On 24 April 2013 16:50, Camillo Bruni <camillobruni@gmail.com> wrote:
I checked the issue under 3.0 with the configuration loading speed.
Main Performance Issues: ------------------------ Gofer: package-cache always used by Gofer => big shared package-cache is very slow RPackageSet: each method/class change trigger the creation of a new RPackageSet for EACH Monticello working copy
What was reported is, that disabling the announcements during configuration loading decreases load time. This is due to the RPackageSet problem. No announcements, no new RPackageSets.
For a typical configuration loading CommandShell and Soup + 5 packages I got 518'124 new instances of RPackageSet!!
Solutions --------- Gofer: removing the package-cache from the default repositories seems like a good thing to me. There are 3 tests failing after that, mostly checking if there is a default repository or not. I would add an explicit statement for the package-cache otherwise.
RPackageSet: With a basic cache for RPackageSets I cut the load time to a third! It will need some more love when it comes to changing classes, but for now I invalidate it each time an RPackage changes.
I got one more: - fix model of RPackage to contain same objects as MC package. As a consequence RPackageSets goes to > dev/null
-- Best regards, Igor Stasenko.