I've recently been in the market for a Java Web framework and have settled on Play. There are a lot of things I really like about it. I have no problem with the Play developers going "off the reservation" (as far as JEE goes).
I'm finding some things are a bit awkward, particularly around templates and routing. Also, the documentation while good is lacking in other parts.
Play is a very prescriptive framework but only on a couple of choices do I have an issue.
The first is about transactions. Each HTTP request is a transaction. Generally I'm fine with this but you need a way to override or control this (IMHO).
Second, I like the mixin approach with controllers (@With(...)). I wish there were something similar with models.
Third, and this is the biggest, it doesn't integrate with Maven, which is something I'm very tempted to fork it for. There's no reason it can't be a Maven plugin/archetype, use Maven for dependencies (directly, rather than tortuous syntax in dependencies.yml), etc. Some guy has an alpha version of something like this already when I last looked.
Fourth, while I don't necessarily mind that controllers are static (as a way of indicating they're stateless it seems), that's actually an issue for parallellizing tests.
Play 2.0 is moving even further away from this and using SBT, which I've never used but heard mixed things about.
I can't speak to many of the specifics of this post (eg I only use Play on Linux).
Whatever (valid) criticisms you can level against Play I'd still far and away prefer to use it over any other Java Web framework I have used or know about.
I've had to work with Play lately, and I find it quite delightful. It's butchering java as a language (or at least what's considered good practice), but since it works so well, you tend to overlook that.
I'm also a fan of maven. There's nothing better than checking out a huge project with a great deal of dependencies and just have it build perfectly with a couple of keystrokes.
Seems like the play developers (Guillaume et co) went for a clean break with legacy java though. And if you want that, you can't depend on maven.
Only had a cursory look at 2.0 so far, but moving further into Scala territory and adopting SBT seems like an interesting move to me. Play is becoming a great vehicle for Web developers who want to ease into a Scala future.
The only good idea in Maven is the standardized repository layout, the rest is useless ceremony although much of it is because of retarded J2EE application servers.
Buildr is a welcome improvement, at least there is no XML on sight.
- We have had a fine experience not using Play's ORM support at all -- we have a Plugin that handles transactions for us, and we use non-Play-integrated persistence. It still works great.
- All of the Controller member variables are ThreadLocals, which is how Play handles concurrency. It seems like multi-threaded tests should work just as multi-threaded requests do. I'm curious what problem you're having with parallelizing tests?
SBT is really just Maven without as much XML boilerplate. The syntax might be a little different but it integrates perfectly with Maven repositories. For my SBT projects I pull in a bunch of maven projects and they interact seamlessly. also it took the huge XML configuration file and replaced it with a few lines of configuration code. The only downside to being a different tool is with some of the plugins. I know maven has a much better plugin landscape then SBT does at this point, but SBT does have the ability to have the same functionality.
I like Play overall but two things concern me, both related: 1) they are moving toward Scala and 2) they are adopting sbt.
This looks to me more like a "picking a shiny toy" move than something that is being done to benefit users (Scala is still quite a marginal language and sbt is very controversial, even among Scala enthusiasts).
I was surprised to see Typesafe throw its weight behind Play.
While I understand they need a web framework in their stack, and Play is more RoR-ish than Lift (which makes it more marketable I think), I dunno, some of the gymnastics Play 1.x used to pull off its magic seemed a little over the top to me.
Like using bytecode rewriting to add language-level features like properties. Yes, Java sucks, but we who use it understand that and accept its imperfections. I'd rather fix Java proper (or use Scala) than have a web framework think that it's their job to somehow fix a lackluster language.
Just seems like it'd lead to a lot of bloat/code in the framework that isn't "just serve this webapp as simply as possible".
That being said, what Play got right did look really awesome (the compilation error reporting in particular).
Anyway, will be interesting to see what the Play 2.x release will look like, given it looks like Typesafe is having some amount of involvement in it?
All the byte-code rewriting has been dropped in Play 2.x. In fact Play 2.x is vastly different beast to Play 1.x, and much superior in my opinion. It's very straightforward, non-magic, and gets stuff done.
If that shop is stuck with debugging into a method in HashMap, then they aren't qualified to evaluate whether or not it's in the framework. That said, the developer community needs to address the problem though education, if it's user error (most likely) or come up with enough manpower to be more active in support.
After having seen their video, I have to say that the Play! framework folks seem very insightful in their design goals. They hit the nail right on the head in terms of the major usability problems of most Java Web AppServer software.
We use play in all our projects and hit some bugs as well. (Like different versions of classes being in jars and in code -> signature mismatch)
It's a great project and no software is without bugs.
I think that any platform has some gotchas especially a 1.x version, however in my experiences with it it was fast and less boilerplate than a normal Java project and we didn't have any of these issues rolling out locally and to Google AppEngine.
I'd still only choose it if I had to use Java though.
If you can roll with Django you can roll with Play. Although, I always prefer a python microframework such as flask nowadays.
I also see how Play is now part of typesafe, so that could get better or worse, but it is the right direction for Java web apps to go.
I've worked quite a lot with both Java and Python and with Play and Django, and before I tried Play I would've agreed with you.
The lightweight nature of Play combined with the speed of the jvm (compared to Python), static typing and IDE support, makes it a winner in my book. I like vim, but wherever I click in a Play project (including in routes, property files and templates), I always end up where I need to be.
The author wrote the Play Framework Cookbook and so it's a presentation from someone that can be considered a bit of an authority on Play (and not someone with an axe to grind). However, the presentation covers a bit of what worries me about Play. Play rewrites the bytecode for things like emulating properties and allowing you to pass whatever you want to the templates and have it automatically name the variables based on their local names in the calling method.
For me, the reason to use Java over Python or Ruby would be to move away from some of the behavior used in those languages that can make debugging and static analysis harder.
I've used Play for some hobby stuff (and kept up on Play 2.0). Play 2.0 seems to remove the former system for passing variables to the templates in favor of a system where templates accept certain parameters with types (a welcome change for me). However, at least the last time I played with it, Eclipse couldn't tell me what variables and types the template was expecting.
I guess Play leaves me feeling like it brings the magic to me without as much ease that Python and Ruby offer (or, at least, without as much familiarity). That might be changing - I think that the changes in the templating for 2.0 show a clear move in that direction. I really wish them well since it's always good to have more good options for development and there's a lot of Java already out there to leverage.
Valid points about bug fixes and outstanding issues. I think the Play 2 announcement was met with a mixture of trepidation and excitement from the community. If support for 1.2.x was to decrease further then I'd be concerned.
For what it's worth, I've been developing a Play (now 1.2.4) app for a year now and it's been a blast. Sometimes you can see the framework's immaturity, but I haven't run into any showstoppers like the author.
It is a very well written opinion piece that gives his ( valid ) reasons for not using play any more. That being said he doesn't say anything bad about the framework or the idealogy behind it and in fact laments having to move away from its ease of use.
He does bring up the point that Play 1.X is essentially out to pasture at this point. Thats not to say typesafe can't come back and once 2.0 is settled and fix the outstanding bugs , but as far as the community is concerned most have moved on to 2.0 or are aren't experiencing these issues.
For my part I have never ran into any serious bugs on my Play apps and services. I would note that I don't have any very large scale deployment, mostly just internal applications. I am however using Play 2.0 as the foundation for my next big project and so far even the RC have proven stable and functional.
Sounds like the author doesn't understand concurrency or how to set up the unit tests from an IDE with a version of play framework which is only weeks away from being completely replaced.
As someone marginally involved with the Play project, I would like to point out that, for better or for worse, Play 2.0 is a complete reimplementation of the original design, with a new API and a totally new codebase. Play 2.0 core is now written in Scala (but also exposes a Java API) and relies much less on bytecode manipulation.
DropWizard, huh? I'd be really careful about anything that overly simplifies opening up CRUD functionality via REST. Be sure to spend some of that time you save looking into the security of your service. Easy generation of RESTful services has bitten many a Rails developer, for example.
Play! is one the best things that has happened to java in the last 10 years!
From the post I understood that she was pissed because they did not accept her fix on testing the app from the IDE.
The second point was that there is a bug in the framework that they could not find... it seems though as the bug is in their application. (concurrent modification of a Map is a problem in java not Play!)
It's a pity to give up such a great framework for silly reasons.
Every time I have to switch to another framework (especially jsf) I want to shoot myself.
Anyway... good luck finding another good java framework.
I'm curious as to specific code examples where the author is having problems. I'm thinking about using Play! or Lift for a pet web app using Scala. Does anyone have experience using either?
I've been using Play 2 recently. I would definitely recommend it over Play 1 for new apps. I've had a very good experience with it. It's pretty easy to get started if you want to try it out and the community on the Google Groups mailing list is very helpful if you run into difficulty. Let me know if you have any more specific questions.
I use Play! for my hobby projects and Lift for work. Coming from MVC lift is very bizarre. Having html and javascript inside my scala classes gives me a queasy feeling.
I also think Lift is dead. The creator of the project left, the documentation isn't great, View first model isn't popular, very little activity in #lift, very little activity in the lift framework google group. Not good signs.
> Having html and javascript inside my scala classes gives me a queasy feeling.
Yeah that's an interesting tradeoff. I like it because it helps keep code completely out of the html templates, which is a design goal of lift. I tend to work with designers who do HTML/CSS only, not javascript, and I do the backend Scala and frontend Javascript, so that separation of roles works. But if your dev team (as most big ones do) consists of frontend HMTL/CSS designers, frontend Javascript UX devs, and backend devs, then that architecture may not be optimal.
> I also think Lift is dead.
Far from it.
> The creator of the project left
Not exactly:
1. He's still a core committer,
2. Advised the founders of Lift Co. [1] - commercial support/SLA for lift projects,
3. Is active on the Lift group [2] (the activity takes place there and at Assembla [3], not Twitter)
4. Continues to present about [4], blog about [5][6], and advocate Lift [7].
5. Even holds personal office hours every week in San Francisco to support Lift devs, for free. He's just not doing direct commercial consulting on Lift anymore, others have grabbed that torch and are running with it, primarily in SF, London, and India.
> the documentation isn't great,
True, but improving. The new book from Manning, Lift In Action [8] is excellent. You can buy it from Manning, or obtain it by other means. There are completely free options as well [9].
> View first model isn't popular
Popular is not necessarily better [10], unfortunately. However, if you grok view first and still believe MVC is better, Lift also includes an MVC version [11] for you.
> very little activity in #lift,
The meat of the discussion is on the google group [2], not Twitter. (And it's #liftweb anyway, but admittedly not much difference there either).
> very little activity in the lift framework google group.
Maybe you're comparing to a mega project like Rails, or maybe you accidentally mistook the stickies at the top for the actual group activity, but there's more than 'very little activity' in the group.
Granted, Lift isn't without issues. Getting started via Maven is painful (Pro-tip: use SBT + Lifty [12]). It has a non-trivial learning curve (I actually had to learn Haskell in order to grok Scala in order to use Lift - all worth it - though people coming from the ML family may have an easier ramp-up); has to deal with Scala's versioning and dependency issues (which David Pollack ticked off the Scala community by vocally complaining about, and which SBT does a decent job of mitigating), and does most things differently than the norm.
However, it's a superb piece of technology more than worth the effort to learn. It's more inherently secure than any other framework out there, enables a simple architecture (3-tier only, web/app/db, no caching layers), is relatively easily scalable (just add more instances to whichever layer you need to scale), does Ajax and Comet better than anything else, is very productive once you know it (minimal code, no testing necessary for classes of errors the type system or compiler catch), and lots of other excellent stuff [13].
It's evolving, maybe not at the pace of Rails, but it's starting with a far stronger technical architecture, runtime, and Java ecosystem interopability.
He didn't just leave Lift, he left Scala altogether. He's now working in Haskell. Check his latest blog posts for the details (and the scala-tools.org controversy).
I would not use Play 1.x for Scala applications; its non-blocking support is complete crap (and I was pretty displeased to hear, when I asked in their IRC channel, that it would never be in 1.x, oh and by the way to use 2.x you'll need to rewrite your app).
You don't need to totally rewrite your app, but it's not backwards compatible either. The biggest change is that they've switched the default views to be written in Scala instead of Groovy. But if you'd still like to write the views in Groovy or are porting an app from 1.x, there are a couple of guys in the community who have written plugins to support writing your views in Groovy.
Ask about Lift on the google group. They're honest about advising whether Lift is a good fit for a particular project or not. They want success stories, and tend to advise against Lift for any project that could have serious issues ramping up on Lift:
I'm switching to Play 2.0 (which is yet in RC3, not yet released) and I think it's a great new framework and I understand them wanting to dedicate more resources to the new framework.
Play made it very very easy for me to target a jar from php. The php-java bridge at the time would start and stop the jvm with every request (I dont know how it stands up today). I did some research and was able to stand up a play server whose routes reflected classes/methods in the jar in about 20 mins.
Author makes great points; refusing to accept a (tested) minor patch to how tests are run should be a no-brainer because of the low threat to existing systems. One thing that was not clear to me (being unfamiliar with Play or Java at large) was wether the HashMap bug was intrinsic to Play or to Play's use of a Java library...
I'm speculating because I've never used play. But I have seen threads hung inside hashmap methods, eating 100% cpu when using a HashMap (which is not thread-safe) concurrently.
I'm positive they don't. Anyone who's ever written anything substantial for a Java App server should know that HashMap isn't thread safe. Magic fix: ConcurrentHashMap....better still write your own.
EDIT: I am also referring to the poster. I'm a fan of Play and echo many of the sentiments others are listing out.
Magic fix: ConcurrentHashMap....better still write your own.
I hope this is a joke. I know plenty of very good Java programmers, but only one or two that I would ever trust to write a concurrent hash map class that matched java.util.concurrent.ConcurrentHashMap's functionality (to say nothing of performance), let alone improve on it.
The fix you're proposing is innadequate for a lot of instances.
A concurrent hashmap is only safe in regards to its own internals, however if you're operating with keys and values that are not thread-safe, then it can still deadlock on get() or other operations.
This is why this model of threading is hard, because it is not composable.
Also, threadsafe data-structures have terrible performance characteristics, unless the implementation is lockfree, and lockfree implementations are hard to get right. Therefore I don't blame devs that use HashMaps, as sometimes is better to put locks in other places, or to make sure that the variables are local to a thread.
Even without multithreading, mutating the value of a key (such that hashCode() changes) after putting it in the map entry won't return the same entry. This is a HashMap basic, not just something special jujitsu you need to learn with multithreading.
I think blaming someone for using a non-thread-safe data structure without sufficient explicit locking is justifiable.
Just FYI, ConcurrentHashMap is indeed lockfree (although it's got a gargantuan memory footprint). Wrapping your HashMap with Collections.synchronizedMap gives you a blocking threadsafe HashMap.
I looked into Lift. Steep learning curve. It would be ok for a single person, but a large team coming up to speed on Lift/Scala looks hard. Then I decided to stick with what I know best Ruby/Javascript/Node.
This confirms my bias against frameworks in general and in fact against reusing more code than necessary. And no, I would not write an encryption library myself or re-implement the HTTP protocol.
But the disparity between what web frameworks actually do for me in any particular application and the amount of layers and dependencies they introduce boggles my mind.
Play is really not a large amount of code though and it's all in one source tree. It generally doesn't have a lot of "layers" compared to even something like Tomcat; the stack just doesn't get as deep. I've had an easy time digging in to the Play source code when needed.
Just one experience fwiw. Sure, it's still a framework.
Agreed. I fixed a handful of bugs in Play 2 while it was still young and immature and didn't have much difficulty tracking them down despite not knowing any Scala.
While OP has met some problems, I don't see how these would drive him away. I'm curious what framework he's moving to.
OP is looking for a silver bullet and there ain't one. Play! is an open source project, so the developers didn't have time to deal with his patch and he got upset. People forgot OS devs don't get paid for their work. Be a little bit more appreciative.
Having spent a lot of time looking at/comparing/building Web 1.0-style server-side frameworks, I've since switched to GWT and really like the client-side model. It'd be hard for me to go back to munging HTML/data back/forth from the stateless server.
Of course, GWT is not perfect. DevMode needs to be faster, and they need to integrate scala-gwt. :-)
I wrote a Backbone-like framework to deal with some of the boilerplate (http://www.tessell.org), especially if you're doing MVP. I don't quite have a solid DTO story down yet, but overall I like it.
If I had to use a server-side framework again, I really liked Click (http://click.apache.org), as it is component based, but has source code you can actually read (vs. both Tapestry and Wicket which are too big/magical IMHO).
I don't know about 'best', but I certainly can tell you what we've done for our latest project, Voost. https://www.voo.st/
Google App Engine cause we hate doing sysadmin work. Objectify. Cambridge Template Engine with JEXL. RestEasy/HtmlEasy. Jackson. Guice to bind it all together. Front end is heavily CoffeeScript, Handlebars, Jquery, LabJS, customized bootstrap and various other libraries. We've got a credits page full of links to the above projects... https://www.voo.st/about#credits
Overall, I'm pretty happy with this choice. We've had to do some weird and semi-complicated stuff to integrate it all together, but at this point, it works quite well.
I've never gotten a chance to try out Play, I'm no longer doing any JVM stuff... but back when I was a Java guy, my favorite java web framework was Stripes: http://www.stripesframework.org/display/stripes/Home
I've never really understood why Stripes hasn't "caught on." It's simplicity is wonderful.
Stripes was pretty much subsumed by JAX-RS. There are a few rough edges surrounding html rendering (thus microframeworks like Htmleasy) but for the most part, Stripes is no longer necessary.
Most of the history of Java web frameworks focused on ways to make form processing easier. Nobody (sane) does html form processing anymore. The "framework" needs to be little more than a way to render html templates and an rpc mechanism.
I was doing basically this before moving to Play 2. I just found there was a lot you had to decide for yourself. You had to pick a persistence layer, roll your own xsrf protection, do your own integration of a JS compiler/minifier, etc. I had something pretty good working, but got tired of having to write the code for every new technology integration myself. The thing I like about Play 2 is that I have an entire community helping me out with these things.
Has Wicket changed to not require so much code? I wrote a handful of apps using Wicket 1.2, 1.3 and 1.4 before realizing I was writing thousands of lines of code to update UI models and backend models that shouldn't be taxing me so much.
Used Play for a few projects, but don't like that 2.0 is more or less getting rewritten... don't like Spring... not sure what is left (JEE6 and JSF2? God I don't want to use JSF again unless it changed significantly)
Re Wicket: have a look at PropertyModel and CompoundPropertyModel. You'll lose a bit of type safety, but it will seriously cut down on boilerplate code. In particular it helps map wicket:ids to your Java bean hierarchies by following simple conventions.
Somehow I managed to use Wicket for well over a year before realizing the benefit to this approach.
IFF latency isn't an issue, just chop the front-end and expose services from the Java container (where it shines and makes perfect sense) and use a suitable non-Java front-end tech to create the presentation layer.
cmiiw, but imho play is too similar with others web framework, like rails.
i'd prefer scala community to push lift instead.
lift has some great ideas, and has many things to improve.
I'm finding some things are a bit awkward, particularly around templates and routing. Also, the documentation while good is lacking in other parts.
Play is a very prescriptive framework but only on a couple of choices do I have an issue.
The first is about transactions. Each HTTP request is a transaction. Generally I'm fine with this but you need a way to override or control this (IMHO).
Second, I like the mixin approach with controllers (@With(...)). I wish there were something similar with models.
Third, and this is the biggest, it doesn't integrate with Maven, which is something I'm very tempted to fork it for. There's no reason it can't be a Maven plugin/archetype, use Maven for dependencies (directly, rather than tortuous syntax in dependencies.yml), etc. Some guy has an alpha version of something like this already when I last looked.
Fourth, while I don't necessarily mind that controllers are static (as a way of indicating they're stateless it seems), that's actually an issue for parallellizing tests.
Play 2.0 is moving even further away from this and using SBT, which I've never used but heard mixed things about.
I can't speak to many of the specifics of this post (eg I only use Play on Linux).
Whatever (valid) criticisms you can level against Play I'd still far and away prefer to use it over any other Java Web framework I have used or know about.