Hi James,
I checked in a test case that confirms this situation works, again as
long as you set the finalName. Would you like to check it is what you
expected?
I haven't tested strong names yet. I believe that they will work, as
you use the classifier to reference them and no version mediation is
used, but I'll add a test case as well.
- Brett
On 01/10/2008, at 9:17 AM, James Carpenter wrote:
I believe the description of a very old JIRA I created against the
original C# support in the maven sandbox is still relevant to your
questions.
---------------------------
The JIRA issue is:
URL: http://jira.codehaus.org/browse/MSANDBOX-13
Date: Aug 23, 2006
--------------------------
Description in MSANDBOX-13:
Further experience with the maven csharp plugins has revealed an
interesting side affect of the current way in which maven built
csharp libraries are used. As mentioned in MNG-2369, the csharp
libraries built by maven have the version number in their name.
Assume the following library heiarchy: A depends upon B which
depends upon C (A->B->C).
Lets assume the initial versioned dependencies are as follows:
A_1.0 (explict dependency upon B_1.0)
B_1.0 (explict dependency upon C_1.0)
C_1.0
Now lets assume C has changed to add some new feature needed by a
new version of A. Lets assume that although A needs the new feature
of C, the interfaces from C used B have not changed and hence no
code changes are necessary to B.
So we now try (Will not work with CSharp even though Java code would
be fine):
A_2.0 (explict dependency upon B_1.0, and C_2.0) Note: 2.0 version
of C superceeds 1.0 in typical mvn fashion
B_1.0 (explict dependency upon C_1.0)
C_2.0
This new configuration fails when the unit tests for A_2.0 are run.
When the unit tests in A_2.0 are run we see that B_1.0 is looking
for C_1.0 which doesn't exist as C_2.0 has taken its place in the
dependency list. Remember that B_1.0 is looking for C_1.0 because
the assembly meta-data in B_1.0 says it needs an assembly named
C_1.0.dll.
If none of the assemblies are strongly-named (assembly meta-data
contains digital signatures for each dependency) it would be
sufficient if the dependencies within the assembly meta-data didn't
contain the version numbers. (Such a change would have synergies
with whatever was done for 3rd party libraries.)
Alternatively, I think one can probably include all versions
mentioned by any of the dependencies. In this case it is important
to maintain version numbers as part of the dependency names as doing
so allows them to co-exist in the same directory. (Could be
problematic for 3rd party dlls without version numbers in their name.)
All of the above solutions require a change to the csharp maven
support in some fashion. The only solution available today is to
create a new release of B which uses the newer version of C.
A_2.0 (explict dependency upon B_2.0)
B_2.0 (explict dependency upon C_2.0)
C_2.0
The inability to override versions is both an advantage and
disadvantage. As you can see there the advantage to the current
solution is that B is now known to work with C_2.0. The disadvantage
is one must re-release B just to get the updated C version.
Summary: Version overriding with CSharp dependencies doesn't work
out. A general solution to the problem is either impossible or at
least awkward. The issue stems from the decision by MS to support
digitally signed libraries, and the particulars of the current mvn
csharp plugin behavior.
--- On Tue, 9/30/08, Brett Porter <[EMAIL PROTECTED]> wrote:
From: Brett Porter <[EMAIL PROTECTED]>
Subject: working around the local repository / filename limitation
To: [email protected]
Date: Tuesday, September 30, 2008, 1:36 PM
Hi,
Trunk previously required a re-compiled NUnit assembly to
work, which
disappeared in the snapshot repository house keeping. I
wanted to try
again with a standard NUnit assembly, and saw that it
failed to run
the tests due to the filename issue. Fixing this is
important, as not
all assemblies can be recompiled.
I added a simple class DotnetArtifactLocator that will load
the POM of
the artifact and check the finalName. If it is set, it is
returned
instead of the previous filename. If requested, it will
copy it to the
$version/$finalName.dll location in the local repository
(ie, one
level deeper in the current directory, as Dan suggested),
if it
doesn't already exist, and return that location
instead. This is meant
to be a very simple utility that plugins can use when they
need a
renamed artifact, and is potentially compatible with a
future local
repository layout adjustment.
This resolved the issue for NUnit (as the file in
target-assemblies
could easily be renamed). I noted it wasn't needed for
compilation at
all, so I didn't include that. On the current trunk,
nothing else uses
artifacts.
Questions:
- any thoughts on this approach?
- is there any reason the rename would be needed for
compilation?
- are there other scenarios on the current trunk where the
rename is
needed?
I'd like to put this matter to bed :) The work is on
the branch
'NMAVEN-8-nunit'. I still have a bit more to do,
but would like to
merge it back into trunk after those changes tomorrow.
The repository I'm building up is at
http://people.apache.org/~brett/repositories/dotnet/
for now - I still need to go back and add licenses, etc
before
pushing this anywhere permanent.
Thanks,
Brett
--
Brett Porter
[EMAIL PROTECTED]
http://blogs.exist.com/bporter/
--
Brett Porter
[EMAIL PROTECTED]
http://blogs.exist.com/bporter/
