[rspec-users] major release required?
dchelimsky at gmail.com
Sat Jan 22 04:47:25 EST 2011
On Fri, Jan 21, 2011 at 5:13 PM, Andrew Premdas <apremdas at gmail.com> wrote:
> On 21 January 2011 13:56, David Chelimsky <dchelimsky at gmail.com> wrote:
>> On Jan 19, 2011, at 6:48 AM, Rick DeNatale wrote:
>> > On Tue, Jan 18, 2011 at 1:31 PM, David Chelimsky <dchelimsky at gmail.com>
>> > wrote:
>> >> On Jan 18, 2011, at 11:08 AM, Rick DeNatale wrote:
>> >>> On Tue, Jan 18, 2011 at 9:15 AM, David Chelimsky
>> >>> <dchelimsky at gmail.com> wrote:
>> >>>> Hi all,
>> >>>> Since the release of rspec-2.0, I've been following Rubygems'
>> >>>> rational versioning  as closely as possible. Patch releases (2.4.x) have
>> >>>> only had bug fixes, and minor releases (2.x.0) have had new features, but no
>> >>>> (intentionally) backward incompatible changes, which should require a major
>> >>>> (3.0) release.
>> >>>> The autotest extension in rspec-2.0 prefixes the command it generates
>> >>>> with 'bundle exec' if it sees a 'Gemfile' in the project root directory. It
>> >>>> turns out that this is not universally helpful, so there was a request to
>> >>>> have an opt-out.
>> >>>> It also turns out that autotest has a bundler plugin that prefixes
>> >>>> the command with 'bundle exec'. To use an autotest plugin, you just require
>> >>>> it in a .autotest file in the project root. In this case:
>> >>>> require 'autotest/bundler'
>> >>>> I think the right thing to do is to rely on the autotest plugin, but
>> >>>> I also think that this would require a 3.0 release, which feels a bit grand
>> >>>> for this situation. My question to you is: do you think this warrants a
>> >>>> major (3.0) release, or would it be an acceptable exception to the rule
>> >>>> (assuming proper fanfare and documentation)?
>> >>> Can't something be done here as a non-breaking change? I can see two
>> >>> things.
>> >>> 1) add the requested option, I think this is optional
>> >>> 2) in lib/autotest/rspec2.rb
>> >>> def using_bundler?
>> >>> File.exists?('./Gemfile') && !defined Autotest::Bundler # and
>> >>> also check for the option if you decide to do #1
>> >>> end
>> >> I actually did implement a --skip-bundler option (not yet released),
>> >> but it has to be passed like this:
>> >> autotest -- --skip-bundler
>> >> Considering that this is a total hack, and that I'd be removing it at
>> >> the next major release anyway, I really don't want to introduce a hack on
>> >> top of a hack. I'd sooner do a 3.0 release now.
>> > I'm still trying to understand what you are proposing to change, and
>> > what it breaks.
>> > I guess you are proposing that the rspec autotest extension would
>> > never prefix the rspec command with 'bundle exec' and this would break
>> > folks using autotest with rspec who haven't changed their .autotest
>> > file.
>> > And that you think that you should bump the whole rspec suite to
>> > version 3 because of this? I guess this is because the autotest
>> > 'extension' isn't really an extension, it's in rspec-core.
>> Correct. And I'm trying to establish a consistent pattern in releases so
>> people can trust that minor and patch releases won't introduce breaking
>> changes. This one is a bit of an outlier, and I started this thread to see
>> what ppl thought of treating it as such, but the more I think of it, the
>> more I'm convinced that this should not be an exception.
>> > What about those of us using other alternatives to autotest, e.g.
>> > guard? I just looked at the guard code and changing rspec as a whole
>> > to version 3 would break guard since it checks specifically for rspec
>> > version 1 vs. 2 in order to determine whether to use 'spec' or 'rspec'
>> > as the base command.
>> That's unfortunate. Whether or not I do an rspec-3 release now, adherence
>> to Rubygems' rational versioning policy will likely result in an rspec-3
>> release in much less time than it took us to get to rspec-2. When it comes
>> out, rspec-3 will not represent a major rewrite or significant API or
>> functional changes. It will simply be an indicator that there are
>> backward-incompatible changes in that release and you should accept that
>> upgrade consciously and carefully.
>> > If you bump rspec to v3 because of this, it looks like guard users
>> > will need to freeze on rspec 2, at least until the author of
>> > guard-rspec catches up. I guess that's OK unless the latter takes too
>> > long, and rspec continues to improve only on the version 3 branch.
>> > That probably wouldn't happen, and if it does I could fork guard-rspec
>> > myself I guess.
>> Thanks for being willing to help out. We should probably hit guard-rspec
>> up with this now, though, so when rspec-3 does come along guard users don't
>> have to take a hit at that point. Do you want to drive that?
>> > But if you do, I think you should also break out the autotest
>> > extension into a separate gem which is NOT required by rspec-core,
>> > much like Rails 3 broke out 'most-favored' things like Test::Unit and
>> > put alternatives like, say RSpec, and Cucumber on more of an equal
>> > footing.
>> Definitely in the plan for rspec-3:
>> rspec-users mailing list
>> rspec-users at rubyforge.org
> Whilst its very noble to try and follow the ruby gems document to the
> letter, some consideration has to be given to the overall effect of rapidly
> changing major version numbers on the project. RSpec has a very large user
> base, a close tie in with Rails major versions, in particular the idea to
> Rails 3 should use RSpec 2, and a history of changing major versions very
> infrequently with major consequences to the vast majority of users. This I
> think is a fair assessment of RSpec's context re version numbers. To move to
> RSpec 3, for such a small change would be completely out of character for
> the project. To end up in 4 months time with RSpec 9 would be very
> detrimental to the projects reputation.
> So I think the pragmatic approach is a minor release with a big caveat in
> the history and a big announcement on the mailing list
The goal here is to set a new standard that people can count on. Right
now, many rspec users count on upgrades breaking things, which is not
exactly the sort of reliability I'd like to convey.
I want to eliminate any perception of a binding between Rails and
RSpec releases. It's one thing to have a compatibility mapping
(rspec-2 supports rails >= 3), but it's entirely another to assume
that we need to wait for rails-4 to release rspec-3.
As for the timing of major releases, I'm thinking more like a major
release every 6 months to a year, not 7 more in 4 months time :) And,
again, the idea is that major releases won't be quite so major. They
simply indicate that there are breaking changes in the release, and
you should upgrade to that release knowingly.
I've found a solution that I think serves these goals well: rspec-2.5
will add the --skip-bundler option and deprecate the implicit addition
of 'bundle exec'. It will still work as 2.4 does, but users will see a
deprecation warning explaining to use the autotest/bundler plugin or
the --skip-bundler option. When either of those options is applied,
the deprecation warning goes away.
When it comes time to release rspec-3, rspec-autotest will be
extracted from rspec-core, the --skip-bundler will no longer do
anything, and you'll see a message saying so. At that point, you
either use the autotest/bundler plugin or not.
More information about the rspec-users