Trevor -

My organization is facing this exact same concern. We are primarily an application platform / hosted-solution company, meaning we don't "shrinkwrap" and support multiple versions of our systems - only one version is in production at a time. Of course, multiple branches are undergoing testing in parallel, each with different release timeframes, etc, and normal branch merging / management occurs at the SCM level. Separate systems, however, have their own release schedules.

SystemA = Dev team A
 -War AWar1, depends on AJar1
 -War AWar2
 -Jar AJar1, depends on SSJar1

SharedStuff
 - Jar SSJar1

SystemB = Dev team B
 - Jar Bjar1, depends on SSJar1
 - Jar Bjar2, depends on AJar1 **note: cross-system dependency
 - War Bwar1
 - Ear Bear1

etc. Organizationally speaking, we have dev teams responsible for one system. Shared stuff is just that, shared - no particular team as owners / maintainers.

Our current solution we're incubating is based around a "floating snapshot" concept. We use the version "current-SNAPSHOT" in all our trunk POMs. So far, we like it. I reiterate the fact that we don't really care about version numbers - once a branch goes to production, the previous branch is retired. Period. And we want to keep our systems relatively up-to-date in the in-house code they share.

- Our CI server (CruiseControl at present) runs a 30 minute build loop. All projects - jars and wars alike - are mavenized (m2). Souce retrieved is tip/trunk at all times, to ensure trunk integrity. - Our "build server" runs builds on-demand, and retrieves only "promoted" source from our SCM. We think of these as release candidate builds, and we create them often.

The ant scripting surrounding our mvn invocations basically mimics the release plugin. The scripts run by our "build server" check out the pom, inserts the CC build number, checks in a pom revision in SCM. The CI build scripting checks back out and re-inserts "current- SNAPSHOT" into the tip revision when triggered.

The result is that whenever CI builds run, they build trunk of everything against trunk (current-SNAPSHOT) builds of all dependencies, creating an early warning system. Whenever "build server" builds run, the packaging of resultant builds include the latest "blessed" (promoted) builds of in-house code. If large amounts of time go between RC builds, then that project's dependencies can get "out of date" so to speak. We're still struggling with that to be honest.

I welcome your thoughts. I think this is a discussion that is central to large-scale adoption of maven in the enterprise. If we need to take this off-list I am fine with that as well.

-Scott Heaberlin
Columbia, SC


On Mar 25, 2009, at 3:10 PM, Trevor Harmon wrote:

Consider this scenario:

Alice and Bob are working independently on two different applications, AppA and AppB. Both applications depend on an in-house shared library, Foo, that Alice and Bob are working on together. They have both checked out Foo's trunk and are regularly committing changes to it.

Because Foo is undergoing heavy development, AppA and AppB depend on Foo 2.1-SNAPSHOT, but now Foo is looking pretty stable, and Alice's AppA needs some of the features scheduled for Foo 2.2, so she decides to perform a release of Foo 2.1 and does the usual release procedure:

1) Changes Foo's version from 2.1-SNAPSHOT to 2.1 and checks it into the trunk
2) Deploys Foo 2.1 to the company's internal repository
3) Tags the Foo trunk as the 2.1 release branch
4) Changes Foo's version from 2.1 to 2.2-SNAPSHOT and checks it into the trunk
5) Changes AppA's dependency to point to Foo 2.2-SNAPSHOT

But what about Bob? He's still working with Foo 2.1-SNAPSHOT for his AppB. If he updates his working copy of Foo's source code, any changes he makes to Foo will be built as a 2.2-SNAPSHOT release, since Foo's trunk is now 2.2-SNAPSHOT. This is a major problem because his AppB has a dependency on 2.1-SNAPSHOT, so the next time he tests AppB, it will pick up the old Foo 2.1-SNAPSHOT, ignoring any changes Bob makes in Foo. He will probably waste a lot of time debugging, at least until he happens to notice that Foo's version has changed.

What can be done to prevent Bob's problem?

Trevor


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]



---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to