I'm not advocating that this would ever be a good idea, but I've found that you can crash Python (2.7 and 3.2 checked) by running eval on a large enough input string:
def kill_python(N):
S = '+'.join((str(n) for n in xrange(N)))
return eval(S)
On my computer S can be generated just fine, but for values of approximately N>74900, Python will fail with Segmentation fault (core dumped). Is there a limit to the length of string (or parse tree) that the interpreter can handle?
Note: I don't need to do this, to me this is a deeper question reflecting my ignorance of what goes on inside the box. I'd like to understand why Python fails here, and so catastrophically (why not throw an exception?)
解决方案
This issue is caused by a stack overflow in the CPython compiler. An easy way to reproduce the same issue is
>>> code = compile("1" + "+1" * 1000000, "", "eval")
Segmentation fault
which proves that the segfault is happening at the compile stage, not during evaluation. (Of course this is also easy to confirm with gdb.)
[Side note: For smaller expressions, the compiler would apply constant folding here anyway, so the only thing happening during the execution of the code is to load the result:
>>> code = compile("1" + "+1" * 1000, "", "eval")
>>> eval(code)
1001
>>> dis.dis(code)
1 0 LOAD_CONST 1000 (1001)
3 RETURN_VALUE
End of side note.]
This issue is a known defect. The Python developers collected several ways to crash the Python interpreter in the directory Lib/test/crashers of the source distribution. The one corresponding to this issue is Lib/test/crashers/compiler_recursion.py.