Re: [boost] Git Modularization example Boost .gitattributes

Subject: Re: [boost] Git Modularization example Boost .gitattributes
From: Niall Douglas (ndouglas_at_[hidden])
Date: 2013-07-16 13:10:12

 > > As a related question, myself and Paul are hoping to have proposed
> > Boost.AFIO sandbox ready for GSoC Milestone 1 end of this month. Do we
> > still submit to Boost.Sandbox SVN and let that mirror to Git, or
> > seeing as it's already in GIT at https://github.com/BoostGSOC do we
> > use some method of mirroring that directly to Git Modular Boost?
> 
> I don't think you need to do any mirroring at all, unless you submit it
for review
> and there's some kind of official policy that all reviews have to use
Boost's SVN
> sandbox... and even if there is, we might want to change that policy
before your
> review date.  I'd say, sit tight for now, and ask the Review Wizards when
you
> apply for the review whether you need to change something.

Ok, I'll drop a line to the Review Wizards now to give them a heads up.
We're actually intending something a bit different for Milestone 1: we're
going to ask this list (and boost-announce if we're allowed) for what are
the most commonly desired new features. Right now, proposed AFIO is
absolutely barebones, and there is a long list of what we /could/ add but we
have no idea what is actually desired.

For example, the existing AFIO dependency based write ordering enforcement
mechanism currently has two implementations, one based on fsync() and the
other based on O_SYNC, and neither of which are fast. BTRFS and ZFS [1] have
the ability to O(1) snapshot the filing system and therefore enforce write
ordering far, far more quickly. The question then becomes, is such a feature
desirable enough to potential AFIO users that we choose implementing that
over let's say, a special Linux kaio driver whose sole benefit is to reduce
thread count overheads on Linux to the same level as Windows?

Anyway that's still a week or two out. Got to get Jenkins working with all
major C++11 compilers on all supported platforms first. That's Paul's next
job for this week.

Thanks Dave.

Niall

[1]: Neither BTRFS nor ZFS do O_DIRECT well (ZFS simply refuses), which
makes O_SYNC performance even worse. An alternative write ordering
enforcement for these COW based filing systems is therefore especially
attractive.

---
Opinions expressed here are my own and do not necessarily represent those of
BlackBerry Inc.