Bug in C++ pytree implementation

[shoyer]

I haven't tracked it down yet, but the first commit in #1256 fails with a core dump on OS X.

I'm guessing this is due to some missing cleanup step when a Python exception is raised.

[hawkinsp]

I think this is actually a pybind11 bug. The issue is that:
a) we are calling a pybind11-wrapped C++ method, but no overload matches
b) pybind11 calls pybind11::repr() to include the value in the error that it prints
c) the implementation of repr() raises a Python exception, which is converted into a C++ exception that pybind11 fails to catch.

I'll send a PR to upstream pybind11.

[shoyer]

Thanks for tracking this down!

…

On Thu, Sep 5, 2019 at 8:35 AM Peter Hawkins ***@***.***> wrote: I think this is actually a pybind11 bug. The issue is that: a) we are calling a pybind11-wrapped C++ method, but no overload matches b) pybind11 calls pybind11::repr() to include the value in the error that it prints c) the implementation of repr() raises a Python exception, which is converted into a C++ exception that pybind11 fails to catch. I'll send a PR to upstream pybind11. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1257>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAJJFVUNG6Y75DBBUYHN3JDQIERNPANCNFSM4IQZRXJQ> .

[hawkinsp]

I believe pybind/pybind11#2389 fixed this. (They ignored my PR. Sigh.)