Community
Participate
Working Groups
copy of my proposal for @AJ if pointcut I came up with today and that everyone agree with --- Here are my thought on the if() pointcut for @AJ stuff. I don't think it is reasonable to have string based if(....) bodies so here is the proposal I am starting on - the main idea is : the @Pointcut annotated method acts as the if body (hence the if body is already at hand etc etc) so simple ! Alex // if() is somehow implicit ie we can know we have an if when we have // a static boolean @Pointcut annotatated method // TODO TBD -may help toString(), parsing and tools if we require the if() marker @Aspect public class IfAspect { static int I; @Pointcut("execution(* Foo.do()) && if()") // if(IfAspect.test()) public static boolean test() { return I > 0; } @Pointcut("execution(* Foo.do()) && if()") // if(IfAspect.test2(thisJoinPoint) public static boolean test2(JoinPoint jp) {// will require JP flags return jp.getSignature().getName().startsWith("baz") && I > 0; } @Pointcut("execution(* Foo.do(int)) && args(val) && if()") public static boolean test3(int val) {//use of formal binding is straightforward return val + I > 0; } @Before("test()") public void doBefore() { System.out.println("IfAspect.doBefore"); } @Before("test2()") public void doBefore2() { System.out.println("IfAspect.doBefore2"); } @Before("execution(* Foo.do()) && if()") // enclosingDef is not a @Pointcut returning boolean, with public static access public void doBefore3() { System.out.println("NO WAY - compiler error"); } } ReplyReply to allForward Adrian Colyer to me, Andrew, Jonas More options 11:25 am (5 hours ago) I like the idea of the method containing the body of the if() pointcut - very elegant! I don't think we want to make the if() completely implicit, so: @Pointcut("execution(* Foo.*(..))") public static boolean fooThings() { return true; } should be an error, and @Pointcut("execution(* Foo.*(..)) && if(true)") public void fooThings() {} would also be an error (not allowed to specify body for if() pointcut in annotation style), but @Pointcut("execution(* Foo.*(..)) && if()") public static boolean fooThings() { return true; } is great! -- Adrian Adrian_Colyer@uk.ibm.com Alexandre Vasseur <avasseur@gmail.com> 08/06/2005 10:02 Please respond to Alexandre Vasseur <avasseur@gmail.com> To Adrian Colyer/UK/IBM@IBMGB, Andrew Clement/UK/IBM@IBMGB cc Jonas Bonér <jboner@gmail.com> Subject if pointcut in @AJ - Show quoted text - Here are my thought on the if() pointcut for @AJ stuff. I don't think it is reasonable to have string based if(....) bodies so here is the proposal I am starting on - the main idea is : the @Pointcut annotated method acts as the if body (hence the if body is already at hand etc etc) so simple ! Alex // if() is somehow implicit ie we can know we have an if when we have // a static boolean @Pointcut annotatated method // TODO TBD -may help toString(), parsing and tools if we require the if() marker @Aspect public class IfAspect { static int I; @Pointcut("execution(* Foo.do()) && if()") // if(IfAspect.test()) public static boolean test() { return I > 0; } @Pointcut("execution(* Foo.do()) && if()") // if(IfAspect.test2(thisJoinPoint) public static boolean test2(JoinPoint jp) {// will require JP flags return jp.getSignature().getName().startsWith("baz") && I > 0; } @Pointcut("execution(* Foo.do(int)) && args(val) && if()") public static boolean test3(int val) {//use of formal binding is straightforward return val + I > 0; } @Before("test()") public void doBefore() { System.out.println("IfAspect.doBefore"); } @Before("test2()") public void doBefore2() { System.out.println("IfAspect.doBefore2"); } @Before("execution(* Foo.do()) && if()") // enclosingDef is not a @Pointcut returning boolean, with public static access public void doBefore3() { System.out.println("NO WAY - compiler error"); } } ReplyReply to allForwardInvite Adrian to Gmail Alexandre Vasseur to Adrian, Andrew, Jonas More options 1:52 pm (2½ hours ago) Great - I reach the same "if() is required" conclusion that was one of the last burden we had to go thru in terms of semantics equivalence. There is one implicit consequence: an anonymous pointcut (@Before(...)) cannot contain an if() in the @AJ style. That's the small special case to pay... Alex - Show quoted text - On 6/8/05, Adrian Colyer <adrian_colyer@uk.ibm.com> wrote: > > I like the idea of the method containing the body of the if() pointcut - > very elegant! I don't think we want to make the if() completely implicit, > so: > > @Pointcut("execution(* Foo.*(..))") > public static boolean fooThings() { > return true; > } > > should be an error, and > > @Pointcut("execution(* Foo.*(..)) && if(true)") > public void fooThings() {} > > would also be an error (not allowed to specify body for if() pointcut in > annotation style), > > but > > @Pointcut("execution(* Foo.*(..)) && if()") > public static boolean fooThings() { > return true; > } > > is great! > > -- Adrian > Adrian_Colyer@uk.ibm.com > > > > Alexandre Vasseur <avasseur@gmail.com> > > 08/06/2005 10:02 > > Please respond to > Alexandre Vasseur <avasseur@gmail.com> > > > To Adrian Colyer/UK/IBM@IBMGB, Andrew Clement/UK/IBM@IBMGB > > cc Jonas Bonér <jboner@gmail.com> > > Subject if pointcut in @AJ > > > > > > Here are my thought on the if() pointcut for @AJ stuff. > I don't think it is reasonable to have string based if(....) bodies so > here is the proposal I am starting on - the main idea is : the > @Pointcut annotated method acts as the if body (hence the if body is > already at hand etc etc) > > so simple ! > Alex > > // if() is somehow implicit ie we can know we have an if when we have > // a static boolean @Pointcut annotatated method > // TODO TBD -may help toString(), parsing and tools if we require the > if() marker > @Aspect > public class IfAspect { > > static int I; > > @Pointcut("execution(* Foo.do()) && if()") // if(IfAspect.test()) > public static boolean test() { > return I > 0; > } > > @Pointcut("execution(* Foo.do()) && if()") // > if(IfAspect.test2(thisJoinPoint) > public static boolean test2(JoinPoint jp) {// will require JP flags > return jp.getSignature().getName().startsWith("baz") && I > 0; > } > > @Pointcut("execution(* Foo.do(int)) && args(val) && if()") > public static boolean test3(int val) {//use of formal binding is > straightforward > return val + I > 0; > } > > @Before("test()") > public void doBefore() { > System.out.println("IfAspect.doBefore"); > } > > @Before("test2()") > public void doBefore2() { > System.out.println("IfAspect.doBefore2"); > } > > @Before("execution(* Foo.do()) && if()") // enclosingDef is not a > @Pointcut returning boolean, with public static access > public void doBefore3() { > System.out.println("NO WAY - compiler error"); > } > } > > ReplyReply to allForward Alexandre Vasseur to Adrian More options 4:12 pm (15 minutes ago) I actually like that a lot: @Pointcut("if()") public static boolean testONE(JoinPoint jp) { System.out.println("\tIfAspect.testONE"); return jp.getSignature().getName().startsWith("do"); } ie a simple method gets turned into a full blown join point testing library using regular Java
Some short note to recap' on the rules: "if()" can only appear in a @Pointcut(..) expression, and appears as is (no if(....)). The such annotated method that usually is a "public void <name>(<formals>) {}" must then be a "public static boolean <name>(<formal || implicit JoinPoint stuff>) { <test body> }" Pointcut composition happens as usual ie formal AND implicit JoinPoint stuff must be then bounded: @Pointcut("if() && args(i)") public static boolean test(JoinPoint jp, int i, JoinPoint.EnclosingStaticPart esjp) { //... see implicit jp and esjp but explicit i return true; } @Before("test(jp, val, enc) && someOtherPointcut()") // all bound public void beforeStuff(JoinPoint jp, JoinPoint.Enclosin.. enc, int val) { ... }
assigned to Adrian since some work for AJ-JDT stuff: 1/ in IfPointcut, testmethod can be null sometime (at least a NPE happens now). The important thing is that I handle the IfPointcut as for @style IF the extraParamFlag == -1. So testMethod can perhaps be there (that would perhaps save a lookup if it is there - but then it must be the user if body and not the ajc_if_xxx one..) 2/ language rules have changed a bit (see below) To work on it use the AtAjSyntaxTest, testIf method, and uncomment in syntax.xml where there is a <!-- FIXME AV - Adrian in JDT stuff --> note. Thanks / (I wish I could do that myself - if so, just tell me if I should and what is the process, where is the readme to do it etc etc) D:\java\jdk1.5.0_01\bin\java -Didea.launcher.port=7534 "-Didea.launcher.library=D:\Program Files\IntelliJ-IDEA-4.5\bin\breakgen.dll" -Dfile.encoding=windows-1252 -classpath "D:\java\jdk1.5.0_01\jre\lib\charsets.jar;D:\java\jdk1.5.0_01\jre\lib\deploy.jar;D:\java\jdk1.5.0_01\jre\lib\javaws.jar;D:\java\jdk1.5.0_01\jre\lib\jce.jar;D:\java\jdk1.5.0_01\jre\lib\jsse.jar;D:\java\jdk1.5.0_01\jre\lib\plugin.jar;D:\java\jdk1.5.0_01\jre\lib\rt.jar;D:\java\jdk1.5.0_01\jre\lib\ext\dnsns.jar;D:\java\jdk1.5.0_01\jre\lib\ext\localedata.jar;D:\java\jdk1.5.0_01\jre\lib\ext\sunjce_provider.jar;D:\java\jdk1.5.0_01\jre\lib\ext\sunpkcs11.jar;D:\aw\cvs_aj\org.aspectj-HEAD\modules\tests\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\testing-drivers\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\bridge\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\lib\junit\junit.jar;D:\aw\cvs_aj\org.aspectj-HEAD\modules\util\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\weaver\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\runtime\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\bcel-builder\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\lib\regexp\jakarta-regexp-1.2.jar;D:\aw\cvs_aj\org.aspectj-HEAD\modules\testing-util\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\lib\jdiff\jdiff.jar;D:\aw\cvs_aj\org.aspectj-HEAD\modules\asm\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\testing\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\lib\ant\lib\ant.jar;D:\aw\cvs_aj\org.aspectj-HEAD\modules\lib\ant\lib\ant-launcher.jar;D:\aw\cvs_aj\org.aspectj-HEAD\modules\testing-client\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\ajde\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\org.aspectj.ajdt.core\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\org.eclipse.jdt.core\jdtcore-for-aspectj.jar;D:\aw\cvs_aj\org.aspectj-HEAD\modules\taskdefs\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\lib\commons\commons.jar;D:\java\jdk1.5.0_01\lib\tools.jar;D:\aw\cvs_aj\org.aspectj-HEAD\modules\loadtime\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\aspectj5rt\bin;D:\aw\cvs_aj\org.aspectj-HEAD\modules\loadtime5\bin;D:\Program Files\IntelliJ-IDEA-4.5\lib\idea_rt.jar" com.intellij.rt.execution.application.AppMain com.intellij.rt.execution.junit2.JUnitStarter -ideVersion5 org.aspectj.systemtest.ajc150.ataspectj.AtAjSyntaxTests,testIfPointcut LOADING SUITE: ..\tests\src\org\aspectj\systemtest\ajc150\ataspectj\syntax.xml TEST: IfPointcutTest ...test "IfPointcutTest" failed Unexpected error messages: error at (no source information available) C:\Temp\ajcSandbox\ajcTest26370.tmp\ataspectj\IfPointcutTest.java:0::0 Internal compiler error java.lang.NullPointerException at org.aspectj.weaver.patterns.IfPointcut.write(IfPointcut.java:120) at org.aspectj.weaver.patterns.AndPointcut.write(AndPointcut.java:119) at org.aspectj.weaver.ResolvedPointcutDefinition.write(ResolvedPointcutDefinition.java:72) at org.aspectj.weaver.AjAttribute$PointcutDeclarationAttribute.write(AjAttribute.java:349) at org.aspectj.weaver.AjAttribute.getBytes(AjAttribute.java:57) at org.aspectj.weaver.AjAttribute.getAllBytes(AjAttribute.java:70) at org.aspectj.ajdt.internal.compiler.ast.EclipseAttributeAdapter.getAllBytes(EclipseAttributeAdapter.java:31) at org.aspectj.org.eclipse.jdt.internal.compiler.ClassFile.addAttributes(ClassFile.java:676) at org.aspectj.org.eclipse.jdt.internal.compiler.ast.TypeDeclaration.generateAttributes(TypeDeclaration.java:594) at org.aspectj.org.eclipse.jdt.internal.compiler.ast.TypeDeclaration.generateCode(TypeDeclaration.java:578) at org.aspectj.org.eclipse.jdt.internal.compiler.ast.TypeDeclaration.generateCode(TypeDeclaration.java:620) at org.aspectj.org.eclipse.jdt.internal.compiler.ast.TypeDeclaration.generateCode(TypeDeclaration.java:560) at org.aspectj.org.eclipse.jdt.internal.compiler.ast.TypeDeclaration.generateCode(TypeDeclaration.java:628) at org.aspectj.org.eclipse.jdt.internal.compiler.ast.CompilationUnitDeclaration.generateCode(CompilationUnitDeclaration.java:182) at org.aspectj.org.eclipse.jdt.internal.compiler.Compiler.process(Compiler.java:513) at org.aspectj.org.eclipse.jdt.internal.compiler.Compiler.compile(Compiler.java:330) at org.aspectj.ajdt.internal.core.builder.AjBuildManager.performCompilation(AjBuildManager.java:727) at org.aspectj.ajdt.internal.core.builder.AjBuildManager.doBuild(AjBuildManager.java:206) ... at java.lang.reflect.Method.invoke(Method.java:585) at com.intellij.rt.execution.application.AppMain.main(AppMain.java:78) error at public static boolean positive(int i) { C:\Temp\ajcSandbox\ajcTest26370.tmp\ataspectj\IfPointcutTest.java:57:0::0 Pointcuts should have an empty method body error at public static boolean positive(int i) { C:\Temp\ajcSandbox\ajcTest26370.tmp\ataspectj\IfPointcutTest.java:57:0::0 Methods annotated with @Pointcut must return void Unexpected fail messages: abort ABORT -- (NullPointerException) null null
I'll look into this shortly - and also write down any process hints for what has to be done in situations like this...
reassign to Adrian for JDT fix
fixed in ValidateAtAspectJAnnotationsVisitor and IfPointcut. When I enable the referenced tests, they still fail but with formal unbound in pointcut errors. Is this expected?? If not we can work some more on this tomorrow.
Yes IfPointcut and IfPointcut2 should pass. They pass when -Xdev:NoAtAspectJProcessing is used (see syntax.xml). When I uncomment the FIXME in syntax.xml : The first one complains with a null body for "positive" method but the method does have a body. The second one complains with a formal unbound, but all is bound (and the bcel weaver when -Xdev:NoAtAspectJProcessing is used does not complains).
ok, will look into it this morning then.
Problem with IfPointcutTest was that when looking for the matching method containing the test, we iterated over getDeclaredJavaMethods. When the @aspect was compiled by ajc, this isn't quite right as the @pointcut method becomes treated like a pointcut declaration. Fix was just to iterate over getMethods. IfPointcut2Test exercised a part of ValidateAtAspectJAnnotationsVisitor that we hadn't gone through before. This was missing the logic to create **Implicit**FormalBindings instead of regular FormalBindings for our special types. Both tests are now passing for me. I'm just running the full suite and then I'll commit if everything goes ok.
ok, fix is committed.
great! I could not figure out those details in order to fix that one.
was M3 remind
