[Vipul-Cariappa]

Example:

>>> x: tuple[i32, f64] = (2, 2.4)
>>> x
(2, 2.400000)
>>> (2, 2.4, "LPython")
(2, 2.400000, "LPython")
>>> @dataclass
... class MyClass:
...   x: bool
...   y: i32
...   z: str
...
>>> MyClass(True, 12, "LPython")
MyClass(x=True, y=12, z="LPython")

[Vipul-Cariappa]

Looks like this line fails on Windows. (1639)

Commenting this one line and running the test does not seg fault in my local machine.

[Vipul-Cariappa]

On further investigation, it looks like when we get the offset of each member field of the struct from LLVM, it returns a random value (reference:

lpython/src/lpython/python_evaluator.cpp

Lines 534 to 535 in c58d31c

	size_t offset = dl.getStructLayout((llvm::StructType*)type)->getElementOffset(i);
	offsets.push_back(al, offset);

).

I am printing out the offsets, and just before it seg faults the offset values are random and high.

(lp16) C:\Users\vipul\Documents\Workspace\lpython\build16>ctest --rerun-failed --output-on-failure
Test project C:/Users/vipul/Documents/Workspace/lpython/build16
    Start 2: test_lpython
1/1 Test #2: test_lpython .....................***Failed    0.68 sec
[doctest] doctest version is "2.4.8"
[doctest] run with "--help" for options
0 0 0 0 0 0 0 0 0 0
0
4
8
0
4
8
0
4
8
0
4
8
0
4
0
4
8
0
4
8
16
0
4
8
16
24
0
2
8
0
0
0
8
0
8
16
0
10376344761626795937
2841615990688
140698833649665
140375097361
===============================================================================
C:\Users\vipul\Documents\Workspace\lpython\src\lpython\tests\test_llvm.cpp(1556):
TEST CASE:  PythonCompiler classes

C:\Users\vipul\Documents\Workspace\lpython\src\lpython\tests\test_llvm.cpp(1556): FATAL ERROR: test case CRASHED: SIGSEGV - Segmentation violation signal

===============================================================================
[doctest] test cases:  55 |  54 passed | 1 failed | 17 skipped
[doctest] assertions: 488 | 488 passed | 0 failed |
[doctest] Status: FAILURE!


0% tests passed, 1 tests failed out of 1

Total Test time (real) =   0.69 sec

The following tests FAILED:
          2 - test_lpython (Failed)
Errors while running CTest

Is this a bug in LLVM?

cc @Shaikh-Ubaid @certik

[ubaidsk]

I am printing out the offsets, and just before it seg faults the offset values are random and high.

Could you please share the test code and the offset printing modification you used here?

Is this a bug in LLVM?

Which LLVM version are you currently using? Could you please check with other LLVM versions?

[Vipul-Cariappa]

I have tested it with both LLVM 11 and 16. It fails similarly in both.

Code used to debug:
debugging_segfault_windows.patch

You can uncomment if (___flag) { std::cout << m->str() << std::endl; }, if you want to see the LLVM IR.

[Vipul-Cariappa]

@Shaikh-Ubaid, please have a look at the PR. I believe it is ready to be merged. I have commented out the failing test and opened an issue #2793.

[ubaidsk]

Looks great, thanks for this!