Le 16 janv. 09 à 23:07, Mitch Gitman a écrit :
Nicolas, correct me if I'm misstating something, but I look at an
ivy:retrieve-like capability as a nice-to-have but not necessary
feature for
IvyDE.
IvyDE itself really just exists as a nice convenience, so that
classpaths
resolve correctly in Eclipse. How it does that--whether with
ivy:cachepath
or with an ivy:retrieve to a specified directory--is somewhat of an
implementation detail in this respect. Either way, the
ClassNotFoundExceptions go away.
The tool of record is still the build script itself. Once you have
IvyDE
doing Ivy retrieves to the lib directory that the build script is
doing
retrieves to, it kind of feels to me like you're letting the IDE (the
second-class citizen) encroach into the realm of the build script (the
first-class citizen).
I agree with you the reference should be the ant build system. For
instance eclipse can only manage one classpath by project, whereas we
usually use it to compile, run and test. So in my projects I ask IvyDE
to get 3 configurations: compile, runtime and test. But in ant the
javac task use an ivy cachepath with only the compile configuration,
for the junit task only the test configuration and for the building of
some war only the runtime configuration.
Eclipse setup should be as close as possible to the build and
deployment environment, but it is not the case.
That concern aside, suppose the developer has blown away their Ivy
cache. I
don't see why the issue Aaron describes cannot be addressed simply by
right-clicking on the Ivy icon in Package Explorer and selecting
Resolve.
Granted, it takes a little getting used to for developers to get in
the
habit of this, but this seems like part of the contract of using
Ivy. As an
Ivy consumer, you need to have some understanding as to when a
resolve is
necessary.
Well, I don't know what size is Aaron's project, how theses projects
are interconnected. But If you have 20 projects all relying on some
maven repository, it will really painful to do a resolve on every
project.
Maybe the real debatable question here is why the cache need to be
cleaned, normally it shouldn't. The cache is not aimed to modify how
the resolve process occurs, just speed it up. Actually I see only two
reasons that should stay exceptional:
* the size of the cache
* a repository has been setup with checkmodified=false but a broken
ivy.xml on that repository has been fixed
Nicolas
On Fri, Jan 16, 2009 at 1:40 PM, Nicolas Lalevée <[email protected]
wrote:
Le 16 janv. 09 à 06:25, Yu, Aaron (IT) a écrit :
Hi,
I have a problem of using Ivy and IvyDE.
At my firm we are in the process of switching to Ivy for managing
our
java dependencies. We do not want thousands of developers to resolve
dependencies in different ways so we provide a centralized ,
read-only
ivy-settings.xml file which is used by all developers. However,
this
presents us with a problem. We would like to make it possible for
our
users to set their own cache locations, because if they use the
same
cache location for all projects, when they clean the cache in one
project, the whole cache location will be cleaned, and this will
crash
the building of other projects.
I know we can create a project-specific ivy-settings.xml file, and
use
'defaultCacheDir', 'resolutionCacheDir' and 'repositoryCacheDir'
attributes in 'caches' element to set the project-based cache
location.
But as I mentioned above, to make all the projects more manageable
and
shareable, we do not encourage developers to have the project-
specific
ivy setting file.
To get around this we have implemented a new properties page for the
ivyde plugin under the existing one called simply 'Cache'. This
allows a
user to choose the cache location. The choices are: Default
($HOME/.ivy2) and Workspace (workspace_dir/.metadata/.ivy2). We then
reset the ivy.home System property and trigger re-resolution of all
project ivy classpath containers when the users changes this setting
(the change affects all projects in the workspace since ivy core
reads
the ivy.home System property to find out where it should do its
caching). However, we are now wondering if what is really needed
is a
private cache area for each project in the workspace, for example:
workspace_dir/.metadata/.ivy2/project_name. But this is not
possible at
the moment due to the way in which ivy core decided where to place
its
cache. (And, as we said, we do not want users to have to specify
their
own ivy-settings for each project for this reason alone.)
Other people at our firm have also pointed out that it might be
better
if ivy core used a version specific cache location by default since
there could be differences in cache metadata between versions. This
would cause cache corruption if different versions of ivy core
were run
on the same system.
All this seems to indicate that the subject of caching needs
rethinking
in ivy core and ivyde, and we would like to ask for some
suggestions on
this.
It seems to me that the way you want to use caching is actually
like using
the retrieve job.
The retrieve job would get the dependencies and put it in your
project lib
directory (somehow your by-project cache), and the IvyDE would
build the
classpath based on the content of that lib directory.
Unfortunately this feature is not implemented yet:
https://issues.apache.org/jira/browse/IVYDE-56
Nicolas
