Segfault when __repr__ throws

[henryiii]

Issue description

If a __repr__ throws, then passing an invalid object causes a segfault.

Reproducible example code

diff --git a/tests/test_exceptions.cpp b/tests/test_exceptions.cpp
index 372d0ae..8244322 100644
--- a/tests/test_exceptions.cpp
+++ b/tests/test_exceptions.cpp
@@ -218,4 +218,7 @@ TEST_SUBMODULE(exceptions, m) {
         }
     });

+    // Test repr that cannot be displayed
+    m.def("simple_bool_passthrough", [](bool x){return x;});
+
 }
diff --git a/tests/test_exceptions.py b/tests/test_exceptions.py
index b794e3d..64bdc63 100644
--- a/tests/test_exceptions.py
+++ b/tests/test_exceptions.py
@@ -179,3 +179,14 @@ def test_nested_throws(capture):
     with pytest.raises(m.MyException5) as excinfo:
         m.try_catch(m.MyException, pycatch, m.MyException, m.throws5)
     assert str(excinfo.value) == "this is a helper-defined translated exception"
+
+
+# This can often happen if you wrap a pybind11 class in a Python wrapper
+def test_invalid_repr():
+
+    class MyRepr(object):
+        def __repr__(self):
+            raise AttributeError("Example error")
+
+
+    m.simple_bool_passthrough(MyRepr())

[YannickJadoul]

This feels familiar, but I can't remember whether that's because a) we talked about it on Gitter or in an issue, in the context of something else, b) there's an open issue about it, or c) there's an open PR dealing with it.

EDIT: I was/am probably thinking about #2317

[jbarlow83]

I thought it might be due there not being a try block around the call to py::repr in pybind11.h:231 but it seems to be something else.

[henryiii]

Line 779, actually.