I've noticed that quite a lot of Node.js packages are tagging version 
number zero for all their releases: 0.4.0, 0.9.9, 0.0.1, 0.27.4, etc (to 
pick from packages that I use). It's as if people think that if the program 
is not fully feature-complete, they shouldn't release version 1.0.0.

You need not feel this way! Semver <http://semver.org/spec/v1.0.0.html> exists 
so that, in addition to providing a unique ID for each release, we can 
infer some basic facts about the compatibility of the release, in 
comparison to other releases. It doesn't mean your code has all the 
features you want, it doesn't mean it has any standard of quality, it 
doesn't even mean "beta" or "production-ready". All semver asks you to do 
is (1) tell us when you break reverse-compatibility of your public API, (2) 
tell us when you release a new feature, and (3) tell us when you patch a 
particular bug. *If you use major version zero, we lose all of this 
information.* By definition, major version zero carries no semantics 
whatsoever. ~0 (major version zero) is supposed to be used for internal 
development and quick iteration where nearly every change breaks of the 
public API. However, if you're releasing software publicly, your users 
expect some stability in your public API. The series of releases that are 
stable against one another should carry the same nonzero major revision 
number, like "1.x.x". If you accidentally make a change that breaks, then 
just release a bugfix release for the breakage, and optionally release a 
new major version that carries the breakage.

If you don't identify when you break your public API, then developers have 
to manually figure out which releases are breaking, and which are safe to 
upgrade to.  We may have to carefully examine changelogs and create and run 
unit tests. This wastes developer time. It's also makes it hard to 
future-proof releases: If I know that 1.0.0 is compatible with my 
application, then so should 1.3.1, and any ~1 version. Unit tests are not a 
replacement for the major version number: When picking an appropriate 
package version to update to, developers (or automated programs) do not 
have access to changelogs or the source code to run unit tests on (nor 
should they). (There's also the corollary, version numbers are not a 
replacement for unit tests, of course.) Nor can per-module or per-function 
version numbers replace a package-wide version number. These sub-versions 
may be a good idea, but they do not tell us anything about which version of 
a package, something installed as a coherent whole, should be installed.

Node.js itself is still releasing major version zero. This is unacceptable 
for all the same reasons. Node.js should be releasing 1.0.0 right now (and 
actually, a long while ago). Then, when a new feature is added (major 
change of an internal library, new core library, etc), increment the minor 
version number. If it breaks reverse-compatibility (crypto finally starts 
using buffers, say), increment the major revision number. It might be a 
minor breakage, in which case we can run all our tests and ensure it's no 
change that breaks the program, and then we can say "My program is 
compatible with Node.js ~2 as well as ~1.2". There is nothing so special 
about any feature like libuv that its release can't be marked with 2.0.0 
instead, it's just a number that tells us something broke. It doesn't mean 
it's conforming to any release schedule, it doesn't mean it's feature 
complete.

Having "stable" and "unstable" branches is fine for Git development, 
however having stable/unstable version numbers is not: The stable branch 
should get it's own major version number. Unstable branches would be 
release candidates for the next major version number: 4.0.0-a1, 4.0.0-a4, 
4.0.0-rc1, etc. (Of course this numbering scheme should be avoided in 
production for all the same reasons, it doesn't mean anything, it's just a 
period of rapid iteration and API breakage.)

It's just a number, numbers are cheap. If you need to make a dozen 
consecutive, breaking releases, then simply number them accordingly, 3.0.0 
through 14.0.0. That's how semver works!

Who else has encountered problems with packages breaking the semantic 
versioning scheme and reverse compatibility?

Austin Wright.

-- 
Job Board: http://jobs.nodejs.org/
Posting guidelines: 
https://github.com/joyent/node/wiki/Mailing-List-Posting-Guidelines
You received this message because you are subscribed to the Google
Groups "nodejs" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/nodejs?hl=en?hl=en

Reply via email to