Skip to main content
Jump to: navigation, search

WTP Build Process and Procedures

Revision as of 00:32, 27 May 2006 by David (Talk | contribs) (Updated to current "out" file ... plus, have learned -F better than -f for what we are doing with it.)

This page is to collect general information and references about the WTP build process and procedures. The idea is that any WTP committer can update it, pretty much "on the fly" or "as you go", or "as needed", so that there will often be a bit of a disorganized, stream of consciousness flavor to it. Hopefully, occasionally, some kind-hearted committer will stop and organize all the miscellaneous notes and tips and references that are provided, into fun-to-read treasure that you just can't put down.

Note: since only committers can edit these pages, if anyone from the community has contributions or suggestions for additions, please open a [feature request] on our releng component.

Thanks, David Williams 01:42, 2 February 2006 (EST)

Build Schedules

As of February 1, 2006

Our WTP 1.0.1 contributions are due by EOD on Monday's, we smoke test that build on Tuesday's, and plan to declare it by Wednesday's at noon (eastern time).

Our WTP 1.5 contributions are due by EOD Wednesday's, we smoke test that build on Thursday's, and plan to declare it by Friday's at noon (eastern time).

Build mechanics

The platform's releng tool should be used to "release" projects to the map files. Be sure to have the appropriate branch of releng project (which has the map files) loaded in your workspace. R1_0_maintenance for 1.0.x builds, and HEAD for 1.5 builds.

As projects are versioned, please use the "standard" format, in UTC time, following vYYYYMMDDHHMM. This is important as these cvs tags become the qualifier field of the plugin's version.

Note: do NOT use underscores in the CVS version, as there are some issues with Eclispe tooling when it finds an underscore in a plugin or version qualifier, see [89428].

How to watch the builds

Do you enjoy watching screens of console logs fly by as much as I? Waiting for that joyous final result? Well then you too can do that with the following information.

We are currently doing our builds on Any committer can logon to that server using their committer ID and password. I use something like


Once there the main work we do is in /shared/webtools/. If you want to "watch" the console log while the compiiler and ant files are running, you can use

   tail -F /shared/webtools/releng.control/out.txt -n200

to watch the last part of the main log (during compile, etc.). This is the fastest way to find out if the build failed. Once it does fail, there's some diagnostic messages that might be informative in the last 20 or 50 lines. (Do be careful, though, the build always says "Build Successful" as its very very last message, and that message just means all the ant scripts were any to completely execute (not that they did so successfully).

During tests, you can watch test results console at,

   tail -F /shared/webtools/test-wtp-S/results/consolelogs/wtptestlog.txt -n200

In fact, I do this so much, I define these aliases in my .bash_proflie

   alias tailout='tail -F /shared/webtools/releng.control/out.txt -n200'
   alias tailtest-i='tail -F /shared/webtools/test-wtp-I/results/consolelogs/wtptestlog.txt -n200'
   alias tailtest-m='tail -F /shared/webtools/test-wtp-M/results/consolelogs/wtptestlog.txt -n200'
   alias tailtest-r='tail -F /shared/webtools/test-wtp-R/results/consolelogs/wtptestlog.txt -n200'
   alias tailtest-s='tail -F /shared/webtools/test-wtp-S/results/consolelogs/wtptestlog.txt -n200'

Here the handy thing to watch for is if a test doesn't seem to make any progress at all, after 10-20 minutes. That's a sure sign of a hang. And, here's how to get some diagnotic information on Linux. First, "break out" of the tail program with ctrl-c. Then get a list of processes that are running by using

    ps -ef | grep java

Usually, torwards the end, you'll see a task that mentions, in the long (wrapped) command line, a recognizable JUnit suite. For that process, looking toward the left of the long (wrapped) line, get its process ID ... usually 4 or 5 digits. Then (almost there!) type "kill -3 <process-id>". I forget exactly what the '-3' means to Unix, but to Java, it means to take a dump of current data and thread states, You'll see a few messages to the console that a java dump wsa greated. That's the file that's handy to attach to a bugzills. (use sftp copy, if you do not have handier tools). That file is usually located in the /eclipse directory in use during the test.

Feel free to edit this document with further tips (or, improve the directions above is I left out some steps, or, you've noticed things are slightly different than written here.

Chronological Notes

This section is just to make quick notes about changes to our build files/scripts that others might find intersting.

Update prereqs to include official Eclipse M5


There was an "odd" note when I downloaded the PPC version ...

Eclipse fails to create a lock file with reason "No locks available."  To launch eclipse you must disable file
locking using the osgi.locking property. For example, you could launch eclipse as follows:

"eclipse -vmargs -Dosgi.locking=none".

Thus, we *might* have to -Dosgi.locking=none on our PPC machine when we move to this M5 version of the base builder.

-- David williams 18:04, 21 February 2006 (EST)

5/23/2006 Changes to build machine and CruiseControl

I've moved the


stuff to another cvs project ... and called it


The idea is that anything that is "machine specific" or "cruise control" specific we can put in releng.control, and have it under its own versioning/update rules. We'll attempt to keep the CVS version in synch with the PPC machine we use, but other machines can be kept in "parallel" files, e.g. see 

and rename them when deployed.

One implication (by design) is that "releng.builder" isn't downloaded and checked automatically by CruiseControl, in other words, changes in it won't trigger a build. Not sure what is still used in "releng.builder", but know some things are, such as API checkers. So, these API checkers will have to be downloaded or triggered by some other means (as has been our direction for a while).

Similar tor the "whatisfixed" tools .. I've turned it "off" by default ... it takes almost as long as the build itself! So, we don't want to run it each build. Naci will be working on how to execute those separately against a specific, declared build.

I've updated to cruise control 2.5 (they fixed that memory leak that was causing me to have to restart it every day or two). (initially the 2.5 version was "crashing" frequently, but I think related to VM version, and seems to be running well now, well ... for at least for 24 hours :)

I got the status is working again ... see

David 21:02, 23 May 2006 (EDT)

drop/drivers moved to arhives/drops/drivers

To avoid mirroring our relatively temporary "drivers" directory, I've completed the move off ...webtools/downloads/drivers to /home/data/httpd/

David 01:27, 27 May 2006 (EDT)

Background and Further Reading References

We in WTP following the basic process and recommendations for versioning as the base Eclipse platform. Plugin Versioning

Very helpful guide to builds and automatic testing. Build and Test Automation for plug-ins and features

Good step-by-step on how to do updates. How To Keep Up To Date

We base our builds on the Eclipse platform's "basebuilder". Platform-releng

With our WTP specific needs addressed in the WTP project called releng.wtpbuilder. While slightly out of date, see the [attachment] to [bug 108259]

Our basic server configuration and cruise control triggers is handled by the WTP project called releng.builder (see especially the tools/cruise directory in that project).

And ... never forget Eclipse Help ... search for things related to update manager, PDE, features, site.xml, etc.

Back to the top