Date: Sat, 16 Sep 2017 12:51:26 +0100 From: Ed Schouten <ed@nuxi.nl> To: Jilles Tjoelker <jilles@stack.nl> Cc: src-committers <src-committers@freebsd.org>, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r323177 - in head: sys/compat/cloudabi sys/contrib/cloudabi usr.bin/truss Message-ID: <CABh_MKnXZiLs5jv8p=zW_4mSCiqL7f5m-kuGk7XU==a_Q3GY9g@mail.gmail.com> In-Reply-To: <20170915115819.GA22099@stack.nl> References: <201709050746.v857kjX9032212@repo.freebsd.org> <20170915115819.GA22099@stack.nl>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi Jilles, 2017-09-15 12:58 GMT+01:00 Jilles Tjoelker <jilles@stack.nl>: > Although this is correct from a functionality point of view (S_IFIFO and > S_IFSOCK are probably expected to be different in a POSIX context, but > this is unlikely to break anything that is not already broken), it is a > partial reversal of previous changes to FreeBSD that created separate > implementations for pipes (first unnamed pipes in 1996, later fifos in > 2012 r232055). The main reason for these changes was performance. > > Unfortunately, I do not have concrete benchmarks in the CloudABI > context. That is important to keep in mind indeed. Thanks for pointing this out! When I discussed this with some of the other folks working on CloudABI, we came to the conclusion that this (potential) loss of performance is acceptable in the nearby future. If it would really affect us negatively, it should in theory be possible to reimplement our streaming sockets and pipes with something as simple and fast as FreeBSD's pipes, but still support file descriptor passing, etc. -- Ed Schouten <ed@nuxi.nl> Nuxi, 's-Hertogenbosch, the Netherlands
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CABh_MKnXZiLs5jv8p=zW_4mSCiqL7f5m-kuGk7XU==a_Q3GY9g>