Repository operations
Mercury replaces the good old Repository.pathOf( GAV ) with an active Repository API and implementations for local and remote M2 repositories.
Repository exposes only GAV-based operations to the clients, leaving storage details inside. This allows us to clearly separate concerns of using repository from gory storage details.
- Artifact is now represented by 3 staged of completeness
- ArtifactBasicMetadata is just bare GAV coordinates, where V could be a VersionRange
- ArtifactMetadata adds dependency as List<ArtifactBasicMetadata>, POM order preserved
- DefaultArtifact adds the reference to a binary file
- Repository
- attributes
- type - a.k.a. layout, it's now used to instantiate a particular implementation
- local - indicates whether it's close by and remote artifacts should be stored in it
- readOnly - if local - don't try to install into it, if remote - do or do not deploy
- releases
- snapshots
- methods
- getReader - below
- getWriter - below
- attributes
- RepositoryReader is the abstraction that does the hiding. Simplified description:
- readVersions(ArtifactBasicMetadata) - call, analogous to reading maven-metadata.xml or all folder names from GA directory, produces a List<ArtifactBasicMetadata> that meet the range criteria of the supplied parameter
- readMetadata(ArtifactBasicMetadata) - call, analogous to reading pom.xml or all folder names from GA directory, produces a ArtifactMetadata with non-transitive dependencies already in. It uses supplied MetadataProcessor to process the pom.xml, ProjectBuilder will have to play that role until we separate dependencies from the rest of project metadata
- readArtifact(ArtifactBasicMetadata) - actually reads the binary and places it into local repositories, if necessary.
- RepositoryWriter a simpler component with only one purpose in life - write a DefaultArtifact into a repository
A picture can substitute 1000 words, what about a sequence diagram ?
This approach, for instance, allows one to easily write a database-based repository, and much more.
RepositoryReader API explanation
RepositoryReader has 3 major calls, that all the rest is based upon:
- readVersions() it receives a version quiery and interprets it depending on the repository (metadata for remote) contents. It is the only API that deals with version ranges and special suffixes like LATEST or SNAPSHOT, all other APIs accept version literally and don't try to interpret it
- readDependencies() - based on the artifact coordinates and environment (passed as Hashtable) returns artifact dependencies. It actually calls out to DependencyProcessor implementation to interpret POM files, thus is just a thin layer that interprets what DependencyProcessor digested and spit back. It repo reader also provides a readMetadata() callback so that DependencyProcessor can access all the metadata it needs
- readArtifacts() is the simplest of them - just reads artifacts based on supplied coordinates. It can also use results of previous calls to readVersions() because the latter transiently records the repository where a version was found.
RepositoryWriter API explanation
RepositoryWriter is a primitive writeArtifacts() implementation. Due to intimate relationship with JettyTransport, implementation of this API call in remote-repository-m2 is also transactional; it ensures all-or-nothing deployment.
Artifact versions special treatment
On top of version ranges, Repository Reader and Writer are responsible for correctly interpreting special version queries, like g:a:1.1-SNAPSHOT, g:a:1.1-20080820.195323-15, g:a:LATEST, g:a:RELEASE because those do not follow standard M2 convention for repository structure. As a result:
- g:a:1.1-SNAPSHOT
- in local repo - I scan the repo dir for latest snapshot
- in remote repo - I rely on GAV-level maven-metadata.xml to find the latest snapshot
- g:a:1.1-20080820.195323-15
- in local repo - I scan the repo snapshot gav dir for this particular snapshot
- in remote repo - I rely on GAV-level maven-metadata.xml to find this snapshot
- g:a:LATEST
- in local repo - I scan the repo ga dir for latest version. Latest is defined by DefaultArtifactVersion.compareTo()
- in remote repo - I rely on GA-level maven-metadata.xml to find latest release or snapshot
- g:a:RELEASE same as LATEST, but excludes SNAPSHOTs
- in local repo - I scan the repo ga dir for latest version, excluding SNAPSHOTs. Latest is defined by DefaultArtifactVersion.compareTo()
- in remote repo - I rely on GA-level maven-metadata.xml to find latest release
Virtual Repository concept
Working with repositories I saw a lot of places where we pass (LocalRepository, Set<RemoteRepository>) and then process them. To abstract this usage pattern in Mercury I introduced VirtualRepositoryReader. This object is constructed out of List<Repository> and then shifts repositories inside so that local ones are upfront, and then performs requested operations that mimic single repository operations.
There is less need in VirtualRepositoryWriter so I still debating the necessity of it. One use case is caused by introduction of repository QualityRange - see in Mercury Version Ranges, here VirtualRepositoryWriter can try several repositories before writing the one, accepting quality of supplied artifact. Beyond that - we almost always write to a particular repository.
3 Comments
Unknown User (ddimitrov)
How does the API accommodate for classifiers?
Sometimes we need to access a secondary artifact - i.e. the repository POM, javadocs, source code, distribution archive, etc.
Olle Jonsson
Dimitar - if a query GAV (expressed as ArtifactBasicMetadata) has classifier - it will be used. The sole purpose of Repository abstraction is to hide storage details and expose useful GAV operations - read/write/lookup.
There is a big debate whether classifiers should contribute to the classpath, but this is outside of this page's topic![(smile)](/confluence/s/p9b4rx/8804/1naq7q0/_/images/icons/emoticons/smile.svg)
Ludovic Claude
VirtualRepositoryWriter could help me to integrate better Maven with a Debian system: in Debian, packaging is already handled for many components, including Java libraries (and later I hope Java EARs and WARs). So it would make sense to have a local repository which works in a slightly different way that currently: the local repository first checks if the artifact requested is installed (usually, in /usr/shared/java). If it is, then it uses the library there and generates the POM on the fly using the metadata in Debian; otherwise the standard repository in ~/.m2/repo is used.