[
https://jira.codehaus.org/browse/MNG-5448?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=341516#comment-341516
]
Markus KARG commented on MNG-5448:
----------------------------------
What I proposed is that Maven learns to deal with binary dependencies in a very
sophisticated and simple-to-use way.
NAR does not do that. It puts up a heavy load on the user and the runtime
system.
Until today, nobody was able to provide a simple and easy to apply config
example how NAR could solve the issue I expressed in the above feature request.
Hence I really beg that Maven will provide a <type>dll</type> DEPENDENCY
support (this is about DEPENDENCIES, not about PRODUCING binary packages)! :-)
If someone knows a dead-simple solution _with_ NAR, then please post the config
NOW AND HERE so we can all be happy without a change in Maven. ;-)
Regards
-Markus
> Putting dll/so-type dependencies on java.library.path by default
> ----------------------------------------------------------------
>
> Key: MNG-5448
> URL: https://jira.codehaus.org/browse/MNG-5448
> Project: Maven 2 & 3
> Issue Type: Story
> Reporter: Markus KARG
>
> Many people are using Maven to build products that are using native
> dependencies (e. g. using JNI directly or indirectly). These people have to
> declare in some way this dependency, so Maven will download the DLL/SO to
> [target]. But what they also need is to tell Java that the target folder has
> to be scanned for the DLL/SO, hence they need to set java.library path.
> There are several solutions but all of them lack one thing: They are no
> default but always do "tricks". This is a problem, because:
> * m2e needs to tell Eclipse where to find DLLs.
> * There might be other tools that want to read or write the POM.
> * How should different tools understand that DLL dependencies are to be put
> on java.library.path if there is no STANDARD way to find this information in
> the POM?
> The only "real" solution would be that Maven 3.1 clearly defines the one and
> only unique way!
> In my opinion the way clearly means:
> * Clearly define that any <packaging>dll</packaging> and
> <packaging>exe</packaging> dependencies MUST be put to [target/native] and
> that [target/native] MUST be part of java.library.path!
> That way, it will become most simple to use any DLL/SO as a dependency,
> either Maven-built or not. And it will be absolutely clear to m2e and any
> other tools that such dependencies must be told to Eclipse and other
> environments as to be put on the java.library.path!
> I do not say that it must exactly work THIS way, but what I really like to
> say is that Maven MUST provide ONE and EXACTLY ONE clear and unambiguous way
> to tell ANY POM-reading tool that a dependency is a DLL/SO and MUST be found
> on java.library.path in turn.
> This is NOT up to the POM author, as it is pretty clear that ANY DLL/SO that
> is a dependency only serves the single and simple purpose of BEING on
> java.library.path -- otherwise one would not have made it a dependency!
> I mean, ANY JAR dependency is put on the CLASSPATH already, so it makes
> pretty much sense to do the same with DLL/SO dependency and java.library.path!
--
This message was sent by Atlassian JIRA
(v6.1.6#6162)
