The attributes that were used previously are not set, gotta use those from travis_build_environment
Suppress PostgreSQL client warning messages in build output
Makes it possible to provision nodejs 0.4.x again
Even though the next versions of Debian, Ubuntu and Fedora ship with OpenJDK 7 by default, and Oracle JDK being based on OpenJDK (so, all new work goes into OpenJDK first), our OpenJDK 7 package is very old (by recent JDK development standards) and ocassionally segfaults with JRuby and Scala. This is a 32-bit problem so our options are: * Move CI env to 64 bit * Move to 12.04 earlier than expected * Upgrade OpenJDK 7 to 7u5 Each of these has problems: * We will likely have to upgrade some hardware to make sure we have the same amount of memory free for I/O buffers, since 64 bit VMs will use more RAM. * There are no 12.04 packages for a number of projects (like PyPy) and we cannot quickly build and maintain them well on our own. * Building OpenJDK 7 is not that hard but takes ~20 minutes on a powerful machine and *packaging* it is a whole different story. We probably won't be able to do it and maintain it well on our own either. So we just switch the default: our Oracle JDK 7 is 7u5 (a recent patch release) and with 7u6, 7u8 and so on it will be pretty easy for us to upgrade it thanks to the PPA we use which downloads packages from the Oracle Technology Network mirrors. We are still committed to helping OpenJDK 7 & 8, but for now we cannot do much about the situation and a non-trivial portion of our users is affected.
…e docs) It was not intentional. Reported by a neo4django developer on the neo4j mailing list.
We are trying it out to see if it helps with OpenJDK 6 and 7 segfaults that ocassionaly happen on JRuby :( It is something very much 32-bit OpenJDK and JRuby specific: Clojure, Java, Groovy and most of Scala projects on travis never cause segfaults and 64 bit JVMs don't seem to segfault either. Per discussion with @rkh.
eth0 with the currently used NIC (virtual network card) sometimes goes down according to dmesg and syslog. The reason for that seems to be that NICs are overwhelmed by a huge number of outgoing network connections that tools like pip and Maven may open simultaneously. ifup eth0 is idempotent so there is no downside to trying this. This is a yet another step towards eliminating stalling SSH connections. Some of the work has to be done on the net/ssh side, too.
…on in #72
This reverts commit 0f81b5d. Execution of that file fails for multiple reasons. One of them (php is not available without phpenv) needs to be addressed fundamentally.
Introduce global Composer for php-build's
To help investigate stalling SSH connections and sshd terminating with code 255
…rovisioning 3 times out of 4" This reverts commit c246ca5. Lets give it another shot with mirroring.
… downloaded archives
This reverts commit 783c8dc. The issue was not with kerl but with the fact that we do not clean up build artifacts and eventually run out of disk space on the / partition.