More specifically, Dart on App Engine via Managed VMs. (Dart was already runnable on Compute Engine instances.)
App Engine is a subset of Cloud Platform services and is a platform-as-a-service (PaaS) offering – auto scaling, a scalable datastore, logging and other managed services and guarantees that aim to substantially reduce your DevOps.
By contrast, Compute Engine (and its closest competitor AWS) is more an infrastructure-as-a-service (IaaS) offering – provision your instances and run anything you'd like on it, as well as choose from available resources like load balancers and firewalls to meet your needs.
Managed VMs attempt to offer the best of both worlds: the managed scalability of App Engine with the choice and flexibility of Compute Engine. Through a Docker-based container approach, App Engine scalability guarantees are still kept while allowing you to run your own language, like Dart or Node.js, and services.
My Dart app, http://woven.co, has been running on Compute Engine, and I plan to move it to App Engine.
Care to elaborate? I believe it has been publicly stated in the past by the team that this is something being actively worked towards once OilPan (or whatever it's called) is finished? Has this changed?
We are working on it. Inbox was designed to be 'mobile first' and use "Material Design" animations, and pulling off 60fps animations of that style in an app like Inbox is very complex, in fact, each new version of Chrome broke assumptions and broke animations we had running previously. The Web, as a platform for competing with mobile-style interfaces, is in a rickety state at the moment, where performance is non-deterministic not only across browsers, but across versions. We often have to work with the engineers on the browser layout engines directly to resolve these issues. We're working full time on Firefox now, with Mozilla engineers helping, which is a lot easier to do now that the project isn't secret anymore.
I do think it is unfortunate that the error page advertised Chrome. I think we could have handled that better by saying "Firefox is coming soon, we'll email you when it's ready, click here" or some such, rather than putting a 'Download Chrome' link.
Thanks for sharing that. It's helpful to understand it from your perspective.
To share my perspective, outside Google:
A similar "error/ad" appears for other some other Google features. Such as trying to edit a picture in Google+ -- I've seen that example for so long now, it seems there's no intent ever to change it.
With each such example, the cumulative effect is to remind me of bad days in the past. My feeling isn't snarky, it's unease. I've seen stuff like Prodigy and CompuServe. I saw Microsoft pivot MSN 1.0 from being another such service, to instead be a web site.
And ever since, I've seen a succession of attempts to "put the open internet genie back in the bottle", including but not limited to Internet Explorer. So far these attempts have failed... but they haven't been from a company as powerful as Google.
So as someone who had loved Google for many years as a supporter of the open internet, it makes me unhappy and confused to see more examples of "Download Chrome!" as an error message in Firefox.
Having said all that, I'm happy to hear that you plan to address this at least for Inbox on Firefox.
Isn't this still fundamentally a problematic approach, though? Yes, being Chrome-only at launch was a bad idea (and the predictable response was that tweet and its thousands of retweets), but even if it were "Chrome + a notice about Firefox soon", and some time later it does run on Chrome and Firefox, that still leaves out all the other browsers!
I understand that you want 60fps flawless performance for your new product. But if you need to work in close cooperation with a browser vendor in order to enable that, then either you work closely with all the browser vendors - which might be possible in theory, but unlikely in practice (you don't have enough resources/interest, they don't have enough resources/interest) - or your product will not run on all browsers.
I honestly think this approach is not a good one. Alternatives which are at least less bad are
1. Let everyone run the app. Yes, some browsers will not hit 60fps. That's already true - even if they run it on your choice of browser, it may be on a slow machine. In this approach, some browsers will do less well, and one may be tempted to write something to the effect of "best experienced in X" but that's just getting back to that tweet (which while snarky, had a valid point). The best you can do here is proactively test on other browsers and file perf bugs on them. Reduced testcases can be filed before your product is public.
or
2. Rely less on browser optimizations. As your linked comment correctly mentions, how browsers do layout, create layers, upload them to the GPU, etc. differs between browsers. But the web is no longer limited to that. You can use WebGL - now present even in modern versions of IE and Safari, which means in all browsers - which gives you explicit control of how to create layers, when to upload to the GPU, etc. Of course, this is no guarantee of 60fps performance everywhere (nothing is, there are slow GPUs, etc.), but it has a far higher chance of running at a predictable and high frame rate than letting the browser's layout engine guess for you.
If we had the resources and didn't have constraints to co-launch with Android Lollipop and iOS, making it work on all the browsers out launch may have been feasible. We had bugs filed for most of the issues in our internal tracker, but we ran out of time before launch.
This is a problem at big companies like Google and probably Apple, is that we time major sets of released functionality for 'big impact' launches like Google I/O, or other media events, and the poor engineers are left working late into the night to meet the deadlines.
As for just rendering everything with WebGL or Canvas, we could do that (Google Sheets uses Canvas), but it kind of rubs me the wrong way, it doesn't feel "webby". You end up reimplementing tons of stuff: Bidi/I18N text handling, A11Y, system behaviors like text selection, clipboard, it just seems like you're making your own Flash runtime in JS, an alien hybrid that displaces the entire browser engine.
I hope apps like Inbox, and the new focus on mobile web performance by both Google and Mozilla can fix many of these layout/rasterization/composite hazards, either at the spec level (e.g. will-change), or at the implementation level, so much like there is a 'war' on JS performance, there can be a 'war' on layout and paint performance.
Long term, if we want the web to succeed and not be destroyed by mobile native apps, we need it to work -- as the web --, and not as simply opaque WebGL + JS, that's basically just a native app model with a different VM.
I get what you're saying about i18n in the alternate approach. It's harder to do those things there.
But isn't "designed for Chrome and Firefox" just a marginally better thing than "designed for Chrome"? It's still tailor-made for specific browsers, just 2 instead of 1. That's very un-"webby".
"... Inbox was designed to be 'mobile first' and use "'Material Design' animations, ... each new version of Chrome broke assumptions and broke animations we had running previously ..."
Cognitive friction.
"... Cognitive friction doesn't come from technology,
but from the people who control technology. The are
the masters because they know how to think in ways
that are sympathetic to silicon, and they imagine
that everyone thinks in the same way. ..."
Alan Cooper
"... The Web, as a platform for competing with mobile-style interfaces, is in a rickety state at the moment, where performance is non-deterministic not only across browsers, but across versions ..."
*"... As engineers, they think more about gears
than aout leather, so the interface to the human
user is expressed in those 'implementation' terms,
which is why I call products designed this way as
having an *implementation model*. ..."
Alan Cooper
For those who have read, "The Inmates are running the Asylum", SAMS, Al Cooper, 2003. Welcome to the dancing bear.
Umm, no, absolutely 100% backwards. Inbox was designed 100% as a goal directed product based on making the user more efficient and satisfying his needs, technical decisions came later. Product design was done by product managers and interaction designers by taking into account usability study feedback. We didn't just have engineers making stuff up cause they liked it, our users told use what their email pain points were. None of the internal presentations on Inbox's development by the designers mention the client implementation details. It's not the inmates running it, it's the people outside the asylum. The basic facts on the ground. Most access to email is going mobile. The design is to fit the users as they are coming.
If we designed product features based on technical decisions, we would not have had so much trouble, because we would have simply picked the lowest common denominator we could get away with to deliver the product and not pushed on sharp edges in the stack.
We take a design based on how the product must or should behave, the engineers do not run the show.
As an aside, I value the Web not because of implementation details (I'm not a particular fan of CSS or JS) but because of high level properties it has that native mobile apps do not have. That's why I'd rather see a return to web applications over time. Web apps produce positive net externalities that native silos do not.
These don't matter for the short term, or the individual, but they matter collectively over the long term for society. I may not be able to access content siloed to apps 30 years from now, but I'll probably still be able to find Web content indexed and stored on archive.org
I view developing for the Web as planting trees in a forest that is logged for value over time by society, by businesses, that extract value. We are caretakers of it, and should put back into it so that the those who come after us can derive value as well.
I'm not sure how developing a native iOS app, for example, will help our grandkids the way making an open ecosystem and publishing lots of content that can be freely indexed, mashed up, and repurposed will. I imagine future archaelogists will have a much easier time digging up value from the Web than trying to dig up App Stores for information.
Dart on Google App Engine via Managed VMs is about server side Dart, not client-side. (Client side Dart might have a runs-best-in-Chrome issue once Google integrates Dart VM into Chrome, but for now all practical web use is dart2js which only runs better on Chrome to the extent that V8 is better than other JS engines.)
App Engine is a subset of Cloud Platform services and is a platform-as-a-service (PaaS) offering – auto scaling, a scalable datastore, logging and other managed services and guarantees that aim to substantially reduce your DevOps.
By contrast, Compute Engine (and its closest competitor AWS) is more an infrastructure-as-a-service (IaaS) offering – provision your instances and run anything you'd like on it, as well as choose from available resources like load balancers and firewalls to meet your needs.
Managed VMs attempt to offer the best of both worlds: the managed scalability of App Engine with the choice and flexibility of Compute Engine. Through a Docker-based container approach, App Engine scalability guarantees are still kept while allowing you to run your own language, like Dart or Node.js, and services.
My Dart app, http://woven.co, has been running on Compute Engine, and I plan to move it to App Engine.