Misc #8835
Updated by knu (Akinori MUSHA) over 10 years ago
=begin [This is a presentation for [[ruby:DevelopersMeeting20130831Japan]].] ((<DevelopersMeeting20130831Japan|URL:https://bugs.ruby-lang.org/projects/ruby/wiki/DevelopersMeeting20130831Japan>)).] Ruby's versioning scheme is currently not well defined or well utilized. First, RUBY_API_VERSION was introduced in 1.9.1, but has not been properly maintained since then. There must have been at least a few ABI incompatible changes (although we tried hard to keep source level compatibility) and many API additions between 1.9.1 and 1.9.3, but RUBY_API_VERSION was never bumped. Secondly, it looks like TEENY version is fixed to zero in the 2.x series without being used effectively. We often find overlooked incompatibility or feature bugs after an official release x.y.0, and would like to fix them in a backward compatible way. I'd suggest we accept such a situation will happen, and bump TEENY when such a fix that affects forward compatibility is made. We have up to nine times of chance for fixing such a situation, and shouldn't be too worried about running out of digits. So, here I propose introducing a new versioning scheme as follows, much with Semantic Versioning (http://semver.org/) in mind: (1) From 2.1.0 and on, RUBY_API_VERSION shall match RUBY_VERSION. Giving them different numbers has been a source of confusion, and introducing this scheme is a way to provide RUBY_VERSION with how to read API compatibility from the numbers. (2) MINOR version is incremented when an incompatible API change is made. In practice, it is incremented before a change to break compatibility is actually made. It is when When a relatively significant amount of change is going to be made, which would break compatibility, or hurts stability for a certain period of time. Before making such a change, a new stable branch ruby_{MAJOR}_{MINOR} is cut off from trunk, and MINOR version is bumped on trunk. MAJOR version may be incremented instead, resetting MINOR to zero. On the ruby_{MAJOR}_{MINOR} branch, when the time comes for prereleasing {MAJOR}.{MINOR}.0, a new branch ruby_{MAJOR}_{MINOR}_0 is cut off from the ruby_{MAJOR}_{MINOR} branch. (3) TEENY version is incremented when functionality is added in a backward-compatible manner. It happens on ruby_{MAJOR}_{MINOR} branches, where a new branch ruby_{MAJOR}_{MINOR}_{TEENY} is cut off when it is released, resetting PATCHLEVEL to zero. (4) PATCHLEVEL is initialized to zero on each ruby_{MAJOR}_{MINOR}_{TEENY} branch where TEENY > 0 on creation, and incremented every time a change is made on the branch. On a ruby_{MAJOR}_{MINOR}_0 branch, PATCHLEVEL is kept at -1 during the prerelease period. It is set to zero when {MAJOR}.{MINOR}.0 is officially released. On other branches, it is fixed to -1. Note that MINOR and TEENY need not be bumped every time an applicable change is made, but once before a new official release is rolled out from the branch. With this scheme introduced, version specific library subdirectory names only need to have {MAJOR}.{MINOR} in it, and user can safely upgrade ruby to a new teeny version without having to rebuild and reinstall already installed libraries. [Figure: Branch Tree] --o-----------------------o----------(trunk) \ \ o--o--o--o--(ruby_2_1) o--o--...--(ruby_2_2) \ \ \ \ \ \ \ o--[v2_2_0_0]--...--(ruby_2_2_0) \ \ \ \ \ o--[v2_1_2_0]--[v2_1_2_ppp]--...--(ruby_2_1_2) \ \ \ o--[v2_1_1_0]--[v2_1_1_ppp]--...--(ruby_2_1_1) \ o--[v2_1_0_0]--[v2_1_0_ppp]--...--(ruby_2_1_0) =end