ADF Faces: Optimizing retrieving beans programmatically

From time to time in JSF and ADF Faces RC applications there’s the need from one managed bean to retrieve another programatically, typically from a lesser scoped bean to a greater, such as a requestScope bean retrieving a sessionScope bean to access its methods. There’s essentially 3 avenues to solving this problem programatically:

1) The following JSF 1.1 createValueBinding method that retrieves the bean using EL:

FacesContext ctx = FacesContext.getCurrentInstance();
Application app = ctx.getApplication();
ValueBinding bind = app.createValueBinding("#{beanName}");
Bean bean = (Bean) bind.getValue(ctx);

Note parts of this code were deprecated since JSF 1.1

2) A JSF 2.0 compliant way using evaluateExpressionGet also evaluating EL to retrieve a bean:

FacesContext context = FacesContext.getCurrentInstance();
Bean bean = (Bean) context.getApplication().evaluateExpressionGet(context, "#{beanName}", Bean.class);

(Thanks to BalusC on StackOverFlow for the technique).

3) Or a direct programmatic method that doesn’t use EL:

ExternalContext exctxt = FacesContext.getCurrentInstance().getExternalContext();

AppBean appBean = (AppBean) exctxt.getApplicationMap().get("appBeanName");
SessionBean sessionBean = (SessionBean) exctxt.getSessionMap().get("sessionBeanName");
RequestBean sessionBean = (RequestBean) exctxt.getRequestMap().get("requestBeanName");

AdfFacesContext adfctxt = AdfFacesContext.getCurrentInstance();

ViewBean viewBean = (ViewBean)adfctxt.getViewScope().get("viewBeanName");
PageFlowBean pageFlowBean = (PageFlowBean)adfctxt.getPageFlowScope().get("pageFlowBeanName");

With these 3 approaches in mind it’s interesting to gather some statistics on how long it takes each method to retrieve each type of bean. The following chart shows 5 consecutive runs of each method, per bean type, where each method is repeated in a loop 10000 times (so we can see the statistical differences).

(Note each run included an initialization section that wasn’t recorded in the statistics, designed to prime the beans and the access methods such that startup times didn’t impact the results).

The numbers represent nanoseconds but their individual values mean little. However the ratios between each method is of more interest and the following conclusions can be made:

a) The older createValueBinding method is slower than the evaluateExpressionGet method, at least initially but this trends to insignificant over subsequent runs.

b) The direct method is faster than the other methods in all cases.

c) Particular gains are to be made in using the direct method to access application, session and request scoped beans. Less so but still positive gains for view and pageFlow scoped beans.

Sample App

The sample app can be downloaded from here.

Platform

MacBookPro 2.3GHz Core i7, 8GB RAM, Mac OS X 10.6.8
JDev 11.1.2 JDK 64bit 1.6.0_26

Thanks

My thanks must go to Simon Lessard who inspired this post sometime back.