Subject: Re: [boost] Review Request: boost.lockfree
From: Tim Blechmann (tim_at_[hidden])
Date: 2009-12-07 06:31:48
> I have been using this library for the past three weeks and have found
> it incredibly useful. Thank you for your excellent work, Tim!
> I have a pre-review question though. Is it possible to decouple
> container and the predefined wait lists so an user may use his own free
> list type? I have a use case which requires various container instances
> to share only one freelist instance, and I would like to implement this
> via an own free list type.
i do understand the need of sharing freelists between different
stacks/fifos, however i am not sure, what is the best way to implement it.
currently, the freelist implementation can be selected via a template
argument using a tag class. i chose this design to hide the actual
impementation of the freelist from the user, so that no user would get
the idea to implement his own freelist, which returns nodes to the os.
otoh, it would be possible to publish the freelist interface so that
people can implement their own freelist types. with a more generic
interface, the freelists may be adapted to use array indices instead of
pointers, which would help with the pointer compression problem pointed
out by andrew venikov.
i would be curious to hear, what other people think about the freelist
implementation, if the interface should be public or not and how
freelists could be shared.
-- tim_at_[hidden] http://tim.klingt.org Just what the hell is the experimental tradition? Morton Feldman
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk