Subject: Re: [boost] Are any of these bugs _really_ show stoppers?
From: Neil Groves (neil_at_[hidden])
Date: 2010-01-27 03:22:33
> Sorry for being so late, but I don't think it's a defect (and it's a
> showstopper because no other has reclaimed such functionality). MAP_FIXED is
> dangerous and isn't portable so if the desired address is already mapped the
> mapping should fail. If you want to use unix behavior you can always use
> munmap to unmap the desired address range and try that address.
I would not argue with a change in the priority. For my use case it is a
show stopper since while I can mmap and pass the address. I cannot guarantee
the mapping occurs at the address I supplied. There is a perfectly valid use
case for mapping the same physical address multiple times that is frequently
used in the implementation of high performance circular buffers. I therefore
believe that there is not an option for me to retain the functionality
present in the earlier versions of Boost.Interprocess, even by adding system
calls in my code. There are only two options available to me to get my code
to work. The first is to patch the library to allow MAP_FIXED, the other is
to not use Boost.Interprocess. I greatly appreciate the library hence I
would rather negotiate a change. I also appreciate that I am only the second
developer to have hit this issue. I propose an opt-in facility to MAP_FIXED
and would be happy to provide an implementation.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk