You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@tapestry.apache.org by hl...@apache.org on 2006/11/24 21:03:48 UTC
svn commit: r478972 [2/2] - in
/tapestry/tapestry5/tapestry-project/trunk/src/site/resources:
tap5devwiki.html tap5devwiki.xml
Modified: tapestry/tapestry5/tapestry-project/trunk/src/site/resources/tap5devwiki.xml
URL: http://svn.apache.org/viewvc/tapestry/tapestry5/tapestry-project/trunk/src/site/resources/tap5devwiki.xml?view=diff&rev=478972&r1=478971&r2=478972
==============================================================================
--- tapestry/tapestry5/tapestry-project/trunk/src/site/resources/tap5devwiki.xml (original)
+++ tapestry/tapestry5/tapestry-project/trunk/src/site/resources/tap5devwiki.xml Fri Nov 24 12:03:47 2006
@@ -6,27 +6,76 @@
<description>The quick and dirty one-stop shopping of random ideas for Tapestry 5.</description>
<language>en-us</language>
<copyright>Copyright 2006 HowardLewisShip</copyright>
-<pubDate>Sat, 04 Nov 2006 17:49:18 GMT</pubDate>
-<lastBuildDate>Sat, 04 Nov 2006 17:49:18 GMT</lastBuildDate>
+<pubDate>Fri, 24 Nov 2006 19:57:26 GMT</pubDate>
+<lastBuildDate>Fri, 24 Nov 2006 19:57:26 GMT</lastBuildDate>
<docs>http://blogs.law.harvard.edu/tech/rss</docs>
<generator>TiddlyWiki 2.0.11</generator>
<item>
-<title>RandomIdeas</title>
-<description>!HTML / XHTML DTDs<br /><br />The template parser should include local (in JAR) copies of the HTML and XHTML DTDs and redirect the parser to use the local copies. This can be a huge performance boost when parsing a template.<br /><br />!final should imply @Retain<br /><br />Final fields should be treated as if they have the @Retain annotation<br /><br />! Exceptions from event handler / phase render methods<br /><br />Tapestry should wrap non-runtime exceptions from these methods. I think today, if you declare that such a method throws an exception, you'll get a runtime exception out of Javassist.<br /><br />! SubForms<br /><br />Perhaps one way to approach highly dynamic, Ajax pages with forms is to have a logical "sub form" concept. A sub form would work inside an existing form, and organize a group of fields within that form. Processing of the fields would occur only
if the sub form was active, which itself<br />would be tracked based on visibility of the sub form (a sub form in an invisible panel would not be processed on the server side). This idea needs a lot of fleshing out, even to see if it is viable.<br /><br />! Ajax Constraints<br /><br />The best way to tackle Ajax features, especially w.r.t. forms, is to put some sensible constraints on what the user can do, then make it easy to implement those things.<br /><br />Basically ... never delete! Deletions are a real pain to handle, unless I suddenly get much smarter. Allow things to be hidden on the client side,<br />and for the corresponding fields to do nothing on the server side, but don't allow them to full out delete. <br /><br />Allow new things to be added, preferable only at the "tail end" of the form. </description>
-<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#RandomIdeas</link>
-<pubDate>Sat, 04 Nov 2006 17:49:18 GMT</pubDate>
+<title>WatchCodeCoverage</title>
+<description>The code coverage tools built into ''mvn site'' are quite useful. Right now, overall coverage is at 93%. Keep an eye on code coverage, including branch coverage (do you test both outcomes of an if statement?). Use unexecuted code to target your efforts.<br /><br />I often do a cursory unit test for "normal behavior", plus more exaustive unit tests for error conditions. I then "back the test up" using an integration test (build with [[Selenium]]) to prove that the normal behavior case really works.</description>
+<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#WatchCodeCoverage</link>
+<pubDate>Fri, 24 Nov 2006 19:57:26 GMT</pubDate>
+</item>
+<item>
+<title>FocusOnTesting</title>
+<description>I'm still not at the stage of test first, but with every line of code I write, I am thiking about how I will test that line of code. Tapestry uses EasyMock extensively, and there's lots of existing code examples to work with.<br /><br />!Dont Despair<br /><br />I occasionally get exhausted by the amount of test code I write for simple chunks of code. And I inevitably find a broken line of code that would be a major pain to locate inside a running application, but easy inside a unit test. Keep your eyes on the big picture.</description>
+<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#FocusOnTesting</link>
+<pubDate>Fri, 24 Nov 2006 19:54:49 GMT</pubDate>
+</item>
+<item>
+<title>DontTouchInternals</title>
+<description>Very few of the Tapestry committers will need to touch anything inside the internals package. When this occurs, some online discussion may be mandated. The goal of Tapestry is to ensure that most work for commiters and for developers is in the domain of creating components, not tinkering with internals.</description>
+<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#DontTouchInternals</link>
+<pubDate>Fri, 24 Nov 2006 19:53:07 GMT</pubDate>
+</item>
+<item>
+<title>Assets</title>
+<description>The concept of Assets is unchanged from Tapestry 4.<br /><br />There will be an "exclusion list" of file name extensions. Files with that extension will require an md5sum as part of the URI. The default exclusion list with be ".class". For other file types (particularily .js, .css, etc.) the file will be the file, and relative file names will work.<br /><br />Some kind of hook will be necessary to support schemes like Akamai, where assets are stored externally. This should also take into account localization.<br /><br />Some kind of folder aliasing will be necessary. I may want "/assets/dojo/" to map on the classpath to "/org/apache/tapestry/dojo/dojo_0_0_4/". When handling updates to big packages such as Dojo, I don't want to have to sort through some ungodly number of changes between 0.0.4 and 0.0.5, I want to create a new dojo_0_0_5 folder, drop the contents in, delete the dojo_0_0
_4 folder, and update the mapping of "/assets/dojo/".</description>
+<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#Assets</link>
+<pubDate>Fri, 24 Nov 2006 19:51:07 GMT</pubDate>
</item>
<item>
<title>MasterIndex</title>
-<description>Top level concepts within Tapestry 5.<br /><br />A //meta-note//: This is where new ideas are first explained, usually before being implemented. In many cases, the final implementation is<br />not a perfect match for the notes. That's OK ... as long as the official Maven documentation does a good job. It's not reasonable to expect developers to jump back in here and dot every i and cross every t if they're already expected to generate good Maven documentation.<br /><br />* PropBinding -- Notes on the workhorse "prop:" binding prefix<br />* TypeCoercion -- How Tapestry 5 extensibly addresses type conversion<br />* FormProcessing<br />* DynamicPageState -- tracking changes to page state during the render<br />* EnvironmentalServices -- how components cooperate during page render<br />* ComponentMixins -- A new fundamental way to build web functionality<br />* RequestTypes -- Requests, requ
est processing, URL formats<br />* ComponentTemplates -- Issues about Component Templates<br />* DeveloperProcedures -- Your a Tapestry committer ... how do you makes changes?<br />* SmartDefaults -- do even more with event less<br />* RandomIdeas -- stuff that doesn't fit elsewhere</description>
+<description>Top level concepts within Tapestry 5.<br /><br />A //meta-note//: This is where new ideas are first explained, usually before being implemented. In many cases, the final implementation is<br />not a perfect match for the notes. That's OK ... as long as the official Maven documentation does a good job. It's not reasonable to expect developers to jump back in here and dot every i and cross every t if they're already expected to generate good Maven documentation.<br /><br />* PropBinding -- Notes on the workhorse "prop:" binding prefix<br />* TypeCoercion -- How Tapestry 5 extensibly addresses type conversion<br />* FormProcessing<br />* DynamicPageState -- tracking changes to page state during the render<br />* EnvironmentalServices -- how components cooperate during page render<br />* ComponentMixins -- A new fundamental way to build web functionality<br />* RequestTypes -- Requests, requ
est processing, URL formats<br />* ComponentTemplates -- Issues about Component Templates<br />* DeveloperProcedures -- Your a Tapestry committer ... how do you makes changes?<br />* SmartDefaults -- do even more with event less<br />* RandomIdeas -- stuff that doesn't fit elsewhere<br />* ProblemsNeedingSolutions<br />* ComponentDocumentation -- Generating Documentation about Components<br />* TapestryLookAndFeel -- Default CSS<br />* [[Assets]]<br /><br /></description>
<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#MasterIndex</link>
-<pubDate>Sat, 04 Nov 2006 00:38:00 GMT</pubDate>
+<pubDate>Fri, 24 Nov 2006 19:44:49 GMT</pubDate>
+</item>
+<item>
+<title>TapestryLookAndFeel</title>
+<description>I think it would be very compelling to create a reasonably sharp default CSS for Tapestry applications. Something pretty, standards based, Web-2.0-ish, that would make even the simplest apps stand out.<br /><br />Of course, the Tapestry default CSS would be the first stylesheet, and could be overridden by additional stylesheets or inline styles.</description>
+<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#TapestryLookAndFeel</link>
+<pubDate>Fri, 24 Nov 2006 19:43:24 GMT</pubDate>
+</item>
+<item>
+<title>DynamicPageState</title>
+<description>Tapestry 4 has left tracking of dynamic page state as an exercise to the developer. Mostly, this is done using the ''parameters'' parameter of the ~DirectLink component.<br /><br />''Update: As I've thought this one through, I don't think it is viable outside of Forms. It will end up with long URLs and a constant ambiguity about whether each link should include or exclude the page state. So this one is unlikely to get implemented.''<br /><br />Dynamic page state is anything that isn't inside a persistent page property. For the most part, this includes page properties updated by a For component<br /><br />It seems likely that this information could be automatically encoded into ~URLs. <br /><br />I'm envisioning a service that accumulates a series of //commands//. Each command is used to store a bit of page state. The commands are serializable. The commands are ultimately serialized into a MIME string and attach
ed as a query parameter to each URL.<br /><br />When such a link is triggered, the commands are de-serialized and each executed in turn. Only when that is finished is any further event processing executed, including calling into to user code.<br /><br />My outline for this is to store a series of tuples; each tuple is a component id plus the command to execute.<br /><br />{{{<br />public interface ComponentCommand<T><br />{<br /> void execute(T component);<br />}<br />}}}<br /><br />These commands should be immutable.<br /><br />So a component, such as a For loop component, could provide itself and a ComponentCommand instance (probably a static inner class) to some kind of PageStateTracker service.<br /><br />{{{<br />public interface PageStateTracker<br />{<br /> void <T> addCommand(T component, ComponentCommand<T> command);<br />
}<br />}}}<br /><br />The commands are kept in the order that they are added, except that new commands for the same component //replace// previous commands for that component.<br /><br />As with the Tapestry 4 For component, some mechanism will be needed to store object ids inside the URLs (that is, inside the commands serialized into URL query parameters) and translate back to //equivalent// objects when the link is triggered.<br /><br />Dynamic page state outside of a Form will overlap with some of the FormProcessing inside the form.</description>
+<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#DynamicPageState</link>
+<pubDate>Fri, 24 Nov 2006 18:32:00 GMT</pubDate>
</item>
<item>
<title>SmartDefaults</title>
<description>As great as the annotations are, allowing things to work without the annotations could be even better.<br /><br />!Event handler methods<br /><br />Methods with the prefix "on" could automatically be considered event handler methods. The string after the prefix, converted to lower case, would be the event type. We could even add "from//~ComponentId//" to the end. Examples (with annotation equivalents):<br /><br />* onSubmit --> @~OnEvent("submit")<br />* onSubmitFromForm --> @~OnEvent(value="submit", component="form")<br />* onUpdateFromSelect --> @~OnEvent(value="update", component="select")<br /><br />!Render phase methods<br /><br />Naming a method the same as the render phase (with the first character lower case). Again, Tapestry could deduce the phase from the method name, as if the annotation were pres
ent:<br /><br />* beforeRender() --> @~BeforeRender<br />* beforeRenderBody() --> @~BeforeRenderBody<br /><br />etc. Again, the methods don't have to be public, they just have to have the correct name. In every other way they are the same as annotated render phase methods except that they don't have the annotation.<br /><br />There may be some minor implications w.r.t. render phase method ordering.<br /><br />etc.<br /><br />!Other Ideas<br /><br />This gets more component specific. I had the idea that a ~TextField whose id was "userId" might want to edit a property named "userId" as the default for when its value parameter is unbound. I think to accomplish this, we need the concept of computed bindings for unbound parameters ... perhaps in the form of methods that return a Binding with a name and/or annotation, for example:<br /><br />{{{<br /><br /&
gt; @Inject<br /> private ComponentResources _resources;<br /><br /> @Inject("infrastructure:bindingSource")<br /> private BindingSource _source;<br /><br /> @Parameter<br /> private Object _value;<br /><br /> Binding valueDefault()<br /> {<br /> ComponentResources containerResources = _resources.getContainer().getComponentResources();<br /> return _source.newBinding("default value", containerResources, "prop", _resources.getId(), null); <br /> }<br />}}}<br /><br />So valueDefault() is invoked if the value parameter is not bound. The component uses its own immediate id ("userId") as the name of a property of its container (typically, the page). ~ComponentResources does not yet implement getContainer(), but the rest would work.<br /><br />If this was widespread, there could be even better optimizations for it.
Perhaps container resources could just be passed into the method as a parameter, to save the code to find it. Ditto with BindingSource. Once again, rather than come up with complex XML-ese to come up with defaults, we're trying to work //with// Java code.<br /></description>
<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#SmartDefaults</link>
-<pubDate>Fri, 03 Nov 2006 23:59:00 GMT</pubDate>
+<pubDate>Thu, 23 Nov 2006 15:35:00 GMT</pubDate>
+</item>
+<item>
+<title>ComponentDocumentation</title>
+<description>Tapestry needs a JavaDoc-like, or JavaDoc-based tool to extract documentation about component parameters (and mixins) and present it in a useable format. In Tapestry 4, you could document the public getters and setters, but in Tapestry 5, the annotations are on the private instance variables, which are generally not documented.</description>
+<category>parameters</category>
+<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#ComponentDocumentation</link>
+<pubDate>Thu, 23 Nov 2006 15:34:00 GMT</pubDate>
+</item>
+<item>
+<title>RandomIdeas</title>
+<description>!HTML / XHTML DTDs<br /><br />The template parser should include local (in JAR) copies of the HTML and XHTML DTDs and redirect the parser to use the local copies. This can be a huge performance boost when parsing a template.<br /><br />!final should imply @Retain<br /><br />Final fields should be treated as if they have the @Retain annotation<br /><br />! Exceptions from event handler / phase render methods<br /><br />Tapestry should wrap non-runtime exceptions from these methods. I think today, if you declare that such a method throws an exception, you'll get a runtime exception out of Javassist.<br /><br />! SubForms<br /><br />Perhaps one way to approach highly dynamic, Ajax pages with forms is to have a logical "sub form" concept. A sub form would work inside an existing form, and organize a group of fields within that form. Processing of the fields would occur only
if the sub form was active, which itself<br />would be tracked based on visibility of the sub form (a sub form in an invisible panel would not be processed on the server side). This idea needs a lot of fleshing out, even to see if it is viable.<br /><br />! Ajax Constraints<br /><br />The best way to tackle Ajax features, especially w.r.t. forms, is to put some sensible constraints on what the user can do, then make it easy to implement those things.<br /><br />Basically ... never delete! Deletions are a real pain to handle, unless I suddenly get much smarter. Allow things to be hidden on the client side,<br />and for the corresponding fields to do nothing on the server side, but don't allow them to full out delete. <br /><br />Allow new things to be added, preferable only at the "tail end" of the form. <br /><br />! SPI Package<br /><br />A number of interfaces, such as Binding
, probably belong in a SPI (Service Provider Interface) package, since they will generally only be used by authors of Tapestry extensions. Perhaps we should just use the oat.services package as the SPI package?</description>
+<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#RandomIdeas</link>
+<pubDate>Thu, 23 Nov 2006 05:01:00 GMT</pubDate>
+</item>
+<item>
+<title>ProblemsNeedingSolutions</title>
+<description>There are a few things that I'm concerned about.<br /><br />!Render Complexity<br /><br />All those states in the render component state machine may be a little much, especially ~PreBeginRender, ~BeginRender and ~PostBeginRender. In addition, it doesn't work for a case I'm interested in ... for link components, I'd like to use the RenderInformals mixin, but also support a disable parameter that turns off the <a> tag (but still renders the body). The state machine currently is set up so that returning false in any of the ~BeginRender states skips all the way to ~AfterRender, bypassing the template and/or body.</description>
+<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#ProblemsNeedingSolutions</link>
+<pubDate>Thu, 23 Nov 2006 04:01:00 GMT</pubDate>
</item>
<item>
<title>WorkInYourOwnBranch</title>
@@ -85,58 +134,6 @@
<description>Page render requests are requests used to render a specific page. //render// is the term meaning to compose the HTML response to be sent to the client. Note: HTML is used here only as the most common case, other markups are entirely possible.<br /><br />In many cases, pages are stand-alone. No extra information in the URL is necesarry to render them. PersistentProperties of the page will factor in to the rendering of the page.<br /><br />In specific cases, a page needs to render within a particular context. The most common example of this is a page that is used to present a specific instance of a database persistent entity. In such a case, the page must be combined with additional data, in the URL, to identify the specific entity to access and render.<br /><br />! URI Format<br /><br />{{{<br />/page-name.html/id<br />}}}<br /><br />Here "page-name" is the LogicalPageName for th
e page. <br /><br />The ".html" file extension is used as a delimiter between the page name portion of the URI, and the context portion of the URI. This is necessary because it is not possible (given the plethora of libraries and folders) to determine how many slashes will appear in the URI.<br /><br />The context consists of one ore more ids (though a single id is the normal case). The id is used to identify the specific data to be displayed. Further, a page may require multiple ids, which will separated with slashes. Example: /admin/DisplayDetail.html/loginfailures/2006<br /><br />Note that these context values, the ids, are simply //strings//. Tapestry 4 had a mechanism, the DataSqueezer, that would encode the type of object with its value, as a single string, and convert it back. While seemingly desirable, this facility was easy to abuse, resulting in long and extremely ugly URIs.<br /><br />Any further informatio
n needed by Tapestry will be added to the URI as query parameters. This may include things like user locale, persistent page properties, applicaition flow identifiers, or anything else we come up with.<br /><br />! Request Processing<br /><br />Once the page and id parameters are identified, the corresponding page will be loaded.<br /><br />Tapestry will fire two events before rendering the page.<br /><br />The first event is of type "setupPageRender". This allows the page to process the context (the set of ids). This typically involves reading objects from an external persistent store (a database)<br />and storing those objects into transient page properties, in expectaion of the render.<br /><br />The @SetupPageRender annotation marks a method to be invoked when this event is triggered. The method may take one or more strings, or an array of strings, as parameters; these will be<br />the co
ntext values. The method will normally return void. Other values are ''TBD''. It may also take other simple types, which will be coerced from the string values.<br /><br />{{{<br />@SetupPageRender<br />void setup(long id)<br />{<br /> . . .<br />}<br />}}}<br /><br /><br /><br />The second event is of type "pageValidate". It allows the page to decide whether the page is valid for rendering at this time. This most often involves a check to see if the user is logged into the application, and has the necessary privileges to display the contents of the page. User identity and privileges are //not// concepts built into Tapestry, but are fundamental to the majority of Tapestry applications.</description>
<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#PageRenderRequest</link>
<pubDate>Sun, 08 Oct 2006 13:33:00 GMT</pubDate>
-</item>
-<item>
-<title>LogicalPageName</title>
-<description>A logical page name is the name of a page as it is represented in a URI.<br /><br />Internally, Tapestry operates on pages using full qualified class names. Technically, the FQCN is the class of the page's root element, but from an end developer point of view, the root element is the page.<br /><br />The logical page name must be converted to a fully qualified class name.<br /><br />A set of LibraryMappings are used. Each library mapping is used to express a folder name, such as "core", with a Java package name, such as org.apache.tapestry.corelib. For pages, the page name is searched for in the pages sub-package (i.e., org.apache.tapestry.corelib.pages). Component libraries have unique folder names mapped to root packages that contain the pages (and components, and mixins) of that library.<br /><br />When there is no folder name, the page is expected to be part of the application, under the pages sub-
package of the application's root package.<br /><br />If not found there, as a special case, the name is treated as if it were prefixed with "core/". This allows access to the core pages (and more importantly, components -- the search algorithm is the same).<br /><br />Finally, pages may be organized into folders. These folders become further sub-packages. Thus as page name of "admin/EditUsers" may be resolved to class org.example.myapp.pages.admin.EditUsers.<br /><br /></description>
-<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#LogicalPageName</link>
-<pubDate>Sun, 08 Oct 2006 13:30:00 GMT</pubDate>
-</item>
-<item>
-<title>OGNL</title>
-<description>The [[Object Graph Navigation Library|http://ognl.org]] was an essential part of Tapestry 4.<br /><br />OGNL is both exceptionally powerful (especially the higher order things it can do, such as list selections and projections). However, for the highest<br />end sites, it is also a performance problem, both because of its heavy use of reflection, and because it uses a lot of code inside synchronized blocks.<br /><br />It will be optional in Tapestry 5. I believe it will not be part of the tapestry-core, but may be packaged as tapestry-ognl.<br /><br />The "prop:" binding prefix is an effective replacement for OGNL in Tapestry 5. See PropBinding.<br /></description>
-<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#OGNL</link>
-<pubDate>Sat, 07 Oct 2006 12:49:00 GMT</pubDate>
-</item>
-<item>
-<title>ComponentMixins</title>
-<description>One of the more exciting ideas in Tapestry 5 is //mixins//; the ability to add behavior to a component without writing code. <br /><br />It is expected that much common behavior, especially for form control components, will be provided by mixins. Further, many Ajax techniques will take the form of mixins applied to otherwise ordinary components.<br /><br />A mixin is an additional component class that operates //with// the main component. For a component element within the page, the functionality is provided by the main component class and by<br />the mixin. <br /><br />Mixins are primarily about rendering. Mixin render methods are //mixed in// to the components' render methods. In effect, the different rendering phases of a component are different AOP-like //joinpoints//, and the mixins can provide //before advice//.<br /><br />Mixins can be specified for an //instance// of a component, or can be specified
as part of the //implementation// of a component.<br /><br />In the former case, the @Component annotation will be supplemented with a @Mixin annotation. The @Mixin is a list of one or more mixin classes for that component.<br /><br />''Todo: Template syntax for mixins?''<br /><br />In the latter case, the @ComponentClass annotation will be supplemented with a @Mixin annotation.<br /><br />Mixins can be configured. They can have parameters, just like ordinary components. When a formal parameter name is ambiguous, it will be prefixed with the unqualified class name. Thus, you might have to say, "MyMixin.parameterName=someProperty" if "parameterName" is ambiguous (by ambiguous, we mean, a parameter of more than one mixin or of the component itself). <br /><br />This disambiguation is //simple//. It is assumed that the unqualified class name will be sufficient to uniquely identify a mixin. That is
, it is expected that you will not have the same class name even in different packages (as mixins, on a single component). In a //degenerate case// where this is not so, it will be necessary to disambiguate the mixin name by create a subclass of the mixin with a new name.<br /><br />''Todo: how are mixins on a component implementation configured?''<br /><br />Mixins may have persistent state, just as with ordinary components.<br /><br /></description>
-<category>mixins</category>
-<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#ComponentMixins</link>
-<pubDate>Thu, 05 Oct 2006 12:43:00 GMT</pubDate>
-</item>
-<item>
-<title>TypeCoercion</title>
-<description>Automatic coercion of types is essential. This primarily applies to component parameters.<br /><br />Parameters are tied to the [[Binding|http://tapestry.apache.org/tapestry5/apidocs/org/apache/tapestry/Binding.html]] interface.<br /><br />Tapestry component parameters look like simple instance variables, but Tapestry's RuntimeTransformation of component classes means that reading the value of a parameter instance variable //may// invoke Binding.get(), and changing the value of a parameter instance variable will invoke Binding.set().<br /><br />!Reading From Parameters<br /><br />Reading a parameter value involves two steps:<br />* Invoking Binding.get()<br />* Converting the result to the type of the parameter (where different)<br /><br />When reading parameters, the binding will provide an object of the type of the bound property. Various kinds of invariant bindings will returned a fixed
type, typically a String.<br /><br />The parameter will be assigned to a variable that has a known type, possibly a primtive type (int, boolean) or an object type (Map, Date).<br /><br />!Writing To Parameters<br /><br />Writing to, or updating, a parameter is in two steps:<br />* Converting the new value into a type appropriate for the binding<br />* Invoking Binding.set()<br /><br />We will be adding a getPropertyType() method to the Binding interface, that will identify the property type of the property bound to the parameter.<br /><br />The component will be responsible for performing a coercion from the value provided to the proper type, before invoking Binding.set().<br /><br />!Coercion Tuples<br /><br />At the core of this will be a service that performs type coercions.<br /><br />Coercions are based on //coercion tuples// that define:<br />* A source ty
pe<br />* A target type<br />* An object to perform the coercion from source to target<br />* A "cost" for the conversion (possibly, but usually with a standard default value) ''(not yet implemented)''<br /><br />As a special case, the type of null will be treated as type void (i.e., void.class). Thus we can use the same mechanism to identify how to convert from null to other types, such as Boolean or Integer.<br /><br />There should be a large number of these tuples available. The most common tuples may be conversions between various types and String.<br /><br />!Coercion Algorithm<br />* Determine the source type (treating null as void.class)<br />* Determine the target type (converting primitive types to equivalent wrapper types)<br />* If the source type is assignable to the target type, then the input value is valid and the process is complete<br />* Find a converter that converts
between the source type and the target type, pass the source value through the converter to get a target value<br /><br />That last part needs a bit of expansion.<br /><br />First off, there will often ''not'' be a tuple for coercing directly form the source type to the target type.<br /><br />In that scenario, the conversion will involve a search to find a sequence of tuples that will perform the coercion. This will take the form a breadth-first search where we look for tuples that coerce from the source type to an intermediate type, then search for tuples from the intermediate type to the target type. This may involve more than two coercions.<br /><br />You can think of the set of tuples as a kind of directed graph. Each type is a node on the graph, and each tuple represents a connection between one type and another type (say, from String to Double). What we're trying to do is find a path form a source type (or some supe
r-class or super-interface of the source type) to some target type (or sub-class or sub-interface of the target type).<br /><br />May need to express a "cost" of the coercion from start type to target type; this might be useful if there are multiple paths for the conversion. Cost may factor in both the computing expense, and any loss of detail. Basic cost is established in terms of the number of steps and enforced by the order in which tuples are considered and combined.<br /><br />For example, a coercion tuple from Number to Float may be represented as the tuple:<br />(Number, Float, {{{ return new Float(input.floatValue()); }}})<br /><br />{{{<br />public interface Coercion<S,T><br />{<br /> T coerce(S input);<br />}<br />}}}<br /><br />If the input type is an Integer, then a search for Integer->Float will find no entries. At that point, it will be necessary to "cl
imb" the inheritance tree and look for coercions from Number (the super class of Integer); this will find the Number->Float tuple.<br /><br />Again, in terms of cost, we might also find a pair of tuples: Object->String and String->Float. This will have a higher cost than the Number->Float tuple and should be rejected in favor of the lower cost coercion.<br /><br />//Note: cost hasn't been implemented, and likely won't be, unless and until the algorithm as it stands is shown to provide less than optimal results.//<br /><br />The algorithm caches the result of this search, with proper guards for concurrent access. The cache is cleared when an invalidation of the component class loader occurs.<br /><br />!Configuring the service<br /><br />This has been implemented as service tapestry.TypeCoercer.<br /><br />The configuration of this service is an unordered collection of CoercionTuple.</d
escription>
-<category>parameters</category>
-<category>types</category>
-<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#TypeCoercion</link>
-<pubDate>Thu, 05 Oct 2006 12:40:00 GMT</pubDate>
-</item>
-<item>
-<title>EnvironmentalServices</title>
-<description>Frequently, different components need to //cooperate// during the rendering process.<br /><br />This is an established pattern from Tapestry 4, which an enclosing component provides services to the components it encloses. By //encloses// we mean, any components that are rendered as part of the Form's body; give the use of the Block/~RenderBlock components, this can not be determined statically, but is instead determined dynamically, as part of the rendering process.<br /><br />The canoncial example of this pattern is Form component, and the complex relationship it has with each form element component it encloses.<br /><br />In Tapestry 4, this mechanism was based on the ~IRequestCycle which could store named attributes. The service providing component would store itself into the cycle using a well known name, and service consuming components would retrieve the service using the same well known name.<br /><br />For
Tapestry 5, this will be formalized. A new service will be used to manage this information:<br /><br />{{{<br />public interface Enviroment<br />{<br /> <T> T push(Class<T> type, T instance);<br /><br /> <T> peek(Class<T> type);<br /><br /> <T> T pop(Class<T> type);<br />}<br />}}}<br /><br />The Environment is unique to a request.</description>
-<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#EnvironmentalServices</link>
-<pubDate>Tue, 26 Sep 2006 01:45:00 GMT</pubDate>
-</item>
-<item>
-<title>DynamicPageState</title>
-<description>Tapestry 4 has left tracking of dynamic page state as an exercise to the developer. Mostly, this is done using the ''parameters'' parameter of the ~DirectLink component.<br /><br />Dynamic page state is anything that isn't inside a persistent page property. For the most part, this includes page properties updated by a For component<br /><br />It seems likely that this information could be automatically encoded into ~URLs. <br /><br />I'm envisioning a service that accumulates a series of //commands//. Each command is used to store a bit of page state. The commands are serializable. The commands are ultimately serialized into a MIME string and attached as a query parameter to each URL.<br /><br />When such a link is triggered, the commands are de-serialized and each executed in turn. Only when that is finished is any further event processing executed, including calling into to user code.<br /><br />My
outline for this is to store a series of tuples; each tuple is a component id plus the command to execute.<br /><br />{{{<br />public interface ComponentCommand<T><br />{<br /> void execute(T component);<br />}<br />}}}<br /><br />These commands should be immutable.<br /><br />So a component, such as a For loop component, could provide itself and a ComponentCommand instance (probably a static inner class) to some kind of PageStateTracker service.<br /><br />{{{<br />public interface PageStateTracker<br />{<br /> void <T> addCommand(T component, ComponentCommand<T> command);<br />}<br />}}}<br /><br />The commands are kept in the order that they are added, except that new commands for the same component //replace// previous commands for that component.<br /><br />As with the Tapestry 4 For component, some mechanism will be ne
eded to store object ids inside the URLs (that is, inside the commands serialized into URL query parameters) and translate back to //equivalent// objects when the link is triggered.<br /><br />Dynamic page state outside of a Form will overlap with some of the FormProcessing inside the form.</description>
-<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#DynamicPageState</link>
-<pubDate>Thu, 21 Sep 2006 16:35:00 GMT</pubDate>
-</item>
-<item>
-<title>FormProcessing</title>
-<description>Form processing in Tapestry 4 had certain strengths and limitations.<br /><br />Basically, any action framework that can do a simple mapping from query parameters to bean property names has advantages in terms of simple forms, and Tapestry 4's approach has huge advantages on more complex forms (with some considerable developer and framework overhead).<br /><br />With a direct mapping of query parameter names to bean names, each query parameter becomes self describing. You map query parameters to property of some well known bean. You do simple conversions from strings to other types (typically, ints and dates and the like). You drop query parameters that don't match up. You leave a lot of validation and other plumbing (such as getting those values into your DataTransferObjects) to the developer.<br /><br />But you never see a ~StaleLinkException.<br /><br />You also have some unwanted loopholes in your application i
n that //any// property can be updated through the URL. This is //one step// towards a security hole.<br /><br />!Tapestry 4 Approach<br /><br />Every form component, as it renders, asks the Form that encloses it to provide a client id. The terminology is a little messed; client id is the unique (within the form) name for //one rendering// of the component. If the component renders multiple times, because of loops, each rendering gets a unique name. This becomes the <input>'s name attribute, and ultimately, the query parameter name.<br /><br />Tapestry attempts to make the client id match the (user provided) component id. This is not always possible, especially in a loop, in which case a numeric suffix may be appended to the id to (help) ensure uniqueness.<br /><br />On render, a sequence of //component activations// occur, guided by the normal render sequence. The exact sequence of activations guides<br />the pro
duction of client ids.<br /><br />Using more advanced Tapestry techniques, including loops, conditionals and the Block/RenderBlock combo, the exact set of components and<br />component activations that will occur for a given rendering of a given form can not be predicted statically. Tapestry must actually render out the form<br />to discover all of these.<br /><br />In fact, while the Form component is producing this series of client ids, it builds up the list and stores it into the rendered page as a hidden form field. It will need it later, when the client-side form is submitted back to the server.<br /><br />An advantage of this approach is the disconnect between the query parameter names (the client ids) and the objects and properties being editted. Often the client ids will be //mneumonic// for the properties, but aren't directly mapped to them. Only the components responsible for each query parameter know how to validate
the submitted value, and what property of which object will need to be updated.<br /><br />When a form submission occurs, we want to ensure that each query parameter value read out of the request is applied to the correct property of the correct object. There's a limit to how much Tapestry can help here (because it has only a casual knowledge of this aspect of the application structure).<br /><br />During this submission process, which endded up with the curious name, //rewind phase//, Tapestry must do two things:<br />* Activate each component, such that the component may re-determine its client id, read its parameter, and update its page property<br />* Validate that the process has not been comprimised by a change of server side state<br /><br />That second element is a tricky one; things can go wonky if a race condition occurs between two users. For example, lets take a simple invoice and line item model. If users A and B b
oth read the same invoice, user A adds a line item, and user B changes a line item ... we can have a problem when user B submits the form. Now that there are three line items (not two) in the form, there will be extra component activations to process query parameters that don't exist in the request. <br /><br />This scenario can occur whenever the processing of the form submission is driven by server-side data that can change between request.<br /><br />Tapestry detects this as a difference in the sequence of client ids allocated, and throws a ~StaleLinkException, which is very frustrating for developers to comprehend and fix.<br /><br />There are also other edge cases for different race conditions where data is applied to the wrong server-side objects.<br /><br />The Tapestry 3 ~ListEdit component, which evolved into the Tapestry 4 For component, attempts to address this by serializing a series of //object ids// into the form
(as a series of hidden fields). This requires a bit of work on the part of the developer to provide an ~IPrimaryKeyConverter that can help convert objects to ids (when rendering) and ids back to objects (during form submission).<br /><br />Generally speaking, the Tapestry 4 approach represents layers of kludge on layers of kludge. It works, it gets the job done, it can handle some very complex situations, but it is less than ideal.<br /><br />!Tapestry 5<br /><br />The goal here is to capture the series of //component activations//, along with any significant page state changes, during the render.<br /><br />These activations will be a series of //commands//. For each component activation there will be two commands: the first command will be used to inform the component of its client id (this command executes during render and during form submission). The second command will request that the client handle the form submission
(this command executes only during form submission).<br /><br />The serialized series of commands is stored as a hidden form field.<br /><br />There's a lot of API to be figured out, especially the relationship between the form components and the form itself.<br /><br />Further, a lot of what the Tapestry 4 For component does, in terms of serializing dynamic page state, will need to fold into this as well.<br /><br />The end result will be a single hidden field with a big MIME string inside it ... but compared to the Tapestry 4 Form component (which has to write out many hidden fields) the whole will be less than the sum of the parts ... due to the overhead of serialization and gzip compression.<br /><br /><br /><br /><br /><br /><br /></description>
-<category>forms</category>
-<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#FormProcessing</link>
-<pubDate>Thu, 21 Sep 2006 15:40:00 GMT</pubDate>
-</item>
-<item>
-<title>SiteUrl</title>
-<description>http://tapestry.apache.org/tapestry5/tap5devwiki.html</description>
-<link>http://tapestry.apache.org/tapestry5/tap5devwiki.html#SiteUrl</link>
-<pubDate>Thu, 21 Sep 2006 07:03:00 GMT</pubDate>
</item>
</channel>
</rss>