IDE choices, coding style and formatting, commit practices, naming conventions and other issues relevant to Tapestry committers & contributers.
It's a free license for all committers and it's just better. Yes, the first few days can be an unpleasant fumble because everything is almost, but not quite, familiar. Pretty soon you'll love IDEA and recognize that Eclipse has been bending you over and doing unspeakable things.
There are shared code formatting settings in the support directory (idea-settings.jar). This will prevent unexpected conflicts due to formatting.
Howard uses this ... because he can't manage to switch IDEs constantly (he uses Eclipse for training). Lately its gotten better.
As with IntelliJ, there are shared code formatting settings for Eclipse in the support directory (tapestry-indent-eclipse.xml).
All source files should have the ASF copyright comment on top, except where such a comment would interfere with its behavior. For example, component template files omit the comment.
As you make changes to files, update the copyright to add the current year to the list. The goal is that the copyright notice includes the year in which files change. When creating a new file, don't back date the copyright year ... start with the current year. Try not to change the copyright year on files that haven't actually changed.
IntelliJ has a great comparison view: Cmd-9 to see the local changes, the Cmd-D to see the differences. You can whip through the changes (using Cmd-forward arrow) and make sure copyrights are up to date as you review the changes prior to a commit.
Always provide a commit message. Howard generally tries to work off the JIRA, so his commit message is often:
TAP5-1234: Make the Foo Widget more Ajax-tastic!
It is very important to include the JIRA issue id in the commit. This is used in many places: JIRA links issues to the Git commits for that issue (very handy for seeing what changed as part of a bug fix). The Hudson CI server does as well, and will actually link Git commits to issues after succesfully building.
All Tapestry committers should be registerred with JIRA and part of the tapestry-developers JIRA group.
Every committer is invited to look at the list of 'Review for closing' issues and review them as it contains probably outdated or no more valid issues.
There's also a list of all Open issue about the project.
Ideally, we would always work top priortity to low priority. Howard sometimes jump out of order, if there's something cool to work on that fits in an available time slot. Alternately, you are always allowed to change the priority of a bug before or as you work it.
As a general rule issues which are "Invalid" or "Won't Fix" shouldn't have a "Fix version".
When you start to work on an issue, make sure it is assigned to you and use the start progress option.
Add comments about the state of the fix, or the challenges in creating a fix. This often spurs the Issue's adder to
provide more details.
Update the issue description to make it more legible and more precise if needed, i.e., "NPE in CheckUpdates" might become "NullPointerException when checking for updates to files that have been deleted". Verbose is good.
Is it a bug fix without tests? No. A good plan is to write a test that fails then work the code until the test passes. Often code works in a unit test but fails unexpectedly in an integration test. As the G-Man says "Expect unforeseen consequences".
When you check in a fix, you should close the issue and make sure the fix release is correct.
We're playing fast and loose – a better procedure would be to mark the bug resolved and verify the fix before closing it. That's ok, we have a community to double check our work .
For anything non-trivial, wait for the Hudson CI server to build. It catches a lot of things ... such as files that were not added to Git. And even IntelliJ has a bit of trouble with wildly refactored code. Hudson will catch all that.
Invalid issues and duplicates
Always provide comments about why_ an issue is invalid ("A Ruby implementation of Tapestry is out of scope for the project."), or at least, a link to the duplicate issues.
Consider writing new tests to prove that an issue is not valid and then leave the tests in place – then close the bug as invalid.
Close the issue but make sure the fix release is blank. Otherwise, the issue will be listed in the release notes, which we don't want.
Public vs. Private/Internal
This is a real big deal. As long as code is in the internal package, we have a high degree of carte-blanche to change it. As soon as code is public, we become handcuffed to backwards compatibility.
Interfaces are public, implementations are private. You can see this is the bulk of the code, where org.apache.tapestry5.services is almost all interfaces and the implementations are in org.apache.tapestry5.internal.services.
Many more services have both the interface and the implementation in org.apache.tapestry5.internal.services.
We absolutely do not want to make Page or ComponentPageElement public. You will often see public service facades that take a page name as a method parameter, and convert it to a page instance before invoking methods on internal services.
We do not have a specific plan for this yet. Future Tapestry 5 will add features to allow clean renames of parameters, and a way to deprecated and eventually remove components.
Tapestry uses interfaces quite extensively.
Interfaces fall into two categories: service interfaces called by user code, and interfaces implemented by user code.
Internal interfaces may be changed at any time. That's why so much is kept internal.
New methods may be added if absolutely necessary, but this should be avoided if at all possible. Don't forget the
@since Javadoc annotation.
Consider having a stable public facade service whose implementation calls into one or more internal service.
These should be frozen, no changes once released. Failure to do so causes non-backwards compatible upgrade problems; that is, classes that implement the (old) interface are suddenly invalid, missing methods from the (new) interface.
Consider introducing a new interface that extends the old one and adds new methods. Make sure you support both.
You can see this with ServiceDef and ServiceDef2 (which extends ServiceDef). Yes this can be a bit ugly.
Howard uses utility methods that convert from ServiceDef to ServiceDef2, adding a wrapper implementation around a ServiceDef instance if necessary:
Use of @since
When adding new classes or interface, or adding new methods to existing types, add an @since Javadoc comment.
Use the complete version number of the release in which the type or method was added: i.e., @since 126.96.36.199.
Code Style & Formatting
Yes, at one time Howard used leading underscores for field names. He has since changed my mind, but this unfortunately infected other people; please try to make your code blend in when modifying existing source.
Long ago, Tapestry (3) code used the regrettable "leading-I-on-interfaces" style. Don't do that. Instead, name the implementation class with an "Impl" at the end.
Howard prefers braces on a new line (and thus, open braces lined up with close braces), so that's what the default code formatting is set up for. It's okay to omit braces for trivial one-liner if statements, such as
if (!test) return;.
Indent with 4 spaces instead of tabs.
Use a lot of vertical whitespace to break methods into logical sections.
We're coding Java, not Pascal; it's better to have a few checks early on with quick returns or exceptions than have ten-levels deep block nesting just so a method can have a single return statement. In other words, else considered harmful. Low code complexity is better, more readable, more maintainable code.
Don't bother alphabetizing things, because the IDE lets you jump around easily.
Final is the new private. Final fields are great for multi-threaded code. Especially when creating service implementations with dependencies, store those dependencies into final fields. Once we're all running on 100 core workstations, you'll thank me. Seriously, Java's memory model is seriously twisted stuff, and assigning to a non-final field from a constructor opens up a tiny window of non-thread safety.
Comments are overwhelmingly important. Try to capture the why of a class or method. Add lots of links, to code that will be invoked by the method, to related methods or classes, and so forth. For instance, you may often have an annotation, a worker class for the annotation, and a related service all cross-linked.
Comment the interfaces and don't get worked up on the implementations. Javadoc does a perfectly good job of copying interface comments to implementations, so this falls under the Don't Repeat Yourself guideline.
Be very careful about documenting what methods can accept null, and what methods may return null. Generally speaking, people will assume that null is not allowed for parameters, and method will never return null, unless it is explicitly documented that null is allowed (or potentially returned).
Try and keep the documentation up-to date as you make changes; it is much harder to do so later. This is now much easier using the Confluence wiki (you're reading the result ).
Documentation was at one point the #1 criticism of Tapestry!
Class and Method Naming Conventions
Naming things is hard. Names that make sense to one person won't to another.
That being said, Howard has tried to be somewhat consistent with naming. Not perfectly.
A factory class creates new objects. Methods will often be prefixed with "create" or "new". Don't expect a Factory to cache anything, it just creates new things.
A source is a level up from a Factory. It may combine multiple factories together. It usually will cache the result. Method are often prefixed with "get".
Find vs. Get
For methods: A "find" prefix indicates that a non-match is valid and null may be returned. A "get" prefix indicates that a non-match is invalid and an exception will be thrown in that case (and null will never be returned).
A data object usually associated with a Tapestry IoC service's configuration.
Part of a pipeline, where there's an associated main interface, and the Filter wraps around that main interface. Each main interface method is duplicated in the Filter, with an extra parameter used to chain the interface.
Often a wrapper around a service configuration, it provides access to the contributed values (possibly after some transformation).
A method prefix that indicates a conversion or coersion from one type to another. I.e.,
An object that peforms a specific job. Workers will be stateless, but will be passed a stateful object to perform some operation upon.
An object whose job is to create other objects, typically in the context of creating a core service implementation for a Tapestry IoC service (such as PipelineBuilder or ChainBuilder).
An object that provides supporting operations to other objects; this is a kind of "loose aggregation".
A data object that holds a number of related values that would otherwise be separate parameter values to a method. This tends to streamline code (especially when using a Filter interface) and allows the parameters to be evolved without changing the method signature.
An object that "plugs into" some other code, allowing certain decisions to be deferred to the Strategy. Often a Strategy is selected based on the type of some object being operated upon.
Captures some stateful information that may be passed around between stateless services.
A non-instantiable class that contains public static fields that are referenced in multiple places.
An object that allows listeners to be registered. Often includes a method prefixed with "trigger" that will send notifications to listeners.
Objects that are exposed to user code should generally implement a meaningful toString() method. And that method should be tested.
You'll notice there isn't a lot of inheritance in Tapestry. Given the function of the IoC container, it is much more common to use some variation of aggregation rather than inheritance.
Where subclassing exists, the guideline for constructor parameters is: the subclass should include all the constructor parameters of the superclass, in the same positions. Thus subclass constructor parameters are appended to the list of super-class constructor parameters.