Hi Allen, I did spend some time a couple months ago experimenting with a custom allocator with inline storage for standard containers. You can find my branch on github at https://github.com/cbodley/ceph/commits/wip-preallocator. I was aiming for a general solution that worked with non-contiguous containers, but couldn't find a good way to handle deallocations. So the approach was really only a good fit for std::vector. I also had trouble getting it to do the right thing with the copy and move operators. I found that it was trying to use the new allocator to release memory that came from the old allocator. Presumably there's a way to make this work using std::allocator_traits; have you had better luck here? I'd love to see your prototype. I tend to agree that we should go with the boost solution if we can. But if a) we can't use boost::small_vector in the near-term due to versioning, and b) we can get a custom version tested and working, it might be worth pursuing. Casey ----- Original Message ----- > Hi, > > I think we should go with small_vector for this role, yes. I found two > issues. First, there are critical defects in small_vector prior to boost > 1.6.0. Second, though it is header-only, the coupling with related > version-specific boost types (container, allocator, respective detail types, > and platform-specific selectors...) seems to make pulling it in separately > pretty unrealistic as an option. I looked at the cost of pulling in and > selectively compiling (libs) an entire boost 1.6, and that went a lot easier > (added less than a minute to my build time). > > Matt > > ----- Original Message ----- > > From: "Sage Weil" <sweil@xxxxxxxxxx> > > To: "Allen Samuels" <Allen.Samuels@xxxxxxxxxxx> > > Cc: ceph-devel@xxxxxxxxxxxxxxx, mbenjami@xxxxxxxxxx, cbodley@xxxxxxxxxx > > Sent: Wednesday, February 3, 2016 3:04:57 PM > > Subject: Re: inline vector container > > > > On Wed, 3 Feb 2016, Allen Samuels wrote: > > > One thing that the code base needs is a simple vector container that is > > > optimized for a small number of elements, i.e., for small element counts > > > we > > > avoid the malloc/free overhead. But fully supports being resized into > > > larger > > > containers that are unbounded. Of course it will need to follow all of > > > the > > > proper object copy/move semantics so that things like unique_ptr, etc > > > work > > > correctly. > > > > > > I heard rumors that there was one available in the boost world, but I > > > can’t > > > seem to find it. Do you know about one being available? > > > > http://www.boost.org/doc/libs/1_60_0/doc/html/boost/container/small_vector.html > > > > > I’ve been prototyping one myself, just to get re-familiar with all of the > > > C++11 move sementics, and trivial constructors, etc. > > > > > > The code is pretty complete, but the testing gets rather involved. > > > There’s > > > no reason to contain this if it’s available elsewhere. But if it isn’t I > > > plan on continuing this… > > > > I think Casey had prototyped something similar using a custom allocator, > > too, but I didn't actually look at the result. And Matt talked about just > > pulling the boost one in-tree until we get updated boost in the supported > > distros, but I forget if he ran into problems there or not... > > > > sage > > -- > -- > Matt Benjamin > Red Hat, Inc. > 315 West Huron Street, Suite 140A > Ann Arbor, Michigan 48103 > > http://www.redhat.com/en/technologies/storage > > tel. 734-707-0660 > fax. 734-769-8938 > cel. 734-216-5309 > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html > -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
