dict") as sf:
summary = eval(sf.read())
commits = summary['builds']
def read_chash_data(commit):
element_hashes = []
try:
with open(commit, 'r') as cf:
commit_data = eval(cf.read())
for ofile_data in commit_data:
element_hashes.extend(ofile_data['element-hashes'])
except:
pass
return element_hashes
stats = {
'data-empty': set(
), # commits with empty info files, e.g. failed to be collected, (first n commits -> missing makefile o.a.)
'commits': {},
'elements': {}, # symbol -> how often did this symbol change
}
total_changed_globals = 0 # How often was any global changed/introduced throughout the history?
total_changed_records = 0 # How often was any record changed/introduced throughout the history?
total_changed_static_funcs = 0 # How often was any static function changed/introduced throughout the history?
total_changed_functions = 0 # How often was any function changed/introduced throughout the history? (incl. static)
prev_commit = None
prev_hashes = None
counter = 1
for info in commits:
print "%d/%d" % (counter, len(commits))
counter += 1
commit = info['commit']
parent = info['parent']
if not parent: # first commit has no parent
print "No parent"
continue
commit_data = read_chash_data(commit)
if not commit_data:
# If the data does not exist, note and skip
#print "Data empty"
stats['data-empty'].add(commit)
continue
local_hashes = {}
for element in commit_data:
local_hashes[element[0]] = element[1]
parent_hashes = {}
if parent == prev_commit:
#print "Reuse prev_commit"
parent_hashes = prev_hashes
else:
#print "Cannot reuse prev_commit"
parent_data = read_chash_data(parent)
for element in parent_data:
parent_hashes[element[0]] = element[1]
if not parent_hashes:
# If the data does not exist, note and skip
stats['data-empty'].add(commit)
# Save data for reuse
prev_commit = commit
prev_hashes = local_hashes
continue
#########################
# LOCAL HASH EVALUATION #
#########################
commit_stats = {
'element-count': len(local_hashes),
'changed-elements': [],
'changed-functions-approx': [],
}
# Get data from approximation
cwd = os.getcwd()
os.chdir(src_path)
commit_stats[
'changed-functions-approx'] = get_changed_functions_from_commit(
src_path, commit)
os.chdir(cwd)
elements = set(local_hashes.keys())
parent_elements = set(parent_hashes.keys())
commit_stats['changed-elements'] = set(
) #TODO here elements ^ parent_elements # elements either added or removed: if this is initialized with the insdel items, causes weird data to show um in result. should perhaps include it and add explanation
# Compare hashes
common_elements = elements & parent_elements
for element in common_elements:
if local_hashes[element] != parent_hashes[element]:
commit_stats['changed-elements'].add(element)
if element.startswith(
'record:'
): # do this here to ignore insertions and deletions
total_changed_records += 1
elif element.startswith(
'variable:') or element.startswith(
'static variable:'):
total_changed_globals += 1
elif element.startswith('static function:'):
total_changed_static_funcs += 1
total_changed_functions += 1
else:
total_changed_functions += 1
commit_stats['changed-element-count'] = len(
commit_stats['changed-elements'])
stats['commits'][commit] = commit_stats
# Count how often each element was changed over the whole history
for element in commit_stats['changed-elements']:
if element not in stats['elements']:
stats['elements'][element] = 0
stats['elements'][element] += 1
# Save data for reuse
prev_commit = commit
prev_hashes = local_hashes
self.build_info['stats'] = stats
eval_info = {
'nr-of-commits':
len(commits),
'change-percentage':
{}, # change percentage -> nr of commits with change < percentage
'hot-commits': {},
'total-changed-globals':
total_changed_globals,
'total-changed-records':
total_changed_records,
'total-changed-static-funcs':
total_changed_static_funcs,
'total-changed-functions':
total_changed_functions,
'total-changed-elements':
total_changed_functions + total_changed_records +
total_changed_globals,
}
# Get most changed elements
eval_info['most-changed-elements'] = {
k: v
for k, v in stats['elements'].iteritems()
if v > self.commits.value / 10
} # arbitrary value (about 10% of commits)
# Calc average nr and percentage of (changed) symbols per commit
summed_avg_change_percentage = 0
summed_changed_elements = 0
summed_total_elements = 0
commits = self.build_info['stats']['commits']
for commit in commits:
commit_stat = commits[commit]
change_percentage = len(commit_stat['changed-elements']) / float(
commit_stat['element-count'])
summed_avg_change_percentage += change_percentage
summed_changed_elements += len(commit_stat['changed-elements'])
summed_total_elements += commit_stat['element-count']
percentage = int(round(change_percentage * 100))
if percentage not in eval_info['change-percentage']:
eval_info['change-percentage'][percentage] = 0
eval_info['change-percentage'][percentage] += 1
# Identify hot commits
if percentage > self.hot_threshold_percentage.value:
eval_info['hot-commits'][commit] = (
percentage, len(commit_stat['changed-elements']),
commit_stat['element-count'])
eval_info[
'avg-change-percentage'] = summed_avg_change_percentage / float(
len(stats['commits']))
eval_info[
'avg-changed-elements'] = summed_changed_elements / eval_info[
'nr-of-commits']
eval_info['avg-total-elements'] = summed_total_elements / eval_info[
'nr-of-commits']
eval_info['nr-hot-commits'] = len(eval_info['hot-commits'])
with open(self.eval_data.path, "w+") as fd:
fd.write(repr(eval_info))
# Output the summary of this build into the statistics file.
with open(self.stats.path, "w+") as fd:
fd.write(repr(self.build_info))
'''
def plot_hash_count_histogram(hash_values, filename):
dictionary = plt.figure()
fig, ax = plt.subplots()
plt.xlabel('Prozentanteil geaenderter Elemente')
plt.ylabel('Anzahl von Commits')
ax.bar(hash_values.keys(), hash_values.values(), align='center')
fig.savefig(filename)
# clean data for plotting
data = {k:v for k,v in eval_info['change-percentage'].iteritems() if k <= 100}
plot_hash_count_histogram(data, self.hot_commits_histo.path)
changed_funcs_approx_list = []
changed_elements_list = []
for commit in commits:
commit_stat = commits[commit]
changed_functions_approx = commit_stat['changed-functions-approx']
changed_elements = commit_stat['changed-elements']
changed_funcs_approx_list.append(len(changed_functions_approx))
changed_elements_list.append(len(changed_elements))
#TODO plot changed elements vs approx. changed functions
# and also changed functions vs approx changed functions
fig, ax = plt.subplots()
ax.plot(changed_elements_list, label='Geaenderte Elemente (lokal)')
ax.plot(changed_funcs_approx_list, 'm', label='Geaenderte Funktionen (Approx)')
lgd = ax.legend(loc='center left', bbox_to_anchor=(1, 0.5)) # legend on the right
plt.xlabel('Commits')
plt.ylabel('Anzahl')
fig.savefig(self.compare_approx_elem.path, bbox_extra_artists=(lgd,), bbox_inches='tight')
'''
def variant_name(self):
return "%s-%s" % (self.project_name(), self.metadata['mode'])
def symlink_name(self):
return "%s-%s" % (self.title, self.variant_name())