Paul Price added a comment - 12/Jun/17 5:28 PM Pim Schellart [X] , would you please review this? price@pap-laptop:~/LSST/afw (tickets/DM-10834=) $ git sub-patch commit 8811f041ba81d4167d46e8bfc93b458038c63e06 Author: Paul Price <price@astro.princeton.edu> Date: Mon Jun 12 17:23:39 2017 -0400   pybind: clear error if we're not going to use it To fail to do so results in a deadlock when using python multiprocessing.   diff --git a/python/lsst/afw/table/schema/schema.cc b/python/lsst/afw/table/schema/schema.cc index 4123f7f3f..a523b04fb 100644 --- a/python/lsst/afw/table/schema/schema.cc +++ b/python/lsst/afw/table/schema/schema.cc @@ -346,6 +346,7 @@ void declareSchema(py::module &mod) { try { self.attr("find")(key); } catch (py::error_already_set &err) { + err.clear(); return false; } return true;

Pim Schellart [X] (Inactive) added a comment - 12/Jun/17 5:45 PM Looks great! Thank you very much for looking into this! Definitely a gold star!

Paul Price added a comment - 12/Jun/17 7:04 PM Thanks Pim. Jenkins passed for py2 and py3 (except for ci_hsc on py3, which doesn't work yet, silly me). Merged to master.

Jim Bosch added a comment - 12/Jun/17 11:56 PM Excellent work, Paul Price . I'm embarrassed I didn't see and recommend this change: I looked at this exact piece of code, complained that it was confusing that ~error_already_set restored the Python error state, and then didn't notice that restoring the Python error state here is entirely the wrong thing to do (since having a Python error set and not returning NULL is considered a serious problem by the interpreter, and moreover we don't want to re-raise this exception). Because the old version of this code is the natural C++ idiom for catching and not re-throwing an excpetion, we should look for other places where we've used it incorrectly. Pim Schellart [X] , Paul Price , I think we should consider asking pybind11 to change the behavior of ~error_already_set , which would then require an explicit call to restore when you want to re-raise in Python instead of an explicit call to clear when you do not. Do you agree?

Pim Schellart [X] (Inactive) added a comment - 13/Jun/17 12:10 AM Jim Bosch , yes I definitely agree.