Hello,
There is a discussion about CI contract for Eclipse Platform
repositories started at
https://git.eclipse.org/r/#/c/145336/
The situation is the following:
1) there is a new bundle to be contributed to eclipse.platform.text
repository
2) The CI Bot fails with
02:42:29 [ERROR] Failed to execute goal org.eclipse.tycho.extras:tycho-eclipserun-plugin:1.4.0:eclipse-run (api-analysis) on project org.eclipse.text.quicksearch: Execution api-analysis of goal org.eclipse.tycho.extras:tycho-eclipserun-plugin:1.4.0:eclipse-run failed: Cannot resolve dependencies of null: See log for details -> [Help 1]
3) The solution suggested is
"... we need both change (merge here + addition in feature) to
happen in the same time, and then a build with both of them. There
will be some red Gerrit builds unavoidably.
The thing is that we don't really need a org.eclipse.search feature,
what would it be good for in term of deployment? The current state
is relatively good as it maps usage of the features in the IDE or
RCP products. Changing the deployment structure to better map the
build issues just puts the bill of build-time complexity on the
runtime, and that's not really something we should do."
I understand that for today it may be the only option to proceed.
But I have generic concern regarding this approach as it prevents to
test changes with CI Bot *before* merging to master.
IMO any repository that has its own build configuration should
support an ability to test the changes inside it with CI Bot, i.e.
*before* merging to master.
Otherwise it should not have its own build configuration and the
"parent" repository should take care about gerrit change for the
child.
Let's continue discussion here.
Regards,
AF
