[OpenWrt-Devel] Proposal: change buildbot's build failure logic to be less restrictive regarding packages from feeds

[Hannu Nyman]

We have seen several times in the past few months that one non-essential 
package from feeds fails to build and stops the whole buildbot build. 
Yesterday it was python-cffi, but it has been ola, ltq-vmmc etc.

It seems strange to me that non-essential packages from feeds, which by 
definition are non-essential, can kill the whole buildbot build for a platform.

Normally a failing package gets marked "broken", but the build still 
continues. However, some packages seem to kill the whole build when they fail.

As far as I have been able to determine, if the failing package contains 
either a kernel module (kmod) or a "host section", then the failure of that 
package is considered so essential that the whole builbot build run is 
halted. The package is not just classified as "broken package", but the whole 
build is really stopped.

I understand that a host section breakage from a tool, toolchain component or 
a core package in the main Openwrt repo may be reason enough to halt the 
build. Same goes for a kmod needed for the device to operate.

But when that failure happens to a package from e.g. telephony feed or a 
python subpackage, I see no reason to stop the whole buildbot build run. The 
more we have packages maintained in Github by a large group of people, the 
more we will have this kind of accidental breakages by small non-essential 
packages.

I propose that devs would change the buildbot build failure logic so that the 
failing packages from the feeds would not kill the whole build. Failing 
packages from feeds should just be marked as broken. That might require 
splitting "package compilation" step to two different steps, one for packages 
from the Openwrt main repo, one for packages from feeds. (or alternatively, 
the essential packages should contain an ESSENTIAL definition on the 
Makefile, or something like that, and the failure logic should take that into 
account.)

This is a slightly reformatted version of my previous message about the same 
subject in January 2015.
The topic is still important, so hopefully developers could think about it.
https://lists.openwrt.org/pipermail/openwrt-devel/2015-January/030719.html
_______________________________________________
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel