As the person who is probably responsible for breaking the most things in HPS Java, I thought I’d share some thoughts about how not to break things going forward.

1) Always run the tests before big checkins

There are integration tests in HPS Java, and at this point some of them actually do useful things, like asserting that a certain number of reconstruction objects must be present in output.  Before making major changes run the build from trunk with the ‘mvn’ command so that all tests are activated.  This will make sure your changes do not fundamentally break anything.

2) Keep an eye on the Hudson builds

When the Hudson builds fail, this usually means that a compilation error occurred, which can mean you have introduced build errors into the project.  (Sometimes Hudson has strange errors having to do with filesystem issues but this is usually easily spotted).  

An “unstable” build means that the build worked but one or more tests failed.  

For instance, you can see the list of the latest test results here.

If you have checked in new code and the tests all of a sudden start failing, it is worth investigating why, as your code may have broken something.

3) Make sure you have checked in all your local, required changes

Sometimes you might checkin code and forget some piece of it, like a file that was never added using ‘svn add’ or a modification to some class that is required with your other changes.  These types of problems are usually easily spotted if your build is working fine but Hudson is failing.  Use ‘svn status’ to tell which files you have locally modified which are not in sync with SVN and then commit the missing changes to fix this.

4) Write test cases

It is worth writing test cases for your code and having them run with the build.  I think I have harped on this enough to not need to explain why.

We are particularly deficient in this area when it comes to converting from the raw EVIO data to LCIO, so we should think about adding some in-depth tests in this area soon, as breakage in the event builders can essentially make the entire reconstruction non-functional.

5) Be careful with changes to LCSim

There are some low level classes in LCSim which when changed can cause fundamental problems in our software.  In particular, any modifications to the base implementations of the reconstruction classes like Track or Cluster, major alterations to detector model code, etc. can cause issues in any classes that depend on them.     Aside from new detector model code, which is sometimes required, any major modifications to this project should be discussed first.

6) Branch the code

The way that SVN works, you cannot stage a commit like with git.  So it is possible to make a series of commits, one of which breaks the build, but altogether they would not.  The proper way to get around this is to branch the code, either the entire trunk or individual modules, and then make incremental commits on those branches where you can leave things broken and not affect anyone else’s copy of the trunk.  Then, when the project is completed on the branch, you can merge it back in.

For instance, you can branch the entire trunk as follows….

svn cp -m “Branching HPS Java for fun and profit.” svn:// svn://

It is a good idea to assign a JIRA item to the project you’re working on and use it for the branch name.

Then when you’re done checking in changes on the branch, you can merge them back into trunk.

cd trunk; svn merge svn:// .

You need to then ‘svn commit’ those changes as merge does not commit them.  It is best to actually check which modifications were made from the merge locally before committing.

Or you can branch a single module…

svn cp -m “Branching the tracking to rewrite it in Ruby.” svn:// svn://

And merge it back similarly…

cd trunk/tracking; svn merge svn:// .

For projects isolated to a single module, it is best to only branch that module.  For anything that touches code in many modules, then you can branch the entire trunk.

If you have any questions about branching and merging, then don’t be afraid to ask.  This is a powerful tool for having a copy of the code that you can work on without introducing unwanted changes into the main “trunk”.

7) Don’t break the prod branch

We have just introduced a “prod” branch where changes will be vetted in trunk before being introduced.  This is for introducing some level of stability to the code running in the counting house.  That said, there is nothing special about this branch that will automatically prevent us from breaking the code on it, so be extra careful about which changes are merged into it going forward, and keep in mind some of the points mentioned above.

If there are any other suggestions here for improving things overall in this area, please let me know!

Use REPLY-ALL to reply to list

To unsubscribe from the HPS-SOFTWARE list, click the following link: