Re: Which Transformation Pack Should I choose?

[Emil Slusanschi <slusanschi@sc.rwth-aachen.de>]

Hi Navindra,

Navindra Umanee wrote:
> Emil Slusanschi <slusanschi@sc.rwth-aachen.de> wrote:
>
> > The thing is, I don't want to insert all those temporary double 
> > variables and use 3 statements for just one assignment. I've solved that 
>
>
> May I ask why you're worried about temporaries? 

I want to be able to "read" and verify the generated/reverse engineered 
code that is being generated (java source files). If I keep generating a 
ton of temporaries (which is perfectly normal at the Jimple level ofc), 
I will eventually loose the overview.

> > by inserting a Grimp statement into the Jimple BodyTransformer that 
> > implements a "jtp.mypahse" phase:
> >
> > z.<DerivType: double value> = staticinvoke <java.lang.Math: double 
> > sin(double)>(x.<DerivType: double value>);
>
>
> I think it would be a bad idea to use Grimp stmts in this phase
> because it violates the 3-address code assumption of Jimple, and in
> more practical terms, you will confuse any Jimple analysis not
> expecting a Grimp statement.  You can pretty much expect undefined
> behaviour.

This is exactly why I asked about your opinions at this point, because 
of course it is not enough to have something "working", it is always 
necessary to have it working properly. Therefore my question about a 
prospective "gtp" pack. If you say I can do it in the "gop" pack, then 
I'll do it there, no problem.

On the other hand, I don't think I will have a Jimple analysis running 
in parallel with my transformations. I plan on making some code 
transformations (either at the Jimple or Grimple level), and only then, 
maybe optimize the resulting transformed piece of code. And yes, I also 
wanted to ask: is there a way of determining the _exact_ order in which 
the passes (packs and analyses) are executed. By this I mean, first a 
jop.cse and then my jtp.transform and then another gop.transform maybe 
and so on... Can it be done through the command line? The order of the 
arguments in the command line or something?

> > All good and well, but shouldn't I do this in a "gtp" pack? Because 
> > after all, this is more or less a Grimp transformation. And here is my 
> > problem: there is no "gtp" pack, or at least not yet ;-) What should I 
> > do? Continue like this, with a mix of Grimp and Jimple statements in a 
>
>
> You may use the "gop" pack for the same effect. The phase is off by
> default but you can enable it from the commandline with "-p gop on".
>
> Cheers,
> Navin.

Cheers,
Emil