[JSR308] Annotations on statements (and expressions?)

[Trevor Harmon]

On Jan 31, 2007, at 8:34 AM, Tom Ball wrote:

> I don't understand why a language like JML needs support from Java.  
> There's no need to turn Java into an intermediate language

Is JML truly a language? I thought it was really just annotations. So  
the issue is not "How can language X be integrated into Java?" but  
rather "How can annotation tool X use Java's annotation standard?"  
The latter question is important because otherwise we have all these  
incompatible annotation standards. For instance, tools that want to  
integrate with JML have to be written specifically for JML's  
proprietary annotations. If JML could be built to use Java's own  
annotation support, other tools could easily read and write its  
annotations, and I suspect JML itself would be smaller and simpler.  
Also, its annotation syntax would be consistent with the Java  
standard, which would make it a bit nicer to use.

WCET analysis tools face the same problem. Currently you find that  
every tool has a completely different syntax to express exactly the  
same semantics. For example, here's how you'd express loop bound  
annotations in some Java WCET tools:

Skånerost: /*$ loop-bound 100 */
XAC: //@ Loopcount(100)
WCA: //@ loop=100

Because of the differing syntax, these statements are incompatible,  
even though they all mean the same thing. If I were to write a lower- 
level WCET tool, it would need to parse these annotations, and I'd  
have to write three different parsers! Likewise, each new WCET tool  
has to bundle its own annotation parser even if it were to use the  
same syntax as one of these existing tools.

Wouldn't it make much more sense if WCET tools standardized on Java's  
built-in annotation support? But the problem for WCET tools is the  
same problem Gary described for JML: There's no way to add an  
annotation on a loop. Simply allowing this would go a long way toward  
convergence on a single annotation standard.

Trevor