|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
import llvm_helper |
|
import json |
|
import os |
|
import dateparser |
|
import time |
|
import datetime |
|
import re |
|
from typing import Tuple, Sequence, Union |
|
|
|
|
|
class TimeCompensationGuard: |
|
def __init__(self, environment): |
|
self.environment = environment |
|
|
|
def __enter__(self): |
|
self.start_time = time.time() |
|
self.environment.interaction_time_compensation_enter += 1 |
|
|
|
def __exit__(self, exception_type, exception_value, exception_traceback): |
|
self.environment.interaction_time_compensation_enter -= 1 |
|
if self.environment.interaction_time_compensation_enter == 0: |
|
self.environment.interaction_time_compensation += ( |
|
time.time() - self.start_time |
|
) |
|
|
|
|
|
class Environment: |
|
def __init__( |
|
self, |
|
issue_id, |
|
base_model_knowledge_cutoff: str, |
|
*, |
|
max_build_jobs=None, |
|
max_test_jobs=None, |
|
): |
|
with open(os.path.join(llvm_helper.dataset_dir, f"{issue_id}.json")) as f: |
|
self.data = json.load(f) |
|
self.base_commit = self.data["base_commit"] |
|
self.knowledge_cutoff = dateparser.parse(self.data["knowledge_cutoff"]) |
|
self.bug_type = self.data["bug_type"] |
|
self.test_commit = self.data.get( |
|
"test_commit", self.data["hints"]["fix_commit"] |
|
) |
|
self.used_knowledge = dict() |
|
self.use_knowledge("base_model", base_model_knowledge_cutoff) |
|
self.interaction_time_compensation = 0.0 |
|
self.interaction_time_compensation_enter = 0 |
|
self.build_count = 0 |
|
self.build_failure_count = 0 |
|
self.fast_check_count = 0 |
|
self.full_check_count = 0 |
|
self.fast_check_pass = False |
|
self.full_check_pass = False |
|
if max_build_jobs is None: |
|
self.max_build_jobs = os.cpu_count() |
|
else: |
|
self.max_build_jobs = max_build_jobs |
|
if max_test_jobs is None: |
|
self.max_test_jobs = max_build_jobs |
|
else: |
|
self.max_test_jobs = max_test_jobs |
|
self.start_time = time.time() |
|
|
|
def use_knowledge(self, url: str, date: Union[str, datetime.datetime]): |
|
if isinstance(date, str): |
|
date = dateparser.parse(date) |
|
if date <= self.knowledge_cutoff: |
|
self.used_knowledge[url] = min(self.used_knowledge.get(url, date), date) |
|
else: |
|
raise ValueError("Knowledge is newer than the cutoff date") |
|
|
|
def reset(self): |
|
with TimeCompensationGuard(self): |
|
llvm_helper.reset(self.base_commit) |
|
|
|
def verify_head(self): |
|
head = llvm_helper.git_execute(["rev-parse", "HEAD"]).strip() |
|
if head != self.base_commit: |
|
raise RuntimeError("invalid HEAD") |
|
|
|
def build(self): |
|
with TimeCompensationGuard(self): |
|
self.build_count += 1 |
|
self.verify_head() |
|
res, log = llvm_helper.build(self.max_build_jobs) |
|
if not res: |
|
self.build_failure_count += 1 |
|
return res, log |
|
|
|
def dump(self, log=None): |
|
wall_time = time.time() - self.start_time - self.interaction_time_compensation |
|
self.verify_head() |
|
patch = llvm_helper.git_execute(["diff", "--", "llvm/lib/*", "llvm/include/*"]) |
|
used_knowledge = [] |
|
for url, t in self.used_knowledge.items(): |
|
used_knowledge.append((url, t.strftime("%Y-%m-%d%z"))) |
|
return { |
|
"wall_time": wall_time, |
|
"knowledge": used_knowledge, |
|
"build_count": self.build_count, |
|
"build_failure_count": self.build_failure_count, |
|
"fast_check_count": self.fast_check_count, |
|
"full_check_count": self.full_check_count, |
|
"fast_check_pass": self.fast_check_pass, |
|
"full_check_pass": self.full_check_pass, |
|
"patch": patch, |
|
"log": log, |
|
} |
|
|
|
def check_fast(self): |
|
with TimeCompensationGuard(self): |
|
self.fast_check_count += 1 |
|
res, reason = self.build() |
|
if not res: |
|
return (False, reason) |
|
res, log = llvm_helper.verify_test_group( |
|
repro=False, input=self.data["tests"], type=self.bug_type |
|
) |
|
if not res: |
|
return (False, log) |
|
self.fast_check_pass = True |
|
return (True, log) |
|
|
|
def check_full(self): |
|
with TimeCompensationGuard(self): |
|
self.full_check_count += 1 |
|
res, reason = self.build() |
|
if not res: |
|
return (False, reason) |
|
res, log = llvm_helper.verify_test_group( |
|
repro=False, input=self.data["tests"], type=self.bug_type |
|
) |
|
if not res: |
|
return (False, log) |
|
self.fast_check_pass = True |
|
res, log = llvm_helper.verify_lit( |
|
test_commit=self.test_commit, |
|
dirs=self.data["lit_test_dir"], |
|
max_test_jobs=self.max_build_jobs, |
|
) |
|
if not res: |
|
return (False, log) |
|
self.full_check_pass = True |
|
return (True, log) |
|
|
|
def get_bug_type(self): |
|
return self.bug_type |
|
|
|
def get_base_commit(self): |
|
return self.base_commit |
|
|
|
def get_tests(self): |
|
return self.data["tests"] |
|
|
|
def get_hint_fix_commit(self): |
|
self.use_knowledge("hint:fix_commit", self.knowledge_cutoff) |
|
return self.data["hints"].get("fix_commit") |
|
|
|
def get_hint_components(self): |
|
self.use_knowledge("hint:components", self.knowledge_cutoff) |
|
return self.data["hints"].get("components") |
|
|
|
def get_hint_files(self): |
|
self.use_knowledge("hint:files", self.knowledge_cutoff) |
|
lineno = self.data["hints"].get("bug_location_lineno") |
|
if lineno is None: |
|
return None |
|
return sorted(lineno.keys()) |
|
|
|
def get_hint_bug_functions(self): |
|
self.use_knowledge("hint:bug_functions", self.knowledge_cutoff) |
|
return self.data["hints"].get("bug_location_funcname") |
|
|
|
def get_hint_line_level_bug_locations(self): |
|
self.use_knowledge("hint:line_level_bug_locations", self.knowledge_cutoff) |
|
return self.data["hints"].get("bug_location_lineno") |
|
|
|
def get_hint_issue(self): |
|
self.use_knowledge("hint:issue", self.knowledge_cutoff) |
|
return self.data.get("issue") |
|
|
|
def get_ir_keywords(self, ir: str): |
|
keywords = set() |
|
|
|
instruction_pattern = re.compile(r"%.+ = (\w+) ") |
|
for match in re.findall(instruction_pattern, ir): |
|
keywords.add(match) |
|
|
|
intrinsic_pattern = re.compile(r"@(llvm.\w+)\(") |
|
for match in re.findall(intrinsic_pattern, ir): |
|
keywords.add(match) |
|
keywords.discard("call") |
|
return keywords |
|
|
|
def get_langref_desc(self, keywords): |
|
self.use_knowledge("llvm/docs/LangRef.rst", self.knowledge_cutoff) |
|
return llvm_helper.get_langref_desc(keywords, self.base_commit) |
|
|