Description Andrew Clement 2008-06-20 13:07:32 EDT

We always support processing of old aspects.  It doesn't matter what level of AspectJ was used to build an aspect, as long as you use that version or a later version of the weaver, we can unpack it and don't require it to be rebuilt from source.  However, I've just encountered a .class apparently built with 1.5.4 that 1.6.1 cannot load.  It crashes deserializing a PointcutDeclaration.

In the data stream we have just read the numbers 1 and 3 indicating 'kinded pointcut' and then 'method-execution' and the next digit is a 0 when it should be 1-9.  We crash with a:

org.aspectj.weaver.BCException: weird kind 0
when batch building BuildConfig[null] #Files=43
	at org.aspectj.weaver.MemberKind.read(MemberKind.java:35)
	at org.aspectj.weaver.patterns.SignaturePattern.read(SignaturePattern.java:682)

The memberkind is a typesafeenum and so can never be other than 1-9.  It is the first part of a signaturepattern so hard to see how it got written out 'wrong' right now.

I've been told 1.5.4 can load this, so about to try that. Wow....1.5.4 did load it back in, how the hell.

I suspect we aren't consuming enough in 1.6.1 which then leaves us some extra that we interpret as a rogue pointcut.

Ok, in a comparison we consume one extra byte from the stream when reading it with 1.6.1 that we do not consume with 1.6.0 - at position 260.

As I got closer to it, I knew what it would be - especially when I knew it was just one byte difference.  The version check for whether the byte for 'annotation pattern relates to a parameter match' was wrong (urgh).