Dear all,
Holger and I (plus later the sysmocom team) had a discussion about whether we should continue to keep the FreeBSD build testing.
Right now, jenkins is set up in a way to test build of a patch on Linux (currently Debian 8) and FreeBSD (currently 10.3). Only if it builds and passes "make check", the patch gets the magical "+V" so it can be merged.
While in general it's of course good practise to write portable code and to test building on multiple operating systems, this adds quite a bit of extra effort: * patches every so often don't build on FreeBSD due to include header differences or the like, requiring extra iterations of a patch * the build takes longer time as the build matrix is multiplied by two operating systems * we cannot easily/safely migrate the way we buld to docker, as docker on FreeBSD is experimental
Particularly during my recent OpenGGSN developments (IPv6) support, this was a big PITA and I'm sure I spent more time in debugging FreeBSD build issues than actually writing the code in the first place.
Given that we arr not aware of anyone using the Osmocom stack on FreeBSD, the suggestion is thus to focus our available time on actual GSM related features rather than fixing up portability issues.
So unless we receive significant complains as a response to this e-mail, we will disable the automatic build testing for patches on FreeBSD. This doesn't mean we will drop related code from our projects. We still appreciate build and other fixes for other platforms, but those would have to come as contributions from users on those platforms, rather than something that's actively maintained and tested by Osmocom upstream.
If we want to extend build testing, I think it's more useful to have e.g. ARM/Linux builds, or clang builds on Linux than the FreeBSD builds.
Let me know what you think.
Regards, Harald
Me personally, I once spent a few minutes extra on a FreeBSD build, a long time ago.
On the other hand, most of the time my patches aren't tested on FreeBSD: particularly openbsc.git as well as the new repositories haven't been tested (for gerrit +V) on FreeBSD for a long time.
So FreeBSD has no significance for my daily work at all and I probably wouldn't notice that it is gone.
Given involvement by some friends of mine I'd actually be more interested in OpenBSD than FreeBSD ... but that's beside the point :)
~N
Hi Neels,
On Tue, Sep 05, 2017 at 11:42:36PM +0200, Neels Hofmeyr wrote:
Me personally, I once spent a few minutes extra on a FreeBSD build, a long time ago.
lucky you.
On the other hand, most of the time my patches aren't tested on FreeBSD: particularly openbsc.git as well as the new repositories haven't been tested (for gerrit +V) on FreeBSD for a long time.
Ah, ok, I missed that.
Given involvement by some friends of mine I'd actually be more interested in OpenBSD than FreeBSD ... but that's beside the point :)
Well, as indicated we'd be more than happy to merge related patches, but I just have doubts whether it's efficient use of our time to actively (build) test the code on other platforms.
Also, if somebody wants to maintain this, we could keep a separate e.g. nightly jenkins job around and then generate build failure notifications to whoever in the community who wants to maintain the (sup)port for other platforms.
Regards, Harald
Hi!
Since there were not any FreeBSD users asking us to keep the FreeBSD builds, I think we should go ahead with removing the build testing on that platform.
For the gerrit jobs it's easy, my proposed patch can be found in https://gerrit.osmocom.org/#/c/4576/
Let's wait if the jenkins-job-builder gerrit jobs work as expected before merging/deploying this.
Regards, Harald
-
Harald Welte -
Neels Hofmeyr