class DatabaseGenerator(Experiment):
inputs = {'mode': Integer(0)}
outputs = {
'keydict':
TableDict(),
"table2":
Table([("version", "integer"), ("method", "text"), ("result", "text")],
keys=["version", "method"],
conflict_strategy="REPLACE")
}
def run(self):
for i in range(0, self.mode.value + 3):
self.keydict[str(i)] = "barfoo " + str(i)
if self.mode.value == 0:
self.table2.insert(version=1, method="GET", result="404") # 1
self.table2.insert(version=1, method="GET",
result="404") # 2: replace 1
self.table2.insert(version=1, method="POST", result="200") # 3
else:
self.table2.insert(version=1, method="GET",
result="200") # 4: replace 2
self.table2.insert(version=1, method="GET",
result="300") # 5: replace 4
self.table2.insert(version=1, method="DROP", result="404") # 6
self.table2.insert(version=1, method="POST",
result="200") # 7: replace 3
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())
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())
class TimingInternal(Experiment, ClangHashHelper):
inputs = {
"clang_hash":
GitArchive("/home/stettberger/w/clang-hash/"),
"project":
GitArchive("/home/stettberger/w/clang-hash/hash-projects/lua",
shallow=True),
"cflags":
String(""),
"jobs":
Integer(4),
"mode":
String("normal"), # Unchangable
}
outputs = {
"stats": File("summary.dict"),
'tex': DatarefDict('data.dref'),
}
def save(self, path, value):
self.tex['/'.join(path)] = value
logging.info("%s = %s", '/'.join(path), value)
def run(self):
with self.clang_hash as cl_path:
logging.info("Cloning clang hash...")
logging.info("Cloning project... %s", self.project_name())
# First, we redirect all calls to the compiler to our
# gcc wrapper
CC = os.path.join(cl_path, "wrappers/gcc-time")
os.environ["CC"] = CC
os.environ["TIMING_REPORT"] = self.stats.path
os.environ["CHASH_EXTRA_FLAGS"] = self.cflags.value
with self.project as src_path:
info = {}
self.call_configure(src_path)
self.rebuild(src_path, info, True)
collect = defaultdict(list)
compiler_calls = 0
with open(self.stats.path) as fd:
for line in fd.readlines():
data = eval(line)
if "name" in data:
compiler_calls += 1
for key in data:
if type(data[key]) is float:
collect[key].append(data[key])
self.save([self.project_name(), "phase", self.cflags.value, "count"],
compiler_calls)
for phase in collect:
if phase in ("preprocessing", "parser (global)",
"phase opt and generate"):
self.save(
[self.project_name(), "phase", phase, self.cflags.value],
np.average(collect[phase]))
def symlink_name(self):
return "%s-%s%s" % (self.title, self.project_name(), self.cflags.value)
class HistoricalCompilation(Experiment, ClangHashHelper):
inputs = {
"clang_hash": GitArchive("/home/stettberger/w/clang-hash/"),
"project": GitArchive("/home/stettberger/w/clang-hash/hash-projects/lua"),
"mode": String("normal"),
"commits": Integer(500),
"jobs": Integer(4),
}
outputs = {
"stats": File("summary.dict"),
"ccache_stats": File("ccache.stats"),
"clang_hash_log": File("clang-hash.log"),
}
def build_parent(self, commit, from_scratch = False):
def eq_hash(a, b):
if len(a) == 0 or len(b) == 0:
return
if len(a) > len(b):
return a.startswith(b)
else:
return b.startswith(a)
src_path = self.project.path
if from_scratch:
shell("cd %s; git clean -dfx -e '*.hash' -e '*.hash.copy'", src_path)
logging.info("Parent [%s^]: clean build", commit)
shell("cd %s; git reset --hard %s^", src_path, commit)
info = {"commit": commit + "^"}
self.call_configure(src_path)
self.rebuild(src_path, info, True)
# Did initial commit fail? Try again
if info.get("failed"):
logging.info("Parent[%s^]: failed", commit)
return False
return True
else:
(lines, _) = shell("cd %s; git rev-parse %s^",
src_path, commit)
parent_revision = lines[0].strip()
if self.current_revision and eq_hash(self.current_revision, parent_revision):
logging.info("Parent[%s^]: resuse good parent", commit)
return True
else:
logging.info("Parent[%s^]: resuse similar build directory", commit)
shell("cd %s; git reset --hard %s^", src_path, commit)
info = {"commit": commit +"^"}
self.call_reconfigure(src_path)
self.rebuild(src_path, info, True)
# Did initial commit fail? Try again
if info.get("failed"):
return self.build_parent(commit, from_scratch=True)
return True
def run(self):
# Determine the mode
modes = ('normal', 'ccache', 'clang-hash', 'ccache-clang-hash')
if not self.mode.value in modes:
raise RuntimeError("Mode can only be one of: %s"%modes)
logging.info("Build the Clang-Hash Plugin")
with self.clang_hash as cl_path:
shell("cd %s; mkdir build; cd build; cmake .. -DCMAKE_BUILD_TYPE=Release; make -j 4", cl_path)
shell("strip %s/build/clang-plguin/*.so", cl_path)
# 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:
(commits, _) = shell("cd %s; git log --no-merges --oneline --topo-order --format='%%H %%P %%s'", src_path)
# [0] is hash. [1] is parent, [2] rest
commits = [x.split(" ", 2) for x in reversed(commits)]
commits = commits[-self.commits.value:]
self.current_revision = None
# First, we redirect all calls to the compiler to our
# clang hash wrapper
self.setup_compiler_paths(cl_path)
time = 0
last_failed = True
while commits:
# Search for a child of the current revision
commit = None
if self.current_revision:
for idx in range(0, len(commits)):
if commits[idx][1] == self.current_revision:
commit = commits[idx]
del commits[idx]
break
# No Child found -> Take the first one.
if not commit:
commit = commits.pop(0)
# Bash initial commit
if commit[0] == "726f63884db0132f01745f1fb4465e6621088ccf":
continue
info = {"commit": commit[0],
"parent": commit[1],
"summary": commit[2]}
# Somehow this commit in musl is weird. It behaves
# totally different, if build with a fresh parent and
# a non-fresh parent. With this we are one the save side
if commit[0] == "416d1c7a711807384cc21a18163475cf757bbcb5":
last_failed = True
# First, we build the parent. In a total linear
# history, this is a NOP. Otherwise, we try to reset
# to the actual parent, and rebuild the project. This
# may fail, since the current commit might fix this.
ret = self.build_parent(commit[0], from_scratch = last_failed)
info['parent-ok'] = ret
# Change to the ACTUAL commit. Call reconfigure, and
# then go on building the commit.
shell("cd %s; git reset --hard %s", src_path, commit[0])
self.call_reconfigure(src_path)
if os.path.exists("/tmp/clang-hash.log"):
os.unlink("/tmp/clang-hash.log")
# Rebuild and Measure
self.rebuild(src_path, info, fail_ok=True)
if os.path.exists("/tmp/clang-hash.log") and not info.get("failed"):
with open("/tmp/clang-hash.log") as fd:
self.clang_hash_log.value += fd.read()
self.build_info["builds"].append(info)
if not info.get("failed"):
time += info['build-time'] / 1e9
# Build was good. Remember that.
self.current_revision = commit[0]
last_failed = False
else:
self.current_revision = None
last_failed = True
logging.info("Rebuild for %d commits takes %f minutes",
self.commits.value, time/60.)
# 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 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 HistoricalCompilation(Experiment, ClangHashHelper):
inputs = {
"clang_hash": GitArchive("/home/cip/2015/yb90ifym/clang-hash/"),
"project": GitArchive("/home/cip/2015/yb90ifym/lua"),
"mode": String("normal"),
"commits": Integer(500), # was 500
"jobs": Integer(1), # was 4
}
outputs = {
"stats": File("summary.dict"),
"ccache_stats": File("ccache.stats"),
"clang_hash_log": File("clang-hash.log"),
}
def build_parent(self, commit, from_scratch=False):
def eq_hash(a, b):
if len(a) == 0 or len(b) == 0:
return
if len(a) > len(b):
return a.startswith(b)
else:
return b.startswith(a)
src_path = self.project.path
if from_scratch:
shell("cd %s; git clean -dfx -e '*.hash' -e '*.hash.copy'",
src_path)
logging.info("Parent [%s^]: clean build", commit)
shell("cd %s; git reset --hard %s^", src_path, commit)
info = {"commit": commit + "^"}
self.call_configure(src_path)
self.rebuild(src_path, info, True)
# Did initial commit fail? Try again
if info.get("failed"):
logging.info("Parent[%s^]: failed", commit)
return False
return True
else:
(lines, _) = shell("cd %s; git rev-parse %s^", src_path, commit)
parent_revision = lines[0].strip()
if self.current_revision and eq_hash(self.current_revision,
parent_revision):
logging.info("Parent[%s^]: resuse good parent", commit)
return True
else:
logging.info("Parent[%s^]: resuse similar build directory",
commit)
shell("cd %s; git reset --hard %s^", src_path, commit)
info = {"commit": commit + "^"}
self.call_reconfigure(src_path)
self.rebuild(src_path, info, True)
# Did initial commit fail? Try again
if info.get("failed"):
return self.build_parent(commit, from_scratch=True)
return True
def run(self):
# Determine the mode
modes = ('normal', 'ccache', 'clang-hash', 'ccache-clang-hash')
if not self.mode.value in modes:
raise RuntimeError("Mode can only be one of: %s" % modes)
logging.info("Build the Clang-Hash Plugin")
with self.clang_hash as cl_path:
shell(
"cd %s; mkdir build; cd build; cmake .. -DCMAKE_BUILD_TYPE=Release; make -j 4",
cl_path)
shell("strip %s/build/clang-plugin/*.so", cl_path)
# 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:
(commits, _) = shell(
"cd %s; git log --no-merges --oneline --topo-order --format='%%H %%P %%s'",
src_path)
# [0] is hash. [1] is parent, [2] rest
commits = [x.split(" ", 2) for x in reversed(commits)]
commits = commits[-self.commits.value:]
self.current_revision = None
# First, we redirect all calls to the compiler to our
# clang hash wrapper
self.setup_compiler_paths(cl_path)
time = 0
last_failed = True
nr_of_commits = len(commits)
original_commits = commits[:]
occurred_errors = {} # map commit -> [error strings]
def gather_local_hashes(src_path):
remove_keys = [
'project', 'return-code', 'start-time', 'run_id',
'compile-duration', 'processed-bytes', 'hash-duration',
'hash-start-time', 'object-file-size'
] # TODO: ofile-size useful?
hashes = read_hash_directory(src_path, remove_keys)
local_hashes = {}
for entry in hashes:
element_hashes = entry['element-hashes']
for element in element_hashes:
local_hashes[element[0]] = element[1]
return local_hashes
def gather_global_hashes(local_hashes, occurred_errors):
global_hashes = {}
for symbol in local_hashes:
symbol = symbol.split(':')[
1] # Remove the prefix ('function:' etc.)
try:
shell("cd %s; %s/clang-hash-global --definition %s",
src_path, self.inputs.clang_hash.path, symbol)
except Exception as e:
occurred_errors[commit[0]] = e
# don't raise exception
return global_hashes
def add_additional_commit_info_to(info):
gitshow = subprocess.Popen(["git", "show"],
stdout=subprocess.PIPE)
dstat_out = subprocess.check_output(('diffstat'),
stdin=gitshow.stdout)
gitshow.wait()
lines = dstat_out.split('\n')
index = -1
while lines[index] == '':
index -= 1
last_line = lines[index]
changedInsertionsDeletions = [
int(s) for s in last_line.split() if s.isdigit()
]
if "insertion" in last_line:
info['insertions'] = changedInsertionsDeletions[1]
if "deletion" in last_line:
info['deletions'] = changedInsertionsDeletions[2]
elif "deletion" in last_line:
info['deletions'] = changedInsertionsDeletions[1]
# Get changed files
changed_files = {}
for line in lines:
if '|' in line:
elems = line.split()
assert elems[1] == '|'
filename = elems[0]
nr_of_changes = int(elems[2])
changed_files[filename] = nr_of_changes
assert len(changed_files) == changedInsertionsDeletions[0]
info['changes'] = changed_files
while commits:
# Search for a child of the current revision
commit = None
if self.current_revision:
for idx in range(0, len(commits)):
if commits[idx][1] == self.current_revision:
commit = commits[idx]
del commits[idx]
break
# No Child found -> Take the first one.
if not commit:
commit = commits.pop(0)
info = {
"commit": commit[0],
"parent": commit[1],
"summary": commit[2]
}
# First, we build the parent. In a total linear
# history, this is a NOP. Otherwise, we try to reset
# to the actual parent, and rebuild the project. This
# may fail, since the current commit might fix this.
ret = self.build_parent(commit[0], from_scratch=last_failed)
info['parent-ok'] = ret
parent_info = {}
add_additional_commit_info_to(parent_info)
info['parent-info'] = parent_info
# Gather hashes of parent
parent_local_hashes = gather_local_hashes(src_path)
parent_global_hashes = gather_global_hashes(
parent_local_hashes, occurred_errors)
#info['parent-local-hashes'] = parent_local_hashes
#info['parent-global-hashes'] = parent_global_hashes
# Change to the ACTUAL commit.
shell("cd %s; git reset --hard %s", src_path, commit[0])
add_additional_commit_info_to(info)
# Call reconfigure, and then go on building the commit.
self.call_reconfigure(src_path)
if os.path.exists("/tmp/clang-hash.log"):
os.unlink("/tmp/clang-hash.log")
# Rebuild and Measure
self.rebuild(src_path, info, fail_ok=True)
# Don't need those atm
del info['clang-hash-hits']
del info['clang-hash-misses']
# Gather hashes
local_hashes = gather_local_hashes(src_path)
global_hashes = gather_global_hashes(local_hashes,
occurred_errors)
#info['local-hashes'] = local_hashes
#info['global-hashes'] = global_hashes
# Compare hashes/search for changed hashes
# The parent's global hashes are copied to find removed symbols
changed_symbols = {}
parent_hashes = deepcopy(parent_global_hashes)
for symbol, global_hash in global_hashes.iteritems():
parent_global_hash = parent_hashes.pop(symbol, None)
if global_hash != parent_global_hash:
# Store it as [before, after]
changed_symbols[symbol] = [
parent_global_hash, global_hash
]
# Add removed symbols
for symbol, parent_global_hash in parent_hashes.iteritems():
changed_symbols[symbol] = [parent_global_hash, None]
# Compare hashes/search for changed hashes
# The parent's global hashes are copied to find removed symbols
local_changed_symbols = {}
parent_hashes = deepcopy(parent_local_hashes)
for symbol, local_hash in local_hashes.iteritems():
parent_local_hash = parent_hashes.pop(symbol, None)
if local_hash != parent_local_hash:
# Store it as [before, after]
local_changed_symbols[symbol] = [
parent_local_hash, local_hash
]
# Add removed symbols
for symbol, parent_local_hash in parent_hashes.iteritems():
local_changed_symbols[symbol] = [parent_local_hash, None]
info['changed-symbols'] = changed_symbols
#info['local-changed-symbols'] = local_changed_symbols
info['local-changed-sym-count'] = len(local_changed_symbols)
# TODO: add more analysis
# TODO: for each changed local hash, the symbol's global hash should also change...
# check every symbol for changed global hash\
# also check the commits, if the correct ones are used...
if os.path.exists(
"/tmp/clang-hash.log") and not info.get("failed"):
with open("/tmp/clang-hash.log") as fd:
self.clang_hash_log.value += fd.read()
self.build_info["builds"].append(info)
if not info.get("failed"):
time += info['build-time'] / 1e9
# Build was good. Remember that.
self.current_revision = commit[0]
last_failed = False
else:
self.current_revision = None
last_failed = True
logging.info("Rebuild for %d commits takes %f minutes",
self.commits.value, time / 60.)
print "\n\noccurred errors:\n"
print occurred_errors
print "\n\nchanged symbols:\n"
print changed_symbols
print "\n\nlocal changed symbols:\n"
print local_changed_symbols
print "\n\n\n"
if len(changed_symbols) or len(local_changed_symbols):
print "!!! success: found one !!!"
# 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 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 IncrementalCompilation(Experiment, ClangHashHelper):
inputs = {
"clang_hash":
GitArchive("/home/stettberger/w/clang-hash/"),
"project":
GitArchive("/home/stettberger/w/clang-hash/hash-projects/musl",
shallow=True),
"touch-only":
Bool(False),
"mode":
String("normal"),
"jobs":
Integer(4),
}
outputs = {
"stats": File("summary.dict"),
}
def get_sources(self, path):
ret = []
for root, dirnames, filenames in os.walk(path):
for filename in filenames:
if filename.endswith(('.h', '.c')):
ret.append(os.path.join(root, filename))
if self.project_name() == "musl":
# We do not touch headers that are external, since they
# are untouchable.
ret = [x for x in ret if x.endswith(".c") or "internal" in x]
return sorted(ret)
def touch(self, path):
if self.touch_only.value:
os.utime(path, None)
else:
with open(path) as fd:
content = fd.read()
content = "#line 1\n" + content
with open(path, "w") as fd:
fd.write(content)
def run(self):
# Determine the mode
modes = ('normal', 'ccache', 'clang-hash')
if not self.mode.value in modes:
raise RuntimeError("Mode can only be one of: %s" % modes)
logging.info("Build the Clang-Hash Plugin")
with self.clang_hash as cl_path:
shell("cd %s; mkdir build; cd build; cmake ..; make -j 4", cl_path)
# 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:
# First, we redirect all calls to the compiler to our
# clang hash wrapper
self.setup_compiler_paths(cl_path)
# Count the number of files
sources = list(self.get_sources(src_path))
nr_files = len(sources)
logging.info("#files: %d", nr_files)
self.build_info['file-count'] = nr_files
# Initial build of the given project
self.call_configure(src_path)
info = {"filename": "FRESH_BUILD"}
self.rebuild(src_path, info)
self.build_info["builds"].append(info)
# Iterate over all files
for fn in sources:
self.touch(fn)
info = {"filename": fn}
self.rebuild(src_path, info)
self.build_info["builds"].append(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 method_name(self):
mod = "append"
if self.metadata['touch-only']:
mod = "touch"
return "%s-%s" % (mod, self.metadata['mode'])
def variant_name(self):
return "%s-%s" % (self.project_name(), self.method_name())
def symlink_name(self):
return "%s-%s" % (self.title, self.variant_name())