Community
Participate
Working Groups
It wwill be nice if AspectJ can avoid generation of runtime check for if primitive pointcut if expression can be eveluated at compile time. The most important example is deecting access to self fields. Such optimization is very important for persistence and remoting aspects. If we access field of "foreign" object, then object has to be loaded first. If we access field of this object (most common case), then no extra checks are needed. Pleas notice that exact check is not needed. For example: MyObject o = this; o.foo = 1; It is not critical that here access to "foo" will be not be treated as "this" access. In AspectJ poincut for matching access to non-this fields can be written in the following way: nonSelfGet(Peristent self, Persistent dst): get(* Persistent.*) && target(dst) && this(self) && if(dst != self); Unfortunatelly this pointcut requires runtime check, which makes this optimization completely useless. Instead of two byye code instructions ApsectJ inserts 20 instruction, which cause increase of byte code about 10 times and about 3 times decrease speed: 0: aload_0 1: getfield #27; //Field x:I -------- 0: aload_0 1: astore_1 2: aload_0 3: aload_1 4: invokestatic #110; //Method DynamicTest$CheckFieldAccess.ajc$if_0: (LDynamicTest;LDynamicTest;)Z 7: ifeq 19 10: invokestatic #102; //Method DynamicTest$CheckFieldAccess.aspectOf:() LDynamicTest$CheckFieldAccess; 13: aload_0 14: aload_1 15: invokevirtual #106; //Method DynamicTest$CheckFieldAccess.ajc$before$DynamicTest$CheckFieldAccess$543: (LDynamicTest;LDynamicTest;)V 18: aload_1 19: getfield #27; //Field x:I But in most cases it is possible to calculate this condtion during compile time! For all "this" accesses (in terms of Java fields access without specifying base object, in terms of byte code - fields of instance method which targt is loaded using aload_0 instruction), specified if condition is always false. So no runtime check needs to be inserted. Also it may be possible to calculate at compile time some other kinds of if conditions, for example instanceof checks. Certainly it is not so important, because most of such checks can be rewritten using other pointcuts which do not require runtime checks. May be there are other useful conditions which can not be expressed in other way and which can be calculated at compile time. So I wonder if this change request will be consiidered to be useful and can be easily implemented.
This is a useful optimization idea, but it is unlikely to be implemented any time soon without a patch contributed that implements it. The only reason not to do this optimization is the time required to implement and test it carefully; however, that's a powerful reason not to do it without a contribution.
Still a perfectly valid enhancement request, but we have no immediate plans to implement this in the near future so moving to "LATER" status for consideration in future planning activities.
LATER/REMIND bugs are being automatically reopened as P5 because the LATER and REMIND resolutions are deprecated.
