Maven 2 & 3

Direct dependencies with scope == provided will not have their transitive dependencies resolved for compiling and testing

Details

  • Type: Bug Bug
  • Status: Closed Closed
  • Priority: Major Major
  • Resolution: Cannot Reproduce
  • Affects Version/s: 2.2.0
  • Fix Version/s: 2.2.1
  • Labels:
    None
  • Complexity:
    Intermediate
  • Number of attachments :
    0

Description

While provided-scoped dependencies should never be passed forward from the current artifact (should not be resolved when the declaring project is itself used as a dependency), it is critical that provided scope and its full dependency sub-tree is resolved for compilation and testing.

Activity

Hide
John Casey added a comment -

The big issue here that needs verification and/or fixing is that when a direct dependency is marked scope == provided, the transitive dependencies below that provided-scope dependency still need to be resolved for the purposes of testing.

For example, assume I'm developing an extension for Maven and I declare a direct dependency on maven-core with scope == provided. My real interest in maven-core is DefaultMaven, to be used to spawn separate, second-level builds (I know, there are many many other reasons this might not work, but just go with it). In this case, maven-core has a dependency on maven-project which is critical.

DefaultMaven depends on DefaultMavenProjectBuilder to read POMs and return MavenProject instances. If I create unit tests that attempt to execute my extension class, those unit tests will require the maven-project artifact (along with many others) to be present before a valid runtime environment for the unit test can be created. If maven-project is excluded from dependency resolution, the unit test will fail with a NoClassDefFoundError.

The place where a problem could arise in all of this is when/if my direct dependency with provided scope itself declares a dependency with scope == provided. Surefire most likely cannot create a valid runtime environment in this case unless it traverses every provided- and non-provided-scope dependency. Since the code makes the assumption that those provided-scope dependencies will be present at runtime, the calculation for creating that runtime environment for the purposes of unit testing may become very difficult.

Show
John Casey added a comment - The big issue here that needs verification and/or fixing is that when a direct dependency is marked scope == provided, the transitive dependencies below that provided-scope dependency still need to be resolved for the purposes of testing. For example, assume I'm developing an extension for Maven and I declare a direct dependency on maven-core with scope == provided. My real interest in maven-core is DefaultMaven, to be used to spawn separate, second-level builds (I know, there are many many other reasons this might not work, but just go with it). In this case, maven-core has a dependency on maven-project which is critical. DefaultMaven depends on DefaultMavenProjectBuilder to read POMs and return MavenProject instances. If I create unit tests that attempt to execute my extension class, those unit tests will require the maven-project artifact (along with many others) to be present before a valid runtime environment for the unit test can be created. If maven-project is excluded from dependency resolution, the unit test will fail with a NoClassDefFoundError. The place where a problem could arise in all of this is when/if my direct dependency with provided scope itself declares a dependency with scope == provided. Surefire most likely cannot create a valid runtime environment in this case unless it traverses every provided- and non-provided-scope dependency. Since the code makes the assumption that those provided-scope dependencies will be present at runtime, the calculation for creating that runtime environment for the purposes of unit testing may become very difficult.
Hide
John Casey added a comment -

this was an incorrect reading of the problem. Once I dug deeper into the case I was working on, we found the real problem

Show
John Casey added a comment - this was an incorrect reading of the problem. Once I dug deeper into the case I was working on, we found the real problem

People

Vote (0)
Watch (1)

Dates

  • Created:
    Updated:
    Resolved: