I upgraded from saxon transformer version 9 to 11.3. My application is facing memory leak issue.
Java spring boot application
Application is multi threaded
The application has 10 different transformer and process different xsl
The memory occupied grows over time and takes up all available memory
Tried to reuse transformer and to use new transformer every time we transform.(No luck)
As Martin says, this kind of question is more suited to the Saxonica support forums than here. The reason is that it probably involves an iterative process of investigation, which doesn't fit well with the StackOverflow question/answer format (where, if I don't immediately have the answer, someone is likely to flag my response as "not an answer").
The term "memory leak" is rather inaccurate when applied to Java. A memory leak is when objects that are no longer referenced are not releasing their memory, but the problem here is (almost certainly) that objects are still referenced when you don't want them to be.
To investigate the problem you need to get a heap dump and find out where the problem lies. It may be obvious that there are millions of objects of a particular class that shouldn't be there. A heap dump analyser will enable you to find out where these objects are referenced from, which will explain why they are not being garbage-collected.
You could proceed either by getting a heap dump and sending it to Saxonica (if you can't analyse it yourself), or by creating a repro of the failing application and sending that. My guess, given that it's a fairly complex application, is that first approach is going to be easier for you.
We try to draw some lines around the level of support we're prepared to provide. One of those boundaries is that we won't attempt to run a repro that has complex dependencies, e.g. that requires third-party products to be installed. That would include Spring.
Note: different people adopt different strategies for problem solving, but personally, when I see something like this that worked in release X and doesn't work in release Y, I will usually investigate under release Y from first principles, rather than exploring what has changed. But there are exceptions, for example we know that Saxon 11 uses a completely new approach to catalog-bsed URI resolution and that will sometimes give clues as to where it's worth looking. So: it probably doesn't matter what release X was, but you should be aware that "Saxon 9" describes a series of ten major releases from 9.0 in 2008 to 9.9 in 2020, and it therefore isn't a very useful definition of your baseline.
Can somebody give me some insights what is the difference between Netflix Zuul version 1.x.x and new version 2.x.x?
Seems that both product line are maintained.
And version 2 is using Guice for DI and there are some difference in Filter implementation. ??
I got really nice answer from #NiteshKant on GitHub from Netflix:
Unfortunately there is no documentation about the motivations for 2.x and what it changes. I am intending to put together something in the coming weeks when time permits. As of today, I hope the following suffices:
What is 2.x?
2.x intends to move zuul from current synchronous execution model to a top to bottom asynchronous processing model. This includes using non-blocking I/O (practically RxNetty as the networking library) and application processing semantics (RxJava as the asynchronous library)
Why 2.x?
Intentionally staying away from proofs and benchmarks, the motivation for 2.x (essentially moving to an async model) is to have better resilience, control and performance characteristics for all applications inside Netflix.
Status
The current status of 2.x is snapshot. We are currently testing the new filter model (async) with blocking I/O inside netflix. Once we are comfortable with this change, we will be testing the changes with non-blocking I/O. After that we will be publishing release candidate and release artifacts.
Should you adopt 2.x now?
2.x is really very bleeding edge (sorry for the cliche) so we will be changing APIs, deployment models and implementations. So, unless you are prepared to take the burden of keeping up with these changes, I would recommend waiting a while.
Also, 2.x comes with lots of changes in usage, so most likely you will have to change all your existing filters, if any. This can be a big task depending on the current usage. So, it is your decision on that front in terms of ROI.
There are more related links to the subject on Zuul 2.x:
https://github.com/Netflix/zuul/issues/121
https://github.com/Netflix/zuul/issues/106
https://github.com/Netflix/zuul/issues/139
https://github.com/Netflix/zuul/issues/130
I am starting out a new project that involves the use of JSF 2.0.
From my initial reading, the Mojarra and Apache Implementation of the
project covers the basic components that you will need.
But I know that user's would seek gui with better presentation such as
panel tab, accordion, slider etc... Currently, there are other implementation that I am seeing, the Primefaces, RichFaces and Icefaces.
But I cant find a good article that discusses which among the three are the best.
I have used Spring MVC before but I use JqueryUI for those widget.
Now that I am into component based framework, I would like to use the best JSF Implementation.
I would like to know metrics such as performance/interoperability/ease of use/support.
Sorry if my question might be vague but I would like to hear comments before I select my JSF Vendor Implementation.
Thanks.
I happily use Primefaces as it is by far the most rich set of open source JSF2 controls out there, but they can be infuriatingly buggy at times. It is best to operate under the assumption that component X will not work correctly in a dialog without heavy tinkering.
I would avoid Primefaces if you operate in a development environment with strict UI design requirements as getting everything exactly the way you would like it to look and operate may not be a possibility.
Further I would avoid Primefaces if you are not comfortable with JSF, HTML, JQuery, JavaScript and CSS as you will need a good bit of JQuery trickery to work around the bugs that crop up.
But on that note, I haven't run into a problem yet that a couple lines of custom Javascript haven't fixed for me, and I have one of the most feature rich applications I have ever wrote in the shortest amount of time.
The speed of development is very fast in this area, and any article gets outdated quickly. I used Primefaces for a new project almost a year ago, because at that time it was the only one that was fully compatible with JSF 2 (both Icefacves and Richfaces have had JSF 2 compatible releases in the meantime).
Primefaces has a lot of powerful components that automatically use AJAX, and even more were added in version 3. Unfortunately this focus on new features led to a lot of bugs, but the developers said they would focus on bugfixing after release 3; I can't say anything about the current status since I left the project after 3 months.
There is one thing against Icefaces: a lot of components and functionality are only available in the Enterprise version, which is commercial, not free (but that might as well be a good thing since you get support etc. if your project can afford it).
Why not play with all three libs for a short time, build a simple project and see how you are getting along with either of them. My personal taste prefers Primefaces, but I haven't tried Richfaces since it turned JSF 2 ready.
I have successfully used JBoss RichFaces on a large online B2B store. RichFaces is a quite good framework for building webapps Web 2.0 style, and have easy to use tags that help you develop features faster.
I do not have any metrics regarding performance between RichFaces and IceFaces/Primefaces, but the ease of development should be approximately the same. All three frameworks have similar components, and they are all working towards more and more logic on the client via JavaScript.
At the current state of the JSF libraries, I am fairly sure that you will be happy with whichever framework your choose. IMO RichFaces and IceFaces are the two frameworks that have been around the longest, and i would put my bet on one of these two. IIRC both frameworks have key developers in the JSF design group as well.
As a general rule of thumb, these framework should work interoperably, but I wouldn't mix and match between them. The frameworks are really ment to be used on their own.
I have a legacy desktop accounting application developed using Delphi 5 & Paradox, which I intend to migrate to a web based Silverlight (for the sake of UX) application with SQL Server.
Can anybody suggest a way to implement this quickly?
I know this is a very open-ended question and I am not looking for concrete answers. Instead opinion/experiences from SO users.
My main concern is about migration approach, possible architecture and design patterns (for SL I know of MVVM) implementation.
Quickly? That's what every manager wants, but I doubt it.
You have fundamentally different models of UIs, and different programming languages.
Unless these applications are small, it is unlikely that will be able to convert them by hand in any short period of time (or even by yourself as it appears the OP implies "I intend").
Gartner Group has analyzed manual migrations, and suggests if everything is "similar" the actual conversion rate is ~~ 150 lines/day, which is possible because you are translating more or less directly from a working, debugged application. (Just how big is the application in SLOC?) So, if you have 75,000 lines of code, you're looking at 500 man-days minimum. You might make the case that Delphi as programming langauges and C# are similar. You cannot reasonably make that case for the Delphi UI and Silverlight, so this estimate is a lower bound.
There are those that say, "just throw it away and recode it from scratch". Unless your productivity exceeds 150 debugged lines of code per day [classic software engineering texts will tell you it is much smaller than this] this will take you even longer. Usually it fails because you end up forgetting what features exist in the current program, and rediscover them late in development or worse after an attempted reployment. Usually what happens is the old application continues to evolve while the new one is being built (remember, you're 500 man-days away from the new one minimum!) and the new one has to play catchup with these changes. If the application has any serious scale (e.g., a million lines) this often prevents the new one from ever being servicable. Another way to think about this, "how long did it take to build the original application?", and "why should building a replacement be enormously easier?". YMMV, if you can work miracles.
My very biased opinion (I build langauge translation tools) is that one of the most practical ways to do this is automated translation. This has its costs, too; they aren't off-the-shelf items no matter what somebody tells you. You have set up the translator, and that also takes a lot of energy, but that energy is proportional to the size of the language and (UI) library features used, rather than the application size, so it is far more effective as the program gets large. This is still on the order of hundreds of man-days to code and test for just for the langauge translation part. The difference is that once set up, you can apply it to the existing application of whatever size in whatever state it happens to be in. There's more complications than this, but this approach overcomes the "can't catch up" problem of manual conversions, and the "can't get enough coders to manually translate it".
For more details, see my answer on how to translate between languages.
If your application is relatively small, there are IMHO no good answers. Hand translation or recoding are likely your only (ugly) choices.
My suggestion would be to create "value add" extras and updates to your application using Silverlight as and when the need for extra functionality comes up until you've got something resembling a full product.
To me developing Silverlight seems to take a very long time and the UX for a business application isn't massively improved over say ASP.NET Ajax (if the Ajax is done properly). I imagine if you were to sit down today and completely re-write a decent size application in Silverlight then Silverlight would be end of life before your development is completed (unless you threw a massive team at it of course)
If your business logic is well separated from the UI, you can start with "porting" your code to Delphi Prism rather than C#. This offers shorter migration path. If your business logic is tightly coupled with UI (as it happened frequently 10-15 years ago), then rewriting everything from scratch could be a better idea.
And once you have all the code in Pascal up and running, rewriting it in C# (if you need it at the end) is almost trivial with help of decompiler.
Yesterday I saw a presentation on Java Server Faces 2.0 which looked truly impressive, even though I am currently a happy ASP.NET MVC / jQuery developer. What I liked most about JSF was the huge amount of AJAX-Enabled UI components which seem to make development much faster than with ASP.NET MVC, especially on AJAX-heavy sites. Integration testing looked very nice too.
Since the presentation only emphasized the advantages of JSF, I'd like to hear about the other side as well.
So my questions are:
What are the main disadvantages of Java Server Faces 2.0?
What might make a JSF developer consider using ASP.NET MVC instead of JSF?
JSF 2.0 disadvantages? Honestly, apart from the relative steep learning curve when you don't have a solid background knowledge about basic Web Development (HTML/CSS/JS, server side versus client side, etc) and the basic Java Servlet API (request/response/session, forwarding/redirecting, etc), no serious disadvantages comes to mind. JSF in its current release still needs to get rid of the negative image it gained during the early ages, during which there were several serious disadvantages.
JSF 1.0 (March 2004)
This was the initial release. It was cluttered with bugs in both the core and performance areas you don't want to know about. Your webapplication didn't always work as you'd intuitively expect. You as developer would run hard away crying.
JSF 1.1 (May 2004)
This was the bugfix release. The performance was still not much improved. There was also one major disadvantage: you can't inline HTML in the JSF page flawlessly. All plain vanilla HTML get rendered before the JSF component tree. You need to wrap all plain vanilla in <f:verbatim> tags so that they get included in the JSF component tree. Although this was as per the specification, this has received a lot of criticism. See also a.o. JSF/Facelets: why is it not a good idea to mix JSF/Facelets with HTML tags?
JSF 1.2 (May 2006)
This was the first release of the new JSF development team lead by Ryan Lubke. The new team did a lot of great work. There were also changes in the spec. The major change was the improvement of the view handling. This not only fully detached JSF from JSP, so one could use a different view technology than JSP, but it also allowed developers to inline plain vanilla HTML in the JSF page without hassling with <f:verbatim> tags. Another major focus of the new team was improving the performance. During the lifetime of the Sun JSF Reference Implementation 1.2 (which was codenamed Mojarra since build 1.2_08, around 2008), practically every build got shipped with (major) performance improvements next to the usual (minor) bugfixes.
The only serious disadvantage of JSF 1.x (including 1.2) is the lack of a scope in between the request and session scope, the so-called conversation scope. This forced developers to hassle with hidden input elements, unnecessary DB queries and/or abusing the session scope whenever one want to retain the initial model data in the subsequent request in order to successfully process validations, conversions, model changes and action invocations in the more complex webapplications. The pain could be softened by adopting a 3rd party library which retains the necessary data in the subsequent request like MyFaces Tomahawk <t:saveState> component, JBoss Seam conversation scope and MyFaces Orchestra conversation framework.
Another disadvantage for HTML/CSS purists is that JSF uses the colon : as ID separator character to ensure uniqueness of the HTML element id in the generated HTML output, especially when a component is reused more than once in the view (templating, iterating components, etc). Because this is an illegal character in CSS identifiers, you would need to use the \ to escape the colon in CSS selectors, resulting in ugly and odd-looking selectors like #formId\:fieldId {} or even #formId\3A fieldId {}. See also How to use JSF generated HTML element ID with colon ":" in CSS selectors? However, if you're not a purist, read also By default, JSF generates unusable ids, which are incompatible with css part of web standards.
Also, JSF 1.x didn't ship with Ajax facilities out of the box. Not really a technical disadvantage, but due to the Web 2.0 hype during that period, it became a functional disadvantage. Exadel was early to introduce Ajax4jsf, which was thoroughly developed during the years and became the core part of JBoss RichFaces component library. Another component libraries were shipped with builtin Ajax powers as well, the well known one being ICEfaces.
About halfway the JSF 1.2 lifetime, a new XML based view technology was introduced: Facelets. This offered enormous advantages above JSP, especially in the area of templating.
JSF 2.0 (June 2009)
This was the second major release, with Ajax as buzzword. There were a lot of technical and functional changes. JSP is replaced by Facelets as the default view technology and Facelets was expanded with capabilities to create custom components using pure XML (the so-called composite components). See also Why Facelets is preferred over JSP as the view definition language from JSF2.0 onwards?
Ajax powers were introduced in flavor of the <f:ajax> component which has much similarities with Ajax4jsf. Annotations and convention-over-configuration enhancements were introduced to kill the verbose faces-config.xml file as much as possible. Also, the default naming container ID separator character : became configurable, so HTML/CSS purists could breathe relieved. All you need to do is to define it as init-param in web.xml with the name javax.faces.SEPARATOR_CHAR and ensuring that you aren't using the character yourself anywhere in client ID's, such as -.
Last but not least, a new scope was introduced, the view scope. It eliminated another major JSF 1.x disadvantage as described before. You just declare the bean #ViewScoped to enable the conversation scope without hassling all ways to retain the data in subsequent (conversational) requests. A #ViewScoped bean will live as long as you're subsequently submitting and navigating to the same view (independently of the opened browser tab/window!), either synchronously or asynchronously (Ajax). See also Difference between View and Request scope in managed beans and How to choose the right bean scope?
Although practically all disadvantages of JSF 1.x were eliminated, there are JSF 2.0 specific bugs which might become a showstopper. The #ViewScoped fails in tag handlers due to a chicken-egg issue in partial state saving. This is fixed in JSF 2.2 and backported in Mojarra 2.1.18. Also passing custom attributes like the HTML5 data-xxx is not supported. This is fixed in JSF 2.2 by new passthrough elements/attributes feature. Further the JSF implementation Mojarra has its own set of issues. Relatively a lot of them are related to the sometimes unintuitive behaviour of <ui:repeat>, the new partial state saving implementation and the poorly implemented flash scope. Most of them are fixed in a Mojarra 2.2.x version.
Around the JSF 2.0 time, PrimeFaces was introduced, based on jQuery and jQuery UI. It became the most popular JSF component library.
JSF 2.2 (May 2013)
With the introduction of JSF 2.2, HTML5 was used as buzzword even though this was technically just supported in all older JSF versions. See also JavaServer Faces 2.2 and HTML5 support, why is XHTML still being used. Most important new JSF 2.2 feature is the support for custom component attributes, hereby opening a world of possibilities, such as custom tableless radio button groups.
Apart from implementation specific bugs and some "annoying little things" such as inability to inject an EJB in a validator/converter (already fixed in JSF 2.3), there are not really major disadvantages in the JSF 2.2 specification.
Component based MVC vs Request based MVC
Some may opt that the major disadvantage of JSF is that it allows very little fine-grained control over the generated HTML/CSS/JS. That's not JSF's own, that's just because it's a component based MVC framework, not a request (action) based MVC framework. If a high degree of controlling the HTML/CSS/JS is your major requirement when considering a MVC framework, then you should already not be looking at a component based MVC framework, but at a request based MVC framework like Spring MVC. You only need to take into account that you'll have to write all that HTML/CSS/JS boilerplate yourself. See also Difference between Request MVC and Component MVC.
See also:
What is the difference between JSF, Servlet and JSP? (just to understand the basics)
Using JSF to develop tableless CSS layouts (another myth about JSF)
JSF vs plain HTML/CSS/JS/jQuery (when JSF is the wrong choice)
Design patterns in web applications (illustrates the ideology behind MVC)
After 5 years of working with JSF, I think that I can add my 2 cents.
Two major JSF drawbacks:
Big learning curve. JSF is complex, that's just true.
Its component nature. Component-based framework tries to hide the true nature of the Web, which comes with a huge amount of complications and disasters (like not supporting GET in JSF within almost 5 years).
IMHO hiding HTTP Request/Response from the developer is an enormous mistake. From my experience, every component-based framework adds abstraction to the Web development, and that abstraction results in unnecessary overhead and higher complexity.
And minor drawbacks that come to my mind:
By default ID of the object is composed of its parents' ids, for example form1:button1.
No easy way to comment-out incorrect page's fragment. Tag <ui:remove> needs syntactically correct content which is parsed anyway.
Low quality 3rd party components which e.g. don't check isRendered() inside processXxx() method before continuing.
Incorporating LESS & Sencha is hard.
Doesn't play well with REST.
Not so easy for UX designers, because ready-to-use components have their own CSS styles, that need to be overwritten.
Don't get me wrong. As a component framework JSF in version 2 is really good, but it's still component-based, and always will be...
Please take a look at the low popularity of Tapestry, Wicket and low enthusiasm of experienced JSF developers (what is even more meaningful).
And for contrast, take a look at the success of Rails, Grails, Django, Play! Framework - they all are action-based and don't try to hide from the programmer true request/response and stateless nature of the web.
For me it's major JSF disadvantage. IMHO JSF can suits some type of applications (intranet, forms-intensive), but for real-life web application it's not a good way to go.
Hope it helps somebody with his/her choices that regards to front-end.
A few drawbacks that pop to mind:
JSF is a component-based framework.
This has inherent restrictions that
have to do with obeying the
component-model.
AFAIK JSF supports only POST, so if you want a GET somewhere you have
to do a plain servlet/JSP.
Most components try to provide abstractions over domains like
relational databases and front-end
JavaScript, and many time these
abstractions are "leaky" and very hard to debug.
These abstractions might be a good starting point for a junior developer or someone not comfortable with a particular domain (e.g. front-end JavaScript), but are very hard to optimise for performance, since there are several layers involved, and most people that use them have little understanding of what is going on under the hood.
The templating mechanisms that are usually used with JSF have nothing to do with how web desigers work. The WYSIWYG editors for JSF are primitive and in any case, your designer will give you HTML/CSS that you'll have to spend ages converting.
Things like EL expressions are not statically checked and both the compiler and IDEs are not doing a good job at finding errors, so you'll end up with errors that you'll have to catch at run-time. This might be fine for dynamically typed language like Ruby or PHP, but if I have to withstand the sheer bloat of the Java ecosystem, I demand typing for my templates.
To sum up: The time you will save with JSF, from avoiding to write the JSP/servlet/bean boilerplate code, you'll spent it x10 to make it scale and do exactly what you want it to do.
To me the biggest disadvantage of JSF 2.0 is the learning curve not only of JSF, but the component libraries that you have to use in order to get it to do useful work. Consider the staggering number of specifications and standards you have deal with to really be proficient:
HTML in the various incarnations. Don't pretend you don't need to know it.
HTTP -- when you can't figure out what is going on you have to open Firebug and see. For that you need to know this.
CSS -- Like it or not. It isn't so bad really and there are some nice tools out there at least.
XML -- JSF will probably the first place you use namespaces to this degree.
Servlet Specification. Sooner or later you will get into calling methods in this package. Aside from that you have to know how your Facelets gets turned into XHTML or whatever.
JSP (mostly so you know why you don't need it in JSF)
JSTL (again, mostly to cope with legacy framework)
Expression Language (EL) in its various forms.
ECMAScript, JavaScript, or whatever else you want to call it.
JSON -- you should know this even if you don't use it.
AJAX. I would say JSF 2.0 does a decent job of hiding this from you but you still need to know what is going on.
The DOM. And how a browser uses it. See ECMAScript.
DOM Events -- a topic all by itself.
Java Persistence Architecture (JPA) that is if you want your app to have any back end data base.
Java itself.
JSEE while you are at it.
The Context Dependency Injection specification (CDI) and how it clashes with and is used with JSF 2.0
JQuery -- I would like to see you get along without it.
Now, once you are done with that you can get on with the proprietary specifications, namely the component libraries and provider libraries you will pick up along the way:
PrimeFaces (my component library of choice)
RichFaces
MyFaces
ICEFaces
EclipseLink (my JPA Provider)
Hibernate
Weld
And don't forget the container! And all those configuration files:
GlassFish (2, 3, etc)
JBoss
Tomcat
So -- THIS IS MAKING IT EASY? Sure, JSF 2.0 is "easy" as long as all you want to do is the most basic web pages with the simplest interactions.
Simply put, JSF 2.0 is the most complicated and cumbersome mishmash of glued together technologies as exists in the software universe today. And I can't think of anything I would rather use.
Inexperienced developers usually will create applications that are painfully slow and code will be really ugly and hard to maintain. Its deceptively simple to start, but actually requires some investment in learning if you want to write good programs.
At least at the start you will often "stuck" on some problem and will spend more time reading balusc posts on internet than actually working :) After a while it will be less and less of that, but it still can be annoying.
Even more annoying when you find out that the problem is not due to you lack of knowledge/mistake but actually a bug. Mojarra was(is?) quite buggy, and another layer of components adds even more problems. Richfaces was biggest piece of crap software ever written :) Don't know how it is now on version 4. We have Primefaces which is better, but still you will run into bugs or lack of features especially with more exotic components. And now you will need to pay for Primefaces updates. So I would say its buggy but its getting better especially after 2.2 version fixed some problems with spec. Framework getting more mature but still far from perfect (maybe myfaces better?).
I don't find it especially flexible. Often if you need something very very customized and there are no components that does that - it will be a bit painful. Again I'm talking from average developer perspective - the one with deadlines, quick reading tutorials, and searching stackoverflow when getting stuck because no time to learn how it really works :) Often some components seems to have "almost" what you need, but not exactly and sometimes you might spend too much time to make it do something you want :) Need to be careful in evaluating if its better to create your own or torture existing component. Actually if you are creating something really unique I would not recommend JSF.
So in short my drawbacks would be: Complexity, Not very smooth development progress, buggy, inflexible.
Of course there are advantages too, but that's not what you asked. Anyway that's my experience with framework others might have different opinions, so best way is to just try it for a while to see if its for you (just something more complex - not naive examples - JSF really shines there:) IMHO best use case for JSF is business applications, like CRMs etc...
"JSF will output View-layer HTML and JavaScript that you cannot control or change without going into Controller code."
Actually JSF gives you the flexibility, you can either use standard/third-party components or create your own which you have full control over what is rendered. It is just one xhtml you need to create your custom components with JSF 2.0.
We developed a sample project with JSF (It was a three week research so we may have lose some things!)
We try to use core jsf, if a component is needed we used PrimeFaces.
The project was a web site with navigation. Each page should be loaded via ajax when the menu is clicked.
The site has two usecases:
A page with a grid. The grid is loaded via ajax and should support sort and paging
A three step wizard page. Each page has client side validation (for simple validations) and server side ajax base validation (for complex validations). Any server exception ( from service layer) should be displayed on the same page of wizard without navigating to next page.
We found that:
You need to use some hacks from omniFaces to make the JSF view state fixed. The JSF state will be corrupted when you include pages via ajax in each other. This seems a bug in JSF and may be fixed on next releases (not in 2.3).
The JSF Flow is not working correctly with ajax (or we could not make it work!) We try to use primeface wizard component instead but the client validation seems not supported and mean while it was not standard JSF flow standard.
When using some jQuery components like jqGird, and you need to load JSON results, then you are advised to use pure servlet, The JSF will do nothing for you. So if you use these kind of components, your design will not fit in JSF.
We try to do some client scripts when ajax complete by ajaxComplete and we found that the PF 4 has implemented its own ajax events. We had some jQuery components and we need to change their code.
If you change the above sample to a non Ajax project ( or at least less ajax project) you will not face lots of above issues.
We summarize our research as:
JSF is not working well in an fully ajax base website.
Of course we find lots of nice features in JSF which may be very helpful in some projects, so consider your project needs.
Please refer to JSF technical documents to review JSF advantages, and in my opinion the biggest advantage of JSF, is the COMPLETE AND HUGE support from #BalusC ;-)
I'm not a Java Server Faces expert at all. But IMHO the main disadvantage is that it's server side. I'm tired of learning and using server side web presentation layer frameworks like ASP.NET Web Forms, ASP.NET MVC, Java Server Faces, Struts, php frameworks and ruby on rails frameworks. I said goodbye to all of them, and I said hello to Angularjs and TypeScript. My presentation layer runs on the browser. I doesn't matter if it is served by Windows IIS running php or ASP.NET, or if it is served by an Apache web server running on Linux. I just need to learn just one framework that works everywhere.
Just my two cents.
For me the biggest shortcoming of JSF is poor support for programmatically (dynamically) generated pages.
If you want to construct your page (create page component model) dynamically from java code. For example if you are working on WYSIWYG web page constructor. Adequate documentation of this use case in not generally available. There are many points where you have to experiment and development is quiet slow. Many things just don't work how you would expect. But generally its possible hack it somehow.
Good thing is that it's not problem in philosophy or architecture of JSF. It's simply not elaborated enough (as far as I know).
JSF 2 brought Composite Components which should make component development easy, but their support for dynamic (programmatic) construction is very poor. If you overcome quiet complicated and almost undocumented process of dynamic Composite Component construction, you will find out that If you nest few Composite components little deeper, they stop working, throwing some exceptions.
But It seems that JSF community is aware of this shortcomings. They are working on this as you can see from these two bugs
http://java.net/jira/browse/JAVASERVERFACES-1309
http://java.net/jira/browse/JAVASERVERFACES_SPEC_PUBLIC-599
Situation should be better with JSF 2.2 at least if we are talking about specification.
Commenting on my last few months of Primefaces/JSF experience:
If you can use components "off the shelf", I guess it's not terrible.
However, it doesn't play well as soon as you step outside and need custom UIs. - For example, we needed to use Twitter's bootstrap for our project. (Not primefaces bootstrap).
Now our pages work as follows:
Page loads.
User interacts with a Primefaces that has ajax functionality
Bootstrap's javascript bindings break
We run extra javascript to rebind everything
The promise of JSF to avoid writing javascript turned into writing more javascript than we would have if not using Primefaces--and that javascript to is fix what Primefaces breaks.
It's a time sink--unless you again use "off the shelf" stuff. Also really ugly (Primefaces) when having to work with Selenium. It can all be done--but again--there's only so much time.
Definitely avoid this if you're working with a UX/design team and need to rapidly iterate on the UI--you can save time by learning jquery/writing straight HTML--or looking at react/angular.
JSF has many advantages, question being on disadvantage let me add couple of points on it.
On a practical scenario of implementing a web project with in a time frame you need to keep an eye on the following factors.
Do you have enough senior members in your team who can suggest best
controls suitable for each scenario?
Do you have the bandwidth to accommodate the initial learning curve?
Do you have enough expertise in your team who can review the JSF
stuff produces by the developers?
If your answer is 'No' for the questions, you may end up in a non-maintainable codebase.
JSF has only one disadvantage: before starting "JSF" development you should clearly understand web development, core java and front-end architecture.
Nowadays "new" JavaScript frameworks just try to copy/paste "JSF" component-based model.
Among all the "mainstream" frameworks such as Spring MVC, Wicket, Tapestry, etc., the JSF of Java EE with its composite components is the most elaborated presentation-layer and component-oriented technology provided. It is a bit cumbersome and incomplete compared to solutions provided by HybridJava.