Agree about mangos being better. I use it quite often and it never
deadlocks. Now that Go can produce shared libraries that look just like C
shared libraries (a Go 1.5 feature, released in August), I'll probably just
use a mangos based library instead of the C for R stuff.
On Thu, Sep 10, 2015 at 12:38 PM, Garrett D'Amore <garrett.damore@xxxxxxxxx>
wrote:
Sorry for the delayed response, this isn't my preferred email account, and
things got a bit backlogged here.
I think some of the things may have been fixed in nanomsg 0.6, but not
all. I still don't have a really good feeling for the stability of it --
yes, I think mangos is vastly better in this regard.
On Sat, Aug 1, 2015 at 11:03 AM, Jason E. Aten <j.e.aten@xxxxxxxxx> wrote:
Hi Garret,
Congratulations on the 0.6 release!
Would you happen to know if this 0.6 version of C-nanomsg release fixes
the server deadlocks? They were intermittent and somewhat difficult to
reproduce (easier with 64 cores), but I was seeing them at least on the
PULL sockets last year.
The side effect of those hangs was that they led me to substitute mangos
which is nicer anyway being all Go.
I don't want to go back now, I'm still a huge mangos fan, but for my
nanomsg R bindings (https://github.com/glycerine/nanomsgardvark) I was
just curious about whether I could use depend now on C-nanomsg to be a
communication substrate in R.
Best,
Jason
On Wed, Jul 22, 2015 at 12:34 PM, Garrett D'Amore <garrett@xxxxxxxxxx>
wrote:
All,
I’ve released (for real this time!) nanomsg-0.6-beta. This release has
quite a few improvements since 0.5, and various fixes, so I highly
recommend updating to it. Notably, the web sockets support was integrated
post-0.5, and there is a wire protocol change for surveyor that makes it
possible for a respondent to peer with multiple surveyors. (And, btw, also
makes it possible to create devices that behave properly with surveyor.)
As of 0.6, the download page has gotten some changes. Notably, I’m
using github releases for actually releasing content, which gives an HTTPS
secure download, which should increase confidence in the content. I’m
still providing MD5 and SHA1 on the download page. The legacy versions are
still served from the download.nanomsg.org site, but new versions
starting with 0.6 are no longer posted there. Feel free to go ahead and
download from the github releases page going forward though.
Also, the build includes continuous integration and deployment provided
by Travis CI and AppVeyor. Again, this should increase overall confidence
in the project and hopefully minimize the likelihood of regressions.
The legacy waterfall testing is still present, but as it is provided
entirely by volunteer resources, I felt it better to rely on services
offered by a commercial concern that focuses on CI as a core delivery
product. This also lets me get the automated deployment integrated, which
is a huge benefit.
Hopefully releases of future versions will have fewer hiccups and we can
resume a more timely release train.
- Garrett
