Hi
On 18/12/2015 11:03, Ed Willink wrote:
The New UpperToLower tests fail intermittently; WIP.
Problem solved. Intermittent due to a Set-orginated freedom. Failing
due to missing code for accumulating result in a connection variable
for a regular mapping.
The schedule is temporarily deterministically stupid to stress the
extra intermediate connection variables for unhelpful mapping
sequencing. I'll add a unit test option so that normal use is good,
but test use can be stupid.
I'll try to ensure that the interpreter shares the same run-time
support code as the CG evaluator.
Done. The Interpreter tests now pass with the New Scheduler.
testExample4_Interpreted remains commented out for model2 that
fails. The constraints that must be observed by the QVTi scheduler
are now documented in the EPackage comment of QVTimperative.ecore.
Sharing the code required opening up the API buried in
org.eclipse.ocl.pivot.evaluation.tx, so the changes require the
following OCL commit:
commit 7242d1a0faf7475fb56cfcc6cfdaa7ae0d509ffe
[480567] Refactor Transformer API for extensibility
Sharing the code solves the CG / Interpreter inconsistencies but
temporarily loses the preliminary support for incremental
transformation execution.
Since the new scheduler now passes all its tests, I have promoted it
to master. If you use it, you must also use OCL master. An I-build
will be available shortly.
The Transformer API will probably evolve slightly once incremental
support is re-instated and fully developed, so if you don't need the
new scheduler stick with OCL M4a and QVTd M4.
Regards
Ed Willink
|