Add triggered field access information analysis and use for call graphs #211
Conversation
|
Looks good at a first glance, but please factor out common code between the two schedulers (e.g., the used and derived properties). |
maximilianruesch
force-pushed
the
feature/add-triggered-field-access-analysis
branch
from
c428580 to
e37fdbd
Compare
This comment was marked as resolved.
This comment was marked as resolved.
|
|
||
| override def derivesEagerly: Set[PropertyBounds] = Set.empty | ||
|
|
||
| override def derivesCollaboratively: Set[PropertyBounds] = derivedProperties |
There was a problem hiding this comment.
Why do you need to override these, can't you specify them in the super trait already? I wouldn't expect us to ever add a lazy scheduler where they wouldn't fit.
There was a problem hiding this comment.
Since these functions stem from the specific BasicEagerFPCFAnalysisScheduler and BasicTriggeredFPCFAnalysisScheduler traits and not from the generic FPCFAnalysisScheduler trait that the super trait of these extends, they are correct here. It also only makes sense that you can specify e.g. eager behavior when you know you have an eager analysis.
There was a problem hiding this comment.
No, they stem from ComputationSpecification, the super trait of FPCFAnalysisScheduler. It's fine like this, though.
There was a problem hiding this comment.
It seems the Github UI is not fit for determining the origin of super functions...
The current field access information analysis is eager. Using it in call graphs usually results in many upfront calculations, growing directly proportional to the number of methods under analysis, regardless of whether they are reachable or not.
This PR adds a triggered analysis for field accesses tor rectify this issue and thus brings the field access information analysis in line with other call graph related analysis and others that are used during call graph computation. Now, field accesses are only computed for methods that are deemed reachable, which is also the only case call graph analyses need such information.
Thus, the points to keys are reconfigured to use the new triggered analysis instead of the previous eager analysis.