Data races in CJK multibyte codec state under free-threading
Chưa có ai nhận issue này.
- Ngôn ngữ chính
- Python
- Star
- 77.2k
- Fork
- 35.9k
- Chỉ số merge pull request
- Chỉ số pull request đang chờ
Mô tả
Bug report
Bug description:
Bug Report
Bug description
On a free-threaded CPython build, concurrent calls on the same CJK
multibyte codec object race on unsynchronized mutable C state. The decoder and stream reader use pending, pendingsize, and
state.c; the encoder and stream writer use pending and state.c.
The .errors getter/setter additionally accesses self->errors
without synchronization.
This is a sub-issue of https://github.com/python/cpython/issues/153852 , gist
Reproducers
Cases and c struct fields has data race.
.errors getter <> setter of all codec classes - on self->errors
import sys, threading, codecs
assert not sys._is_gil_enabled(), "run free-threaded: PYTHON_GIL=0"
NT = 8
ROUNDS = 4000
box = [None]
enter = threading.Barrier(NT + 1)
leave = threading.Barrier(NT + 1)
ERROR_NAMES = ("tsan_error_handler_a", "tsan_error_handler_b")
def worker(wid):
for _ in range(ROUNDS):
enter.wait()
dec = box[0]
for _ in range(6):
try:
dec.errors # reads self->errors; custom name takes Py_NewRef() path
except Exception:
pass
try:
dec.errors = ERROR_NAMES[wid & 1] # replaces and decrefs self->errors
except Exception:
pass
leave.wait()
ts = [threading.Thread(target=worker, args=(i,)) for i in range(NT)]
for t in ts:
t.start()
factory = codecs.getincrementaldecoder("euc_jp")
for r in range(ROUNDS):
dec = factory(ERROR_NAMES[r & 1])
box[0] = dec # publish the shared decoder for this round
enter.wait()
leave.wait()
for t in ts:
t.join()
print("done, no crash")
IncrementalDecoder.decode <> .reset — on pendingsize
import codecs, sys, threading
assert not sys._is_gil_enabled(), "run free-threaded: PYTHON_GIL=0"
NT = 8
ROUNDS = 4000
box = [None]
enter = threading.Barrier(NT + 1)
leave = threading.Barrier(NT + 1)
DECODER = codecs.getincrementaldecoder("euc_jp")
DECODER_STATE = (b"\xa4", 0)
def new_decoder():
decoder = DECODER()
decoder.decode(b"\xa4") # leaves one byte in self->pending
return decoder
def worker(wid):
for _ in range(ROUNDS):
enter.wait()
dec = box[0]
try:
dec.decode(b"\xa4", final=False)
except Exception:
pass
try:
dec.reset()
except Exception:
pass
leave.wait()
threads = [threading.Thread(target=worker, args=(i,)) for i in range(NT)]
for thread in threads:
thread.start()
for _ in range(ROUNDS):
box[0] = new_decoder()
enter.wait()
leave.wait()
for thread in threads:
thread.join()
print(f"done, no crash")
IncrementalDecoder.getstate <> .setstate - on pending, pendingsize, state.c
import codecs, sys, threading
assert not sys._is_gil_enabled(), "run free-threaded: PYTHON_GIL=0"
NT = 8
ROUNDS = 4000
box = [None]
enter = threading.Barrier(NT + 1)
leave = threading.Barrier(NT + 1)
DECODER = codecs.getincrementaldecoder("euc_jp")
DECODER_STATE = (b"\xa4", 0)
def new_decoder():
decoder = DECODER()
decoder.decode(b"\xa4") # leaves one byte in self->pending
return decoder
def worker(wid):
for _ in range(ROUNDS):
enter.wait()
dec = box[0]
for _ in range(6):
try:
dec.getstate() # reads self->pending/self->pendingsize
except Exception:
pass
try:
state = DECODER_STATE if wid & 1 else (b"", 0)
dec.setstate(state)
except Exception:
pass
leave.wait()
threads = [threading.Thread(target=worker, args=(i,)) for i in range(NT)]
for thread in threads:
thread.start()
for _ in range(ROUNDS):
box[0] = new_decoder()
enter.wait()
leave.wait()
for thread in threads:
thread.join()
print(f"done, no crash")
IncrementalEncoder.encode <> .reset - on pending
import codecs, sys, threading
assert not sys._is_gil_enabled(), "run free-threaded: PYTHON_GIL=0"
NT = 8
ROUNDS = 4000
box = [None]
enter = threading.Barrier(NT + 1)
leave = threading.Barrier(NT + 1)
ENCODER = codecs.getincrementalencoder("euc_jis_2004")
empty_state = ENCODER().getstate()
def new_encoder():
encoder = ENCODER()
encoder.encode("\u00e6") # leaves a Unicode string in self->pending
return encoder
def worker(wid):
for _ in range(ROUNDS):
enter.wait()
enc = box[0]
for _ in range(6):
try:
enc.encode("\u0300")
except Exception:
pass
try:
enc.reset()
except Exception:
pass
leave.wait()
threads = [threading.Thread(target=worker, args=(i,)) for i in range(NT)]
for thread in threads:
thread.start()
factory = ENCODER
for _ in range(ROUNDS):
box[0] = new_encoder()
enter.wait()
leave.wait()
for thread in threads:
thread.join()
print(f"done, no crash")
IncrementalEncoder.getstate <> .setstate - on pending
import codecs, sys, threading
assert not sys._is_gil_enabled(), "run free-threaded: PYTHON_GIL=0"
NT = 8
ROUNDS = 4000
box = [None]
enter = threading.Barrier(NT + 1)
leave = threading.Barrier(NT + 1)
ENCODER = codecs.getincrementalencoder("euc_jis_2004")
empty_state = ENCODER().getstate()
def new_encoder():
encoder = ENCODER()
encoder.encode("\u00e6") # leaves a Unicode string in self->pending
return encoder
def worker(wid):
for _ in range(ROUNDS):
enter.wait()
enc = box[0]
for _ in range(6):
try:
enc.getstate()
except Exception:
pass
try:
enc.setstate(empty_state)
except Exception:
pass
leave.wait()
threads = [threading.Thread(target=worker, args=(i,)) for i in range(NT)]
for thread in threads:
thread.start()
factory = ENCODER
for _ in range(ROUNDS):
box[0] = new_encoder()
enter.wait()
leave.wait()
for thread in threads:
thread.join()
print(f"done, no crash")
StreamReader.read/readline/readlines <> .reset - on pendingsize
import codecs
import sys
import threading
assert not sys._is_gil_enabled(), "run free-threaded: PYTHON_GIL=0"
NT = 8
ROUNDS = 4000
box = [None]
enter = threading.Barrier(NT + 1)
leave = threading.Barrier(NT + 1)
class InputStream:
def read(self, size=-1):
return b"\xa4"
def readline(self, size=-1):
return b"\xa4"
def worker(wid):
for _ in range(ROUNDS):
enter.wait()
dec = box[0]
try:
# Because StreamReader.read, .readline, .readlines all use mbstreamreader_iread,
# the code below proves every thread-conflict case of StreamReader.
box[0].reset() if wid & 1 else box[0].read(1)
except Exception:
pass
leave.wait()
threads = [threading.Thread(target=worker, args=(i,)) for i in range(NT)]
for thread in threads:
thread.start()
factory = codecs.getreader("euc_jp")
for _ in range(ROUNDS):
box[0] = factory(InputStream())
enter.wait()
leave.wait()
for thread in threads:
thread.join()
print(f"done, no crash")
StreamWriter.write/writelines <> .reset - on pending
import codecs, sys, threading
assert not sys._is_gil_enabled(), "run free-threaded: PYTHON_GIL=0"
NT = 8
ROUNDS = 4000
box = [None]
enter = threading.Barrier(NT + 1)
leave = threading.Barrier(NT + 1)
class OutputStream:
def write(self, data):
return len(data)
def worker(wid):
for _ in range(ROUNDS):
enter.wait()
# Because StreamWriter.write, .writelines, all use encoder_encode_stateful,
# the code below proves every thread-conflict case of StreamWriter.
for _ in range(6):
try:
box[0].write("\u00e6")
except Exception:
pass
try:
box[0].reset()
except Exception:
pass
leave.wait()
threads = [threading.Thread(target=worker, args=(i,)) for i in range(NT)]
for thread in threads:
thread.start()
factory = codecs.getwriter("euc_jis_2004")
for _ in range(ROUNDS):
box[0] = factory(OutputStream())
enter.wait()
leave.wait()
for thread in threads:
thread.join()
print(f"done, no crash")
CPython versions tested on:
CPython main branch
Operating systems tested on:
macOS
Hướng dẫn đóng góp
Bắt đầu từ đâu
- Đọc hết issue, rồi đọc hướng dẫn đóng góp của dự án.
- Bình luận trên issue rằng bạn sẽ nhận — tránh hai người làm cùng một việc.
- Fork repository và làm thay đổi trên một nhánh.
- Mở pull request có tham chiếu số hiệu của issue.
Hướng nghiên cứu
Bắt đầu bằng cách chạy các reproducer được cung cấp cho .errors, incremental decoder/encoder, StreamReader và StreamWriter trên một bản build CPython free-threaded. Theo dõi các lượt truy cập pending, pendingsize, state.c và errors được liệt kê trong các điểm vào codec đó; công việc được xem là hoàn tất khi các race được loại bỏ trên các thao tác đã báo cáo mà không gây hồi quy.
Do mô hình lập chỉ mục viết ra từ nội dung của issue.
Đánh giá
- Công nghệ
- c, python
- Lĩnh vực
- operating-systems
- Loại issue
- Lỗi
- Độ khó
- 5/5
- Thời gian dự kiến
- Hơn một tuần
- Mức độ hoạt động
- Sôi nổi
- Độ rõ ràng
- Khá rõ ràng
- Mức phù hợp với người mới
- 35/100