276 lines
8.6 KiB
Python
Executable File
276 lines
8.6 KiB
Python
Executable File
#!/usr/bin/env python3
|
||
"""confirm_check.py - 确认提示上的每一种回答都必须被正确处理
|
||
|
||
重点是 Ctrl-C.
|
||
|
||
`confirm_tool` 用 fgets 读输入, 而 SIGINT 会让 fgets 以 EINTR 失败、返回
|
||
NULL; 代码把 NULL 当成"没人能回答, 按默认值放行". 于是**在确认提示
|
||
上按 Ctrl-C 等于批准这次操作**. run_command 侥幸自己查了中断标志才没执行,
|
||
write_file / edit_file 没查, 文件照写.
|
||
|
||
这个脚本把三种回答都跑一遍, 直接看磁盘上有没有留下东西:
|
||
Ctrl-C -> 必须拒绝
|
||
n -> 必须拒绝
|
||
y -> 必须执行 (否则就是"修过头"了)
|
||
|
||
用法:
|
||
confirm_check.py <ligent 二进制> <mock base-url>
|
||
"""
|
||
import os
|
||
import pty
|
||
import select
|
||
import sys
|
||
import time
|
||
|
||
# 探针文件放在 build/ 下, 别弄脏仓库根目录
|
||
# WRITE_TEST 写这一个; WRITE2_TEST 写 PROBE1 / PROBE2 (一轮里两次)
|
||
PROBE = "build/tmp/confirm_probe.txt"
|
||
PROBE1 = "build/tmp/confirm_probe1.txt"
|
||
PROBE2 = "build/tmp/confirm_probe2.txt"
|
||
# 只认稳定前缀: 选项列表是会变的 (现在是 Y/N/Always/neVer),
|
||
# 写死整串会在改提示文案时把测试弄坏.
|
||
MARK = "继续？"
|
||
PROMPT_READY = b"\x1b[?2004h"
|
||
|
||
|
||
def run_case(binary, base, answer, wait_confirm=6.0):
|
||
"""跑一轮: 触发 write_file → 等确认提示出现 → 回答 → 看文件在不在. """
|
||
try:
|
||
os.unlink(PROBE)
|
||
except OSError:
|
||
pass
|
||
|
||
os.environ["LIGENT_API_KEY"] = "sk-test"
|
||
pid, fd = pty.fork()
|
||
if pid == 0:
|
||
os.execv(binary, [binary, "--base-url", base, "--no-color",
|
||
"--confirm", "--no-auto-save", "--no-history"])
|
||
os._exit(127)
|
||
|
||
out = bytearray()
|
||
|
||
def pump(seconds, idle=0.3):
|
||
"""读到"安静"或超时为止"""
|
||
end = time.time() + seconds
|
||
last = time.time()
|
||
while time.time() < end:
|
||
r, _, _ = select.select([fd], [], [], 0.02)
|
||
if r:
|
||
try:
|
||
d = os.read(fd, 65536)
|
||
except OSError:
|
||
return False
|
||
if not d:
|
||
return False
|
||
out.extend(d)
|
||
last = time.time()
|
||
elif time.time() - last > idle:
|
||
return True
|
||
return True
|
||
|
||
def wait_for(text, seconds):
|
||
"""一直读到 text 出现 (或超时). 按标记等, 不靠 sleep 猜. """
|
||
end = time.time() + seconds
|
||
while time.time() < end:
|
||
if text.encode("utf-8") in bytes(out):
|
||
return True
|
||
r, _, _ = select.select([fd], [], [], 0.05)
|
||
if r:
|
||
try:
|
||
d = os.read(fd, 65536)
|
||
except OSError:
|
||
return False
|
||
if not d:
|
||
return False
|
||
out.extend(d)
|
||
return text.encode("utf-8") in bytes(out)
|
||
|
||
pump(1.2) # 横幅 + 第一个提示符
|
||
os.write(fd, "WRITE_TEST 写个文件\n".encode())
|
||
got_prompt = wait_for(MARK, wait_confirm)
|
||
if got_prompt:
|
||
os.write(fd, answer)
|
||
pump(1.5)
|
||
os.write(fd, b"\x03")
|
||
pump(0.3)
|
||
try:
|
||
os.close(fd)
|
||
except OSError:
|
||
pass
|
||
try:
|
||
os.waitpid(pid, 0)
|
||
except OSError:
|
||
pass
|
||
return got_prompt, os.path.exists(PROBE)
|
||
|
||
|
||
def run_turn_scope(binary, base):
|
||
"""Always / neVer 的作用域必须是"这一轮", 不是整个会话.
|
||
|
||
一轮里让模型连写两个文件: 第一次问、答 Always, 第二次不该再问.
|
||
然后开新的一轮 (下一条用户消息), 必须**重新问** —— 会话级的开关是
|
||
/yolo, 不是这个.
|
||
|
||
全程按标记同步 ("继续？"出现 / 回到提示符), 不靠 sleep 猜时间.
|
||
"""
|
||
for p in (PROBE1, PROBE2):
|
||
try:
|
||
os.unlink(p)
|
||
except OSError:
|
||
pass
|
||
|
||
os.environ["LIGENT_API_KEY"] = "sk-test"
|
||
pid, fd = pty.fork()
|
||
if pid == 0:
|
||
os.execv(binary, [binary, "--base-url", base, "--no-color",
|
||
"--confirm", "--no-auto-save", "--no-history"])
|
||
os._exit(127)
|
||
|
||
out = bytearray()
|
||
state = {"ready": 0, "tail": b""}
|
||
READY = b"\x1b[?2004h" # 行编辑器每次回到提示符都会发它
|
||
|
||
def note(chunk):
|
||
buf = state["tail"] + chunk
|
||
state["ready"] += buf.count(READY)
|
||
state["tail"] = buf[-(len(READY) - 1):]
|
||
|
||
def pump(seconds, idle=0.3):
|
||
end = time.time() + seconds
|
||
last = time.time()
|
||
while time.time() < end:
|
||
r, _, _ = select.select([fd], [], [], 0.02)
|
||
if r:
|
||
try:
|
||
d = os.read(fd, 65536)
|
||
except OSError:
|
||
return
|
||
if not d:
|
||
return
|
||
out.extend(d)
|
||
note(d)
|
||
last = time.time()
|
||
elif time.time() - last > idle:
|
||
return
|
||
|
||
def wait_for(pred, seconds):
|
||
end = time.time() + seconds
|
||
while time.time() < end:
|
||
if pred():
|
||
return True
|
||
r, _, _ = select.select([fd], [], [], 0.03)
|
||
if r:
|
||
try:
|
||
d = os.read(fd, 65536)
|
||
except OSError:
|
||
return False
|
||
if not d:
|
||
return False
|
||
out.extend(d)
|
||
note(d)
|
||
return pred()
|
||
|
||
def has_prompt():
|
||
return "继续？".encode("utf-8") in bytes(out)
|
||
|
||
def prompt_count():
|
||
return bytes(out).count("继续？".encode("utf-8"))
|
||
|
||
pump(1.2)
|
||
|
||
# ---- 第一轮: 连写两个 ----
|
||
base_ready = state["ready"]
|
||
os.write(fd, "WRITE2_TEST 写两个文件\n".encode())
|
||
wait_for(has_prompt, 8.0)
|
||
os.write(fd, b"Always\n")
|
||
# 等这一轮彻底跑完 (回到提示符), 别用固定时长猜
|
||
wait_for(lambda: state["ready"] > base_ready, 10.0)
|
||
pump(0.3)
|
||
first_turn_prompts = prompt_count()
|
||
both_written = os.path.exists(PROBE1) and os.path.exists(PROBE2)
|
||
|
||
|
||
# ---- 第二轮: 换一条用户消息, 必须重新问 ----
|
||
base_ready = state["ready"]
|
||
before = prompt_count()
|
||
os.write(fd, "WRITE_LAST_TEST 再来一轮\n".encode())
|
||
got_again = wait_for(lambda: prompt_count() > before, 8.0)
|
||
second_prompts = prompt_count() - before
|
||
os.write(fd, b"N\n")
|
||
pump(1.0)
|
||
|
||
os.write(fd, b"\x03")
|
||
pump(0.3)
|
||
try:
|
||
os.close(fd)
|
||
except OSError:
|
||
pass
|
||
try:
|
||
os.waitpid(pid, 0)
|
||
except OSError:
|
||
pass
|
||
|
||
for p in (PROBE1, PROBE2):
|
||
try:
|
||
os.unlink(p)
|
||
except OSError:
|
||
pass
|
||
return first_turn_prompts, both_written, got_again, second_prompts
|
||
|
||
|
||
def main():
|
||
if len(sys.argv) < 3:
|
||
print("用法: confirm_check.py <二进制> <base-url>")
|
||
return 2
|
||
binary, base = sys.argv[1], sys.argv[2]
|
||
os.makedirs(os.path.dirname(PROBE), exist_ok=True)
|
||
|
||
fails = 0
|
||
checks = 0
|
||
|
||
def check(name, ok, detail=""):
|
||
nonlocal fails, checks
|
||
checks += 1
|
||
if ok:
|
||
print(" ✓ %s" % name)
|
||
else:
|
||
fails += 1
|
||
print(" ✗ %s%s" % (name, (" " + detail) if detail else ""))
|
||
|
||
# 三种回答: (说明, 发送的字节, 期望文件是否被写)
|
||
cases = [
|
||
("在确认提示上按 Ctrl-C", b"\x03", False),
|
||
("答 n", b"n\n", False),
|
||
("答 y", b"y\n", True),
|
||
]
|
||
|
||
for label, answer, expect_written in cases:
|
||
got, written = run_case(binary, base, answer)
|
||
check("%s: 等到了确认提示" % label, got, "没等到确认提示")
|
||
check("%s: 文件%s" % (label, "被写入" if expect_written else "没被写入"),
|
||
written == expect_written,
|
||
"实际 %s" % ("写了" if written else "没写"))
|
||
|
||
# 作用域: Always 只活这一轮
|
||
n1, both, again, n2 = run_turn_scope(binary, base)
|
||
check("Always: 同一轮里只问一次", n1 == 1, "实际问了 %d 次" % n1)
|
||
check("Always: 同轮的第二次改动被放行", both, "第二个文件没写")
|
||
check("新一轮会重新询问 (作用域仅限本轮)", again and n2 >= 1,
|
||
"新一轮没有询问")
|
||
|
||
for p in (PROBE, PROBE1, PROBE2):
|
||
try:
|
||
os.unlink(p)
|
||
except OSError:
|
||
pass
|
||
# 这里**不要** rmtree("build/tmp"): 那是 e2e 各段共用的临时目录, 删掉它
|
||
# 会连带清掉别人放进去的东西 (我就因此丢过两次日志), 而它自己的三个文件
|
||
# 上面已经删干净了. 测试辅助脚本只该收拾自己造的东西.
|
||
|
||
print("\n%d 项检查, %d 项失败" % (checks, fails))
|
||
return 1 if fails else 0
|
||
|
||
|
||
if __name__ == "__main__":
|
||
sys.exit(main())
|