Re: [Boost-users] [Boost.Test] BOOST_CHECK_THROW behaviour

Subject: Re: [Boost-users] [Boost.Test] BOOST_CHECK_THROW behaviour
From: Gavin Lambert (gavinl_at_[hidden])
Date: 2014-06-23 01:19:01

On 20/06/2014 01:03, Ghyslain Leclerc wrote:
> Then, I had the idea to check what the behaviour would
> be if the wrong exception is thrown.  I simply changed the first call
> like this :
>
> BOOST_CHECK_THROW( testRow.setCol( 30, 12 ), std::runtime_error );
> BOOST_CHECK_THROW( testRow.setCol( 30, 12l ), std::out_of_range );
>
> I was under the impression (and I might be wrong here, sorry  :-(   )
> that the Boost.Test library would (the numbers are not to mean that
> order, simply there for enumeration):
> 1- output a message saying the test has failed, that an exception was
> indeed thrown, but it was not a runtime_error
> 2- continue the rest of the test (single test currently)
> 3- terminate
>
> but I actually never get the message of step 1, the execution of the
> rest of the test seems never occur (the second BOOST_CHECK_THROW) and
> the program terminate because of my exception and not very gracefully.
>
> I guess my questions to the other users here are:
>    - Was my understanding of what BOOST_CHECK_THROW does flawed ?
>    - Is the behaviour normal ?
>    - Is there an obvious problem I am missing with the design of my
> simple class ?

If you look at the definition of BOOST_*_THROW in 
boost/test/test_tools.hpp, you can see that it only tries to catch the 
named exception; any other exceptions will be left to bubble up (and 
will likely be uncaught and result in terminate() getting called).  So 
it sounds like it's behaving as expected from that viewpoint.

(Having said that, there is some evidence of plumbing to catch all 
exception types at a higher level and do something more useful than 
calling terminate(); I haven't looked too closely at it so I don't know 
if it's involved in the behaviour you're seeing or not.  Related docs at 
http://www.boost.org/doc/libs/1_55_0/libs/test/doc/html/execution-monitor.html 
suggest that this is optional and must be explicitly enabled to operate, 
though, but I could easily be missing something.  But even if it's 
"working", it would still result in aborting at least the test case 
despite the CHECK level.)

I agree with you that it would be more useful for this to be handled 
locally.  You could try submitting that as a bug report in Trac (if it's 
not already there), or waiting for Gennadiy Rozental's response.  In the 
meantime, you'd need to write a more intelligent test yourself.

(I consider it a bug rather than a feature request because the reference 
docs for BOOST_*_THROW state that "check fails" if an unrelated 
exception is thrown, which is at least misleading at present if not 
incorrect.)