Localized Version Control

The earliest version control systems in wide use were sccs and the open source rcs.

• We’ll focus on rcs

• The “repository” of historical information is kept as a “special” subdirectory, named RCS

• Sharing of repositories is possible only via the operating system’s underlying mechanism for sharing access to directories
  • permissions, linking

Basic rcs Operations

• ci Check In a file from the working directory into the repository

• co Check Out a file from the repository into the working directory

• rcsdiff Compare two versions of a file.

• rcsmerge

1. History

• mkdir RCS

  Creates an RCS repository for the files in the current directory (only)

  • The repository is currently empty

• ci filename

  Checks files in to the repository

  • If the file is not in there yet, it is added

  • If it is in there, then this becomes the new/current revision

  • Each check in is assigned a new, ascending revision number

    

  • Somewhat surprisingly, deletes the file from the current directory

• co -l filename

  Checks out the most recent version of that file from the repository, storing it in the working directory.

  • Adding a -r v option allows check out of a specific revision number

Revision Numbers

• Clearly there was an intent that revision numbers also serve as version numbers.

  • A special option allows you to force a change to the leading digit,

    e.g., to move from version 1.12 to 2.0

• Problem is that each file’s revision number changes independently
  • So your intended release “version 2.1” might use revision 2.1, revision 2.5 of adt.cpp, revision 2.3 of main.cpp, etc.

• Versions can be checked out by date instead:

  “check out whatever version was current as of 12/13/2012”

  • Repeated over all files, would give a coherent view of the project status as of that date

Naming Revisions

• Revisions can be named:

  ci -N "v1.2" -t "Public release 1.2" *.h *.cpp
  
  

  and later checked out by name instead of by exact revision number

• Note also the option to add explanatory text at the time of the checkout
  • Later version managers would make this “mandatory”

Implementation

rcs is essentially a systematic way of creating and organizing patches.

• The repository always contains the current version of the file plus enough diffs/patches to move back to any prior revision.

  
• The current version is always available immediately.
  • Diffs are used to go back in time
    • Originally considered an important point in supporting efficient access to the most commonly needed file.
    • Now, probably not so important

2. Exploration

Exploring Alternatives

We can start a branch to explore our idea while others continue work on the main trunk.

ci -r1.3.1 _filename_

Checks in our current version of filename as a new branch of development, numbered 1.3.1.1

• 1.3.1.1 is the trunk version from which we branched out

• 1.3.1.1 is the branch number

• 1.3.1.1 is the revision number within the branch

Working in a Branch

Subsequent check-ins of both the main trunk (1.3) and of our branch version will maintain separate revision numbers:

• Note that checking out the most recent version along a branch is not as efficient as checking out the most recent version on the trunk.

Merging a Branch

• If the idea in the branch does not pay off, the branch can simply be abandoned.

• You decide to adopt the changes in the branch, you can elect to merge it back into the trunk.
  • The rcsmerge command is used to conduct the merge,
    • Need to resolve any conflicts introduced by continued development along the trunk.
  • then the resulting combined file checked in with a trunk number
    

Multiple Merges

After a merge

• We might opt to discontinue using the branch

• Or we might continue working along it, eventually generating more changes to be merged into the system

  

Combating Drift

Over time, a long-running branch can get so far out of sync with changes being made to the trunk that the final merge becomes difficult or even impossible.

• An effective strategy for combating this is to periodically merge the trunk into the branch
  • the reverse of the “normal” merge direction
    

3. Collaboration

rcs supports collaboration by locking files

• Most checkouts like this

  co filename
  

  obtain a read-only copy of the file. * nix permissions 400 Can be used to compile system, but cannot be changed * (Of course, you can always chmod, but that’s cheating.

Locks

• A checkout like this

  co -l filename
  

  requests a locked version of the file. * Request fails if a locked version already exists somewhere. * If successful, programmer receives a copy with write permission. * Lock persists until the programmer checks in changes or explicitly releases the lock (which deletes the file from their directory, forcing them to check out an unlocked, read-only version again).

4. Strengths and Weaknesses

• rcs addresses history, exploration, & collaboration concerns

• but has weaknesses in each area

History

• rcs tracks files in a directory.
  • Each file is tracked separately.
• No support for deletion of file
  • Unless you know not to request a file, you will always get the last version before it was deleted.
• No support for creation of new files
  • If you request revisions associated with very old dates, you will get version 1.1 even if the file did not actually exist as of that date.
• No support for renaming files
  • Appears to be a deletion and a subsequent creation of a new, unrelated file
• Each directory is tracked separately
  • Poor support for multi-directory projects

Exploration Issues

• Branching and merging is often confusing.

Collaboration Issues

• Locks are frequently abused
  • e.g., people forget to release a lock, forcing team members to wait
  • People grab locks they don’t really need.
• Cheating on locks is easy
  • People get in the habit of cheating to cope with lock abuse
  • And eventually start cheating with less and less provocation.