Anthony Laforge, the Google program manager in charge of Chrome releases, was kind enough to give a talk at Mozilla's all-hands meeting in December. He gave a version of this presentation: http://techcrunch.com/2011/01/11/google-chrome-release-cycle...
Each Chrome release is in development for 12 weeks. There are always two 12-week cycles in progress at once, staggered by 6 weeks. So users on the stable channel see a new version every 6 weeks.
This has benefits much more profound than just a large number before the dot. For example, developers working on new features actually feel less schedule pressure. If a feature misses the deadline for version N, the team knows they'll have another chance just six weeks later for version N+1. And for users, the browser becomes more like a web app. Do you know what version of GMail you use? Do you notice every time they add fix bugs or improve performance on Google Maps? No - every time you open the app, you simply have the latest version.
Laforge mentioned some things about Chrome development that make this work:
- Silent background updates. When an update is available on your channel, Chrome downloads it in the background and installs it. The next time you launch the browser, it is running the new version and deletes the old one. Because of this, the majority of Chrome users are up-to-date within days of each release.
- Feature switches. Development is done on trunk, but each work-in-progress feature can be disabled with a switch (either a preference, a build option, or a command-line flag). This lets developers work for several cycles before turning a feature on for the stable channel. It also lets product management disable a new feature if bugs are found during beta, without changing any code. Web startup folks might recognize this as the "always ship trunk" pattern used by sites like Flickr: http://news.ycombinator.com/item?id=1463751
- No support for old versions. Instead of deciding which "version" to install, Chrome users decide which "channel" to follow. If you are a developer and want to test upcoming changes, use the canary or dev channels. If you want a more tested browser, use the beta or stable channels. New features roll out to each channel in turn. If you are on one of these four channels, you get updates. If you are not, then you don't. (Just like you can't choose to run an old version of GMail or Google Reader.)
Mozilla is just starting to talk about the new roadmap and how to achieve it. Chrome's experience is useful, and it was really helpful of Laforge to share it with us. But we're different both technically and organizationally from Chrome, so just blindly copying what they do would not be a good idea. The wiki shows a plan for releases every 3 months (not every 1.5 months like Chrome). We use Mercurial with long-lived project branches (unlike Chrome which uses Subversion and does all development on trunk). And our product differs in some important ways, like the ABI for binary extensions. (Even non-binary Firefox extensions may reach deeper into the browser internals than Chrome extensions.)
But we do think that the benefits for users (more frequent improvements to the browser) and developers (more predictable schedule, fewer release delays) are worth pursuing. There are benefits for web developers too, since it reduces the time from when we start implementing new web-facing features (like IndexedDB, or CSS3 transitions) to when you can deploy them to more of your users.
Arguably, the concept that ties major releases with major purchases is old and broken. It discourages companies from making major updates to software between such releases, as it will actually make it harder for them to make new revenue off the next major release. So you get point releases that are mostly minor bug fixes, and then a huge release every year or two that the maker hopes will squeeze you into opening your wallet, even if you're mostly happy with the last release.
Web apps are clearly going to break this version treadmill. Even if you're downloading a package to the desktop, modeling software more as a service - you pay for the right to use software for a period including all updates, rather than for a perpetual license that in reality will expire and need to be re-purchased as soon as the next major version hits.
Pay as you go is better for both users and developers. It means you can charge less up front, spend less on marketing (since the cost of trialing the software drops) and focus on delivering the _best_ experience to all your users rather than denying some features to existing customers in order to create a future revenue event.
That assumes a series of minor improvements equals a major release, but that isn't always the case. Sometimes something might need a year's work because it is a change to the core of the programme(and old features have to be re-added in a different way for the new core and users are never happy about losing features) but with your system that gets completely disincentivised and minor improvements get prioritised. All the incentives point to bloatware, you create an expectation of constant releases that update or add features so you have to fulfil them, but you never have long enough to change the core and get back to where you started feature wise. That or they cheat and don't meet expectations of what version changes are and instead of point releases that are "mostly minor bug fixes" you get version changes that are so and then they need a new way of marketing that "huge release every year or two".
"...the maker hopes will squeeze you into opening your wallet, even if you're mostly happy with the last release."
I don't understand, what is the problem with that? They can hope all they want, if you're happy with the last release you don't need to update in the "old fashioned" system, it's yours forever. In the "new" subscription system you could be perfectly happy with the product as it is but you are forced to keep paying for updates just to use it.
"Changing the core" isn't generally any more encouraged by either system, if it doesn't result in marketable features.
The idea of the "complete overhaul" is something engineers love to dream about, business managers hate to do and end users generally just don't care that much about. That's why they happen so seldom - and usually only if the existing package is a pile of unmaintainable crap.
The shrink-wrap model encourages chasing adding more bullet points to the outside of the new box over actually improving the day-to-day user experience. The service model encourages making your actual customers happy with their product AFTER the purchase, so they keep buying.
The whole concept of a large, up-front fee is driven by traditional mass-media marketing strategy: spend a bunch of money making something sound really great to buy, sell it for a bunch of money up front so you can get a return on your advertising spend. Don't worry about what happens after that.
"I don't understand, what is the problem with that?"
The problem with that is if there is some small feature change that would really improve the current product (say supporting an additional new file format on import) there is little reason for the maker to add it to the old release. Instead, they lump it in with the new version and hope it forces you to buy a whole new license. So a feature that might only add a small marginal cost but would make current users happy for longer doesn't get released to them.
Firstly I said nothing of "complete overhaul", core changes are for example, the internet moving to IPv6, OSes moving from 32-bit to 64-bit.
Anything that is constantly updated with anything more than bug-fixes will become "a pile of unmaintainable crap", or just obsolete if the core isn't changed. Core changes normally happen before that point though. That is if business managers' incentives aren't all wrong. An example of that fairly recently was twitteriffic for iOS, they felt it was heading towards being a pile of crap so they had to stop and go back to the core.(even though it meant unhappy current customers)
Users not caring about it unless it immediately comes with new features or at least all the old features is exactly my point! And if what the user is paying for is updates they are absolutely not going to accept losing any features(and not going to be happy with not updating since that is the whole reason that they are paying in your method). If the user owns a version and sees that the next version doesn't have features they need they'll stick with the old version until the new version has those features. In your model they have already paid for that update.
"So a feature that might only add a small marginal cost but would make current users happy"
Prices aren't just based on costs, they are usually(in the IT sector) based on how much more they "make users happy", in some cases they are almost entirely based on that, clothes, shoes and apple products being the most obvious examples. Which to repeat in another way is the problem with core updates in your system, they are high "cost" but they only prevent the programme from going to shit so they don't "make current users happy"(just prevent future unhappiness).
That's not to say that your model doesn't work in some cases, anti-virus programmes need constant updates but very seldom core changes.(although oddly enough at least for norton it's much cheaper to buy the newer version in amazon or a shop and get continued subscription that way then to update a subscription) And you mentioned GPS apps the same applies to them, no core changes needed.
Angry Birds on iOS is absolutely not a case of it though, you don't pay for new levels, new levels are free and are an example of the opposite of what you're saying happening, you pay upfront and they have continued adding levels way beyond a point where it would have been reasonable for them to make an Angry Birds 2 and putting the new levels in that. It is a special case but it is the traditional model just with a new attitude (on iOS, on android it's the ad revenue model which is different again).
Sounds great but I am never going to pay a subscription for my IDE, or my FTP program or my Twitter client. Relationships require trust and effort from both parties and it's simply not a perfect model for every piece of software.
I don't want a relationship with a company. For most apps I want to buy and own a thing as it is now.
This "old and broken" model persists because it's something that customers get instantly and can commit to knowing in advance what their full investment will be.
With Android apps (and iPhone I presume), it seems that you pay once and automatically get any available updates. I'd guess that would work reasonably well for small utilities, so long as they continue to get new customers.
That hasn't always held true for some of the major iPhone apps, at least - when there's a major version release, they sometimes require a new purchase.
Also, iPhone and Android apps have the potential for ongoing in-app purchases by users. So you buy the GPS app once for a very low price (or free) but you have to keep buying the data updates if you're using it. Or new levels for Angry Birds.
I don't mind faster releases if they're doing things we traditionally associated with minor version numbers, such as performance improvements or slightly UI tweaks.
I really hope they don't change any rendering components at the same pace, though. I already have to strike any contractual obligation to test/support Chrome in commercial web development projects, because it's a moving target and Google do release breaking changes. If Mozilla browsers go the same way, then I really will start coding for the nice, stable, standards-friendly world of IE -- and that's never a statement I thought I would write on a serious forum with a straight face!
Recently I worked on a website with a lot of forms, and decided to use HTML5 form validation to make my life easier, with a JS fallback for older browsers. It turns out that Webkit added support for HTML5 form validation, then found a problem, so they switched off the implementation without switching off the exposed API. So every JS fallback that tests for the presence of the HTML5 methods on form elements will assume that HTML5 form validation is implemented, and leave the validation up to the browser... which doesn't actually do it at all.
That means server-side validation, which of course one needs to do anyway. But it's not to hard to imagine somebody making a site during the brief window when Chrome supported HTML5 form validation, and then discovering their site broken in Chrome a few months later.
They didn't "switch it off" they merely turned off form submission blocking when the form was invalid. It's back on in chrome now because they have the error notification bubble. The constraint validition API, the error pseudo-classes and the extra attributes all work in Safari.
If the answer to "are the contents of this form valid" is always "yes", I don't really care how much of the validation API the browser supports; it's effectively going to waste.
I'm glad to hear it's turned back on; I hope the form-validation interface is as sleek and professional-looking as the one in Firefox 4.
> Do you notice every time they add fix bugs or improve performance on Google Maps? No - every time you open the app, you simply have the latest version.
Since you asked... not every time, but often. You see, it's apparent from my experiences that Google makes incremental changes to the live system.
So, somewhat frequently, I'll see mis-loaded pages. Sometimes I fire up the web inspector and see that someone's typoed a JavaScript function, or a CSS element. Very annoying. This happens with GMail and GMaps fairly often.
For clarification, are all the different channels simultaneously present in stable release that regular Joe's dl and install, just hidden by by pref/build option/cl-flag switches?
For example, when you say developers don't have to decide which version to install, but only which channel to follow, is that how they switch among channels - just restart the stable version using a different switch?
No, the update channel determines which code is downloaded to your computer. The "stable" and "beta" channels receive builds from release branches, while the "dev" and "canary" channels receive snapshots from trunk.
Regardless of your channel, you can turn on different features in Chrome by going to "about:flags" (or sometimes by passing command-line switches) - but only if those features are in the build you are running.
Anthony Laforge, the Google program manager in charge of Chrome releases, was kind enough to give a talk at Mozilla's all-hands meeting in December. He gave a version of this presentation: http://techcrunch.com/2011/01/11/google-chrome-release-cycle...
Each Chrome release is in development for 12 weeks. There are always two 12-week cycles in progress at once, staggered by 6 weeks. So users on the stable channel see a new version every 6 weeks.
This has benefits much more profound than just a large number before the dot. For example, developers working on new features actually feel less schedule pressure. If a feature misses the deadline for version N, the team knows they'll have another chance just six weeks later for version N+1. And for users, the browser becomes more like a web app. Do you know what version of GMail you use? Do you notice every time they add fix bugs or improve performance on Google Maps? No - every time you open the app, you simply have the latest version.
Laforge mentioned some things about Chrome development that make this work:
- Silent background updates. When an update is available on your channel, Chrome downloads it in the background and installs it. The next time you launch the browser, it is running the new version and deletes the old one. Because of this, the majority of Chrome users are up-to-date within days of each release.
- Feature switches. Development is done on trunk, but each work-in-progress feature can be disabled with a switch (either a preference, a build option, or a command-line flag). This lets developers work for several cycles before turning a feature on for the stable channel. It also lets product management disable a new feature if bugs are found during beta, without changing any code. Web startup folks might recognize this as the "always ship trunk" pattern used by sites like Flickr: http://news.ycombinator.com/item?id=1463751
- No support for old versions. Instead of deciding which "version" to install, Chrome users decide which "channel" to follow. If you are a developer and want to test upcoming changes, use the canary or dev channels. If you want a more tested browser, use the beta or stable channels. New features roll out to each channel in turn. If you are on one of these four channels, you get updates. If you are not, then you don't. (Just like you can't choose to run an old version of GMail or Google Reader.)
Mozilla is just starting to talk about the new roadmap and how to achieve it. Chrome's experience is useful, and it was really helpful of Laforge to share it with us. But we're different both technically and organizationally from Chrome, so just blindly copying what they do would not be a good idea. The wiki shows a plan for releases every 3 months (not every 1.5 months like Chrome). We use Mercurial with long-lived project branches (unlike Chrome which uses Subversion and does all development on trunk). And our product differs in some important ways, like the ABI for binary extensions. (Even non-binary Firefox extensions may reach deeper into the browser internals than Chrome extensions.)
But we do think that the benefits for users (more frequent improvements to the browser) and developers (more predictable schedule, fewer release delays) are worth pursuing. There are benefits for web developers too, since it reduces the time from when we start implementing new web-facing features (like IndexedDB, or CSS3 transitions) to when you can deploy them to more of your users.