Fix enforce test failed and demangle stack info

[gangliao]

Note: If no symbol with a suitable value is found, both this field and dli_saddr shall be set to NULL.

Fix: #4072

[gangliao]

fixed enforce

[gangliao]

bug is here:

dli_sname		The name of the nearest runtime symbol with value less than or equal to addr. Where possible, the symbol name shall be returned as it would appear in C source code.If no symbol with a suitable value is found, both this field and dli_saddr shall be set to NULL.

[gangliao]

Pass under internal test machine

69: Test command: /home/liaogang/github/Paddle/build/paddle/platform/enforce_test
69: Test timeout computed to be: 9.99988e+06
69: Running main() from gtest_main.cc
69: [==========] Running 19 tests from 9 test cases.
69: [----------] Global test environment set-up.
69: [----------] 3 tests from ENFORCE
69: [ RUN      ] ENFORCE.OK
69: [       OK ] ENFORCE.OK (0 ms)
69: [ RUN      ] ENFORCE.FAILED
69: [       OK ] ENFORCE.FAILED (1 ms)
69: [ RUN      ] ENFORCE.NO_ARG_OK
69: [       OK ] ENFORCE.NO_ARG_OK (0 ms)
69: [----------] 3 tests from ENFORCE (1 ms total)
69:
69: [----------] 2 tests from ENFORCE_EQ
69: [ RUN      ] ENFORCE_EQ.NO_EXTRA_MSG_FAIL
69: [       OK ] ENFORCE_EQ.NO_EXTRA_MSG_FAIL (0 ms)
69: [ RUN      ] ENFORCE_EQ.EXTRA_MSG_FAIL
69: [       OK ] ENFORCE_EQ.EXTRA_MSG_FAIL (0 ms)
69: [----------] 2 tests from ENFORCE_EQ (0 ms total)
69:
69: [----------] 2 tests from ENFORCE_NE
69: [ RUN      ] ENFORCE_NE.OK
69: [       OK ] ENFORCE_NE.OK (0 ms)
69: [ RUN      ] ENFORCE_NE.FAIL
69: [       OK ] ENFORCE_NE.FAIL (1 ms)
69: [----------] 2 tests from ENFORCE_NE (1 ms total)
69:
69: [----------] 2 tests from ENFORCE_GT
69: [ RUN      ] ENFORCE_GT.OK
69: [       OK ] ENFORCE_GT.OK (0 ms)
69: [ RUN      ] ENFORCE_GT.FAIL
69: [       OK ] ENFORCE_GT.FAIL (0 ms)
69: [----------] 2 tests from ENFORCE_GT (0 ms total)
69:
69: [----------] 2 tests from ENFORCE_GE
69: [ RUN      ] ENFORCE_GE.OK
69: [       OK ] ENFORCE_GE.OK (0 ms)
69: [ RUN      ] ENFORCE_GE.FAIL
69: [       OK ] ENFORCE_GE.FAIL (0 ms)
69: [----------] 2 tests from ENFORCE_GE (0 ms total)
69:
69: [----------] 2 tests from ENFORCE_LE
69: [ RUN      ] ENFORCE_LE.OK
69: [       OK ] ENFORCE_LE.OK (0 ms)
69: [ RUN      ] ENFORCE_LE.FAIL
69: [       OK ] ENFORCE_LE.FAIL (0 ms)
69: [----------] 2 tests from ENFORCE_LE (0 ms total)
69:
69: [----------] 2 tests from ENFORCE_LT
69: [ RUN      ] ENFORCE_LT.OK
69: [       OK ] ENFORCE_LT.OK (0 ms)
69: [ RUN      ] ENFORCE_LT.FAIL
69: [       OK ] ENFORCE_LT.FAIL (1 ms)
69: [----------] 2 tests from ENFORCE_LT (1 ms total)
69:
69: [----------] 2 tests from ENFORCE_NOT_NULL
69: [ RUN      ] ENFORCE_NOT_NULL.OK
69: [       OK ] ENFORCE_NOT_NULL.OK (0 ms)
69: [ RUN      ] ENFORCE_NOT_NULL.FAIL
69: [       OK ] ENFORCE_NOT_NULL.FAIL (0 ms)
69: [----------] 2 tests from ENFORCE_NOT_NULL (0 ms total)
69:
69: [----------] 2 tests from ENFORCE_USER_DEFINED_CLASS
69: [ RUN      ] ENFORCE_USER_DEFINED_CLASS.EQ
69: [       OK ] ENFORCE_USER_DEFINED_CLASS.EQ (0 ms)
69: [ RUN      ] ENFORCE_USER_DEFINED_CLASS.NE
69: [       OK ] ENFORCE_USER_DEFINED_CLASS.NE (0 ms)
69: [----------] 2 tests from ENFORCE_USER_DEFINED_CLASS (0 ms total)
69:
69: [----------] Global test environment tear-down
69: [==========] 19 tests from 9 test cases ran. (3 ms total)
69: [  PASSED  ] 19 tests.
1/1 Test #69: enforce_test .....................   Passed    0.01 sec

The following tests passed:
	enforce_test

100% tests passed, 0 tests failed out of 1

Total Test time (real) =   0.02 sec

[qingqing01]

我在Python单测里测了下：

case 1 可以正常显示：

126: Traceback (most recent call last):
126:   File "test_mul_op.py", line 13, in setUp
126:     self.outputs = {'Out': np.dot(self.inputs['X'], self.inputs['Y'])}
126: ValueError: shapes (32,84) and (80,100) not aligned: 84 (dim 1) != 80 (dim 0)

case 2 只报 Exception: SegFault ：

126: Test timeout computed to be: 9.99988e+06
1/1 Test #126: test_mul_op ......................***Exception: SegFault 45.98 sec

0% tests passed, 1 tests failed out of 1

Total Test time (real) =  46.00 sec

[Xreki]

@qingqing01 求问case 1是哪个，case 2是哪个？我碰到过很多次SegFault，大概情况是python没有设置正确的Output。

[qingqing01]

@Xreki 用MulOp测的：

case 1： shape改成不匹配:

'Y': np.random.random((84, 100)) -> 'Y': np.random.random((80, 100))

class TestMulOp(OpTest):
    def setUp(self):
        self.op_type = "mul"
        self.inputs = {
            'X': np.random.random((32, 84)).astype("float32"),
            'Y': np.random.random((80, 100)).astype("float32")
        }
        self.outputs = {'Out': np.dot(self.inputs['X'], self.inputs['Y'])}

    def test_check_output(self):
        self.check_output()

case 2: 改错名字Y->Y0

class TestMulOp(OpTest):
    def setUp(self):
        self.op_type = "mul"
        self.inputs = {
            'X': np.random.random((32, 84)).astype("float32"),
            'Y0': np.random.random((84, 100)).astype("float32")
        }
        self.outputs = {'Out': np.dot(self.inputs['X'], self.inputs['Y0'])}

    def test_check_output(self):
        self.check_output()

case2有可能是C++代码内部check做的不好。

[qingqing01]

LGTM.

[qingqing01]

但其实希望case2也能有错误栈信息~

[Xreki]

case2很奇怪，就是C++端createOp时，名字写错，是有提示和stack信息的。python端写错就是SegFault，很奇怪。这两种情况我都在#3899 里面提到过。

[Xreki]

@qingqing01 case1的错误，应该是python报的，因为单测里面会执行np.dot()，会检查数据的规模。