|
|
This consists of the following changes:
1. Add a top_block parameter to control exception handling
This restores past behavior where the scheduler catches exceptions
raised in block threads, allowing flowgraphs to continue running after
the failure of an individual block. It also adds optional new behavior,
selected by setting catch_exceptions=False to the top block, which causes
exceptions to not be caught. In this mode of operation, a std::terminate
handler can be installed to print a stack trace before the flowgraph exits.
2. Add terminate_handler to top_block_impl
This adds a terminate_handler function that prints information
about any uncaught exceptions and also attempts to print a backtrace
for said exception after which it calls std::abort. The backtrace is
printed via libunwind, which is a new optional dependency. If libunwind
is not found/installed, then the terminate handler will still print
information about the exception and then exit.
Co-Authored-By: Scott Torborg <storborg@gmail.com>
|
