> On Sept. 26, 2019, 5:59 p.m., Benjamin Mahler wrote:
> > 3rdparty/stout/include/stout/posix/dynamiclibrary.hpp
> > Lines 70-74 (original), 68-72 (patched)
> > <https://reviews.apache.org/r/71388/diff/2/?file=2166909#file2166909line75>
> >
> >     hm.. should maybe release / reset to nullptr if dlclose fails as well?
> >     
> >     E.g. if some other part of the program closes it (bug), I guess whether 
> > we reset here will just mean whether we issue an additional bad close call 
> > in the destructor.
> >     
> >     Didn't mark as an issue since the error case here already means 
> > something is badly broken.

I think if the close fails it is really hard to recover in some general way 
here. If we did that we'd either implement enough intelligence to understand 
all `dlclose` error scenarios (since `dlclose` does not define its own errors 
it looks like that would involve understanding error messages?). I think 
surfacing the error to the caller instead is a better idea.


- Benjamin


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/71388/#review217951
-----------------------------------------------------------


On Sept. 25, 2019, 11:23 p.m., Benjamin Bannier wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/71388/
> -----------------------------------------------------------
> 
> (Updated Sept. 25, 2019, 11:23 p.m.)
> 
> 
> Review request for mesos, Benjamin Mahler and Jan Schlicht.
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> This not only simplifies our implementation of `DynamicLibrary`, but
> also removes the potential for accidental file descriptor leaks.
> 
> 
> Diffs
> -----
> 
>   3rdparty/stout/include/stout/posix/dynamiclibrary.hpp 
> 6a50592632246b06152bee177e933f65438c07ca 
> 
> 
> Diff: https://reviews.apache.org/r/71388/diff/2/
> 
> 
> Testing
> -------
> 
> `make check`
> 
> 
> Thanks,
> 
> Benjamin Bannier
> 
>

Reply via email to