Nick, On 8/20/13 8:24 PM, Nick Williams wrote: > I ran in to a roadblock with this idea. Part of the byte code of a > class includes the fully-qualified class name. If I create a class, > say UnweavedClass, and replace its byte code in my fake transformer > with that of another class, the FQCN changes. This results in a > NoClassDefFoundError because the class loader is looking for > UnweavenClass in be in the byte code when really some other class > is. > > My backup idea is slightly less clean but, IMO, still more clean than > adding ASM as a test-time dependency and trying to figure all of that > out. I locally compiled fake "weaved" versions of the UnweavedClass > (with the modified behavior) and then translated each version into a > Java byte array definition. (These are extremely simple on-line, > one-method classes, so the byte arrays are relatively short.) I then > simply embedded the byte array definitions as static final byte[] > fields the test class and replaced the byte code in my fake > transformer with those embedded fields' content. I've tested this and > it works great.
Any reason not to simply compile some .java source into a .class file and read it from the disk instead of shoving it into a byte array? There's nothing stopping you from doing an offline-compile of a .java file into a .class file and committing both to svn. You don't have to compile the .java source as part of the test -- just load it off disk as part of the test. This will allow easier inspection of the .class file, and not be such a pain in the neck to change the bytecode if necessary. -chris
signature.asc
Description: OpenPGP digital signature