Re: [eigen] Memory Leak When Constructor of User-Defined Type Throws Exception

[Christoph Hertzberg <chtz@xxxxxxxxxxxxxxxxxxxxxxxx>]

On 19.07.2014 00:00, Moritz Klammler wrote:
> thank you for looking at the patch and sorry for my late reply.  I have
> (finally) applied your suggestions now (see attached).

Yes, at first glance all looks good. I think this evening I have a 
deeper look and push it to the dev-branch.

> The unit test is now using a static object count as suggested.  However,
> this can only check that the member-wise destruction happens.  I don't
> know of a way to check from within C++ that a 'std::malloc()'ed pointer
> was 'std::free()'ed again.  (Eigen does not use operator 'new' /
> 'delete' to manage the storage array.)

I found that ctest is actually able to run tests through valgrind. I 
failed to set this up so far, but it would certainly be nice in general.

> While adding '#ifdef EIGEN_EXCEPTIONS' to the code, I found that there
> is an existing try-catch in 'Memory.h' that is not surrounded by such
> guards.  I've left it unchanged since it has nothing to do with the
> reason for this patch.

I didn't spot any. There is only one try-catch block which is entirely 
inside an #ifdef EIGEN_EXCEPTIONS, because it is inside a macro.

> [...] Then, we can write the above code as
>
>      EIGEN_TRY
>        {
>          do_stuff();
>          if (bad_things_happened())
>            EIGEN_THROW(Ball("oh no!"));
>        }
>      EIGEN_CATCH(const Ball&)
>        {
>          do_other_stuff();
>        }

Yes, that looks somewhat cleaner. We don't have a lot of try-catch 
blocks, so maybe it's not really worth the effort.

(off-topic: `throw Ball` always reminds me of http://xkcd.com/1188/)

Christoph


--
----------------------------------------------
Dipl.-Inf., Dipl.-Math. Christoph Hertzberg
Cartesium 0.049
Universität Bremen
Enrique-Schmidt-Straße 5
28359 Bremen

Tel: +49 (421) 218-64252
----------------------------------------------