revising BGL code (isomorphism.hpp) that uses abort() about graph HOT 4 OPEN

Quite possibly. I'll ask about it. Thank you for bringing this to attention.

from graph.

Looks like the general advice is to avoid using aborts, with Peter Dimov pointing me towards https://github.com/boostorg/graph/blob/develop/include/boost/graph/isomorphism.hpp#L316 . This seems like an easy PR, so would you like the credit? If not, I should be able to get to it this weekend.

from graph.

I am not concerned with getting credit. There are a few other instances in graph:

%vjcair> # #define BOOST_LIB_VERSION "1_69"
%vjcair> pwd
/Library/Frameworks/R.framework/Resources/library/BH/include/boost/graph
%vjcair> grep abort *
grep: detail: Is a directory
grep: distributed: Is a directory
isomorphism.hpp:              abort();
grep: parallel: Is a directory
grep: planar_detail: Is a directory
grep: property_maps: Is a directory
%vjcair> grep abort */*
distributed/adjacency_list.hpp:      abort();
grep: distributed/adjlist: Is a directory
grep: distributed/detail: Is a directory
grep: parallel/detail: Is a directory
%vjcair> grep abort */*/*
distributed/detail/mpi_process_group.ipp:    abort();
distributed/detail/mpi_process_group.ipp:    abort();

from graph.

I just happened to notice this issue and yeah, wow, library code should never call abort() or exit(), except in the guise of an assert, which the code in graph/include/boost/graph/isomorphism.hpp does as well. Weird.
The abort() was introduced in 2012 as part of a fix to the algorithm and I guess it escaped review, as the same change also introduces considerable use of goto. :\ Now I'm not saying that goto is universally bad, but these dubious changes together are... suspicious.

I wouldn't recommend replacing it with throwing an exception; the abort() is really just an over-eager debugging tool.

from graph.