Re: [Boost-users] Re: Re: Re: Re: boost::thread::joincausesassertion(pthreads)

From: Peter Dimov (pdimov_at_[hidden])
Date: 2005-03-23 08:55:51

Vladimir Prus wrote:
> Peter Dimov wrote:
>
>> None (if you are sure that this is what assert does, of course -
>> this is platform specific), except that the asserts should not fire.
>> So spending time and effort to make something happen when the
>> asserts fire is not productive.
>
> I disagree. Here, two people report that they got assertion failures.
> Good, but we have no idea why. If there were additional diagnostic,
> we'd knew the exact system error which was reported.

Hm.

Thorsten Froehlich reports that on Mac OS X pthread_join on a thread that 
has completed returns EINVAL. He is given a lecture on pthread_join. Nobody 
with Mac OS X access bothers to investigate, at least not visibly.

Kevin Wheatley reports that the assert( res == WAIT_OBJECT_0 ) is failing 
for him on Windows. Printing res (WAIT_FAILED) or GetLastError() ("invalid 
handle" would be my bet) won't give us much information as to whether the 
assert is failing due to double thread::join, thread::join on a 
default-constructed thread, or an internal error in Boost.Threads.

Your response is to add perror on Linux. :-)

However I do see your point. In these situations I usually replace

    assert( r == 0 );

with

    assert( r != EINVAL );
    assert( r != ESRCH );
    assert( r != EDEADLK );
    assert( r == 0 );

I don't claim that it's better than perror/abort when this is what assert 
does, print a message to stderr and call abort, but adding platform-specific 
error reporting sections to every assert would probably be a slippery slope.