dev@glassfish.java.net

Re: [Fwd: GlassFish build failure]

From: Bill Shannon <bill.shannon_at_sun.com>
Date: Tue, 03 Apr 2007 16:10:02 -0700

Terena Chinn-Fujii wrote:
> Bill Shannon wrote:
>> Terena Chinn-Fujii wrote:
>>> The RE builds updates the image with snapshot pointing to the
>>> latest nightly build. However, it's possible that your 3:00am build
>>> was still picking up the previous night's build since the trunk's
>>> snapshots generally don't get updated until 4:00am-8:30am depending
>>> on the platform being built (and the network traffic going on).
>>
>> So there's a window between when someone checks in a change of
>> this sort and 8:30am the next day during which GlassFish might
>> not be buildable?
>
> The snapshots don't get updated if the build is broken or if
> the tests are failing so it's always pointing to a build that should
> be buildable.

The problems I'm running into seem more like a lack of synchronization
between two different parts of the build.

If I change an API in one module, and change another module to depend
on that API change, I need to build the first module before I can build
the second module. But which module to build first depends on which
modules I change and how I change them.

If the build of the second module depended on an existing build of the
first module, it may or may not work depending on whether the existing
build uses the new API or the old API. If the existing build is updated
every night, then this problem corrects itself every day. But in the
mean time it's impossible to build reliably.

> RE nightly builds starts at 11:59pm PDT with cvs checking out all the
> source files which takes about an hour or so. If someone checks in
> something during that time, the build may/may not pick it up
> depending on where the build is with checking out the modules.

Sure would be nice if these operations were atomic. :-)