[tei-council] Commits, builds, errors and Jenkins

Martin Holmes mholmes at uvic.ca
Tue Apr 19 20:59:22 EDT 2011

> When Jenkins is already engaged in doing a build, it will not start
> another one until the current one is finished.

No sooner did I send that than I saw Jenkins start a second build while 
one was already running. But I think it holds true for two builds: it 
doesn't seem to run more than two (as far as I can see, so far...). 
Presumably the number of concurrent builds is configurable.


On 11-04-19 03:23 PM, Martin Holmes wrote:
> Hi all,
> I made my first few commits today, and I thought I'd report my
> experiences in case they help anyone else who's intending to do this but
> hasn't tried yet. This is all on Ubuntu Linux.
> First I followed these instructions to install the TEI packages for
> Debian on my machine:
> <http://wiki.tei-c.org/index.php/TEIDebian>
> I made a directory and checked out the full source my local machine:
> mkdir sf_repo
> svn co https://tei.svn.sourceforge.net/svnroot/tei sv_repo
> Then I moved to the P5 directory and ran a build using "make":
> cd sf_repo/trunk/P5
> make
> This took a long while the first time, and showed me lots of messages
> that turned out to be a mixture of the incomprehensible, the ignorable,
> and the relevant. Relevant messages (for me) are ones that I can imagine
> a way to fix; or that were caused by one of my changes (this hasn't
> happened yet).
> I soon realized that watching the messages go by in my terminal was not
> really useful, so I started saving the error output to a separate file
> incorporating a datetime stamp in its name every time I ran a build.
> This is how I'm working now:
> #Go to the root directory of the SVN tree
> cd sf_repo
> #Check out the latest changes
> svn up
> #Go into the P5 directory
> cd trunk/P5
> #Run a preliminary build
> make 2>  /path/to/error/file/make_errors_`date +"%Y-%m-%dT%H:%M:%S"`.txt
> #Make my own changes...
> #then run another build
> make 2>  /path/to/error/file/make_errors_`date +"%Y-%m-%dT%H:%M:%S"`.txt
> #Finally, diff the two latest error files to see if I've fixed
> #any previous errors, or introduced any new ones. For this I
> #use Meld.
> #Commit my changes if all is OK
> cd Source/Specs #(for instance, assuming that's where I edited a file)
> svn commit -m "Message explaining my changes."
> #Go to Jenkins and watch it build the changed source,
> #then check for errors again.
> Watching Jenkins today answered a number of questions for me about how
> it works, especially with regard to queueing up builds:
> When Jenkins is already engaged in doing a build, it will not start
> another one until the current one is finished. When the current build
> does finish, it will initiate a new build which incorporates ALL the
> changes which have been committed during the time it was engaged in the
> previous build. What this means is that some revisions are never built
> independently. For instance, I committed two changes to SVN while a
> build was in progress (revisions 8816 and 8817); when Jenkins was free,
> it commenced a build of the documentation that incorporated both those
> changes, meaning that no independent build of revision 8816 was created.
> Only revision 8817 was built.
> What this means in practice is that, if there are many of us making
> changes and commits, we will only see the results of our commits (in
> terms of errors and warnings) mixed in with those from other commits
> which have happened around the same time. That will make it slightly
> more difficult to tell whether you caused an error or not, and how you
> might fix it. So I think it's probably even more important to have a
> local build environment you can use to test your changes before
> committing them. While errors can easily be undone by means of
> reverting, there is a danger that in order to do this, you might have to
> revert past other people's changes that were committed just after your own.
> I put all the above into a blog post here in case it's useful:
> <http://hcmc.uvic.ca/blogs/index.php?blog=15&p=8102>
> Cheers,
> Martin

More information about the tei-council mailing list