/ 
VCI-0.1.0_1
River stage zero No dependents
/ VCI::VCS::Cvs

NAME

VCI::VCS::Cvs - The CVS implementation of VCI

DESCRIPTION

This is a "driver" for VCI for the CVS (Concurrent Versioning System) version-control system. You can find out more about CVS at http://www.nongnu.org/cvs/.

For information on how to use VCI::VCS::Cvs, see VCI.

CONNECTING TO A CVS REPOSITORY

For the repo argument to "connect" in VCI, choose what you would put in the CVSROOT environment variable.

The constructor also takes two additional, optional parameters:

x_cvs

The path to the "cvs" binary on your system. If not specified, we will search your PATH and throw an error if cvs isn't found.

x_cvsps

The path to the "cvsps" binary on your system. If not specified, we will search your PATH and throw an error if cvsps isn't found.

REQUIREMENTS

In addition to the Perl modules listed for CVS Support when you install VCI, VCI::VCS::Cvs requires that the following things be installed on your system:

cvs

The cvs client program, at least version 1.11. You can get this at http://www.nongnu.org/cvs for *nix systems and http://www.cvsnt.org/ for Windows systems.

cvsps

This is a program that interacts with CVS to figure out what files were committed together, since CVS doesn't normally track that information, and VCI needs that information.

You can get it from http://www.cobite.com/cvsps/. (Windows users have to use Cygwin to run cvsps, which you can get from http://www.cygwin.com/.)

REVISION IDENTIFIERS

cvsps groups file commits that are close together in time and have the same message into "PatchSets". Each of these PatchSets is given a unique, integer identifier.

Since VCI::VCS::Cvs uses cvsps, the revision identifiers on Commit objects will be these PatchSet ids.

For File objects, the revision identifiers will be the actual revision identifier as returned by CVS for that file. For example 1.1, etc.

For Directory objects, the revision identifier is currently always HEAD.

LIMITATIONS AND EXTENSIONS

  • Currently VCI doesn't understand the concept of "branches", so you are always dealing with the HEAD branch of a project. This will change in the future so that VCI can access branches of projects.

  • cvsps needs to write to the HOME directory of the current user, you must have write access to that directory in order to interact with the History of a Project.

  • VCI::VCS::Cvs has to write files to your system's temporary directory (/tmp on *nix systems), and many operations will fail if it cannot. It uses the temporary directory returned by "tmpdir" in File::Spec.

  • If your program dies during execution, there is a chance that directories named like vci.cvs.XXXXXX will be left in your temporary directory. As long as no instance of VCI is currently running, it should be safe to delete these directories.

In addition, here are the limitations of specific modules compared to the general API specified in the VCI::Abstract modules:

VCI::VCS::Cvs::Repository

get_project doesn't support modules yet, only directory names in the repository. Using a module name might work, but operations on that Project are likely to then fail.

VCI::VCS::Cvs::Commit

CVS doesn't track the history of a Directory, so Directory objects will never show up in the added, removed, modified, or contents of a Commit.

VCI::VCS::Cvs::Directory

  • For the time accessor, we return the time of the most-recently-modified file in this directory. If there are no files in the directory, we return a time that corresponds to time() == 0 on your system, probably January 1, 1970 00:00:00. Currently this is a fairly slow operation, but it may be optimized in the future.

  • All Directory objects have a revision of HEAD, even if you get them through the parent accessor of a File.

  • If you manually create a Directory with a revision other than HEAD, the contents will be incorrect.

PERFORMANCE

VCI::VCS::Cvs performs fairly well, although it may be slower on projects that have lots of files in one directory, or very long histories.

Working with a local repository will always be faster than working with a remote repository. For most operations, the latency between you and the repository is far more important than the bandwidth between you and the repository.