Hi, Céline,
Indeed, not difficult, but not quite so straightforward as using variables: the *.tpd/*.target files don’t support variable substitution for this purpose, but it was easy enough to use Ant to automate a transformation of the *.target files. Also, it was easy enough to have the releng POM detect automatically whether it the build is running on the Papyrus-RT HIPP or some other environment, using profiles to choose between the Eclipse-local or portable variant of the target. This does turn out to have drastically improved build times when HTTP responses from download.eclipse.org were slow, so that’s good.
Christian
On 5 October, 2016 at 11:03:08, Céline JANSSENS (celine.janssens@xxxxxxxxxxx) wrote:
+1.
That can be done into the TP files with a specific property like in Papyrus : transfer.protocol=”file” or “http”
By default it would be “http”, and in the jobs, we can override it to “file”.
It seems not too difficult to put in place.
Céline
Now it looks like we just have some kind of severe performance problem in accessing
download.eclipse.org via HTTP: fetching the metadata from p2 repositories hosted there is extremely slow. I’m working on generation of targets localized for use on
Eclipse.org that access the downloads via the local filesystem, as is done in Papyrus. The build would automatically detect when it is running in the HIPP on
Eclipse.org to activate that option.
On 5 October, 2016 at 08:47:30, Christian Damus (give.a.damus@xxxxxxxxx) wrote:
It appears that we now have Gerrit builds failing on a time-out error accessing the latest successful build results of the Papyrus Developer component:
I don’t think that anything has changed in the Papyrus builds that could cause this. Is it possible that today's plug-in installations and/or
restart of the Papyrus-RT HIPP could have broken HTTP connections to the other HIPPs at
eclipse.org?
And I don’t suppose it will be just the Gerrit builds that fail, but the main component and product builds, too, because they all use the same
target-platform files. Papyrus has two forms of every TP, one using HTTP access and one using local disk access for use on the Hudson build servers. Should we be doing the same? I have no trouble accessing the required resources in the Papyrus Hudson instance
from home; it’s only our builds that are timing out.
This would be another thing to account for in the single-sourcing of TPs for the build and Oomph setups for the developers.
_______________________________________________
papyrus-rt-dev mailing list
papyrus-rt-dev@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit
https://dev.eclipse.org/mailman/listinfo/papyrus-rt-dev
|