[great-users] Udm CopyObjectHierarchy function duration increases during transformation

[Kevin Smyth]

Hi Tim,

Unfortunately, GME is slow and gets slower under the default configuration,
which enables it to roll back illegal operations. Here's code to disable
the roll-back capability in the GReAT code generator's ...._main.cpp output
file:

#include "Mga.idl" // from $(GME_ROOT)\Interfaces
....

// get objects from DNs
UMLModelTransformer::RootFolder rootUMLModelTransformer_552=
UMLModelTransformer::RootFolder::Cast(
uMLModelTransformer_ref_54b.GetRootObject());
StructureTestDSL::RootFolder rootStructureTestDSL_553=
StructureTestDSL::RootFolder::Cast(
structureTestDSL_ref_54e.GetRootObject());

// new code starts here:
// disable nested transactions for output model
IUnknown* gmeFco =
UdmGme::Udm2Gme(sdnUMLModelTransformer_549.GetRootObject());
IMgaObject* gmeFcoPtr = NULL;
gmeFco->QueryInterface(__uuidof(IMgaObject), (void**)&gmeFcoPtr);
IMgaProject* project = NULL;
gmeFcoPtr->get_Project(&project);
long prefs;
project->get_Preferences(&prefs);
project->put_Preferences(MGAPREF_NO_NESTED_TX | prefs);
gmeFcoPtr->Release();
project->Release();
gmeFco->Release();
// new code ends

// Add objects to input packets.
templateModels_1.push_back( rootStructureTestDSL_553);
trafos_3.push_back( rootUMLModelTransformer_552);
// Get access to temporary root object(s).
g_Crosslinks__GlobalRootContainer= Crosslinks::GlobalRootContainer::Cast(
crosslinks_ref_551.GetRootObject());
// transformation
Start_0 start_0;
start_0( templateModels_1, trafos_3);



Alternatively, you can commit the GME transaction before running
UdmUtil::CopyObjectHierarchy:
#include "UdmGme.h"
....
((UdmGme::GmeDataNetwork*)fromObj.__impl()->__getdn())->CommitEditSequence();

GME will be able to roll back operations, but this will reduce the slowdown.


You should add error checking, e.g. in case the output file is not a GME
file.

Kevin



On Thu, Sep 14, 2017 at 6:48 AM, Belschner, Tim <
tim.belschner at ils.uni-stuttgart.de> wrote:

> Hello  everybody,
>
>
>
> based on the generated code for the GReAT Group operator, we built up a
> cpp template for copying an abstract source object including its children.
> Since we would like to pass the copied object to an out-port within the
> same rule, the approach is to create a dummy object within the rule’s
> effector and replace it during the AttributeMapping by calling the
> aforementioned template:
>
>
>
> //++++++++++++++++++++++++++
>
> template <class T1, class T2, class T3 > bool
> UserLibCloneObjectAbstractSource( T1& fromObj, T2& toObj, T3& toParent)
>
> {
>
>                 T1 newObj= T1::Create( toParent );
>
>
>
>                 UdmUtil::copy_assoc_map copy_map;
>
>                 UdmUtil::CopyObjectHierarchy(fromObj.__impl(),
> newObj.__impl(), newObj.__impl()->__getdn(), copy_map);
>
>
>
>                 toObj.DeleteObject();
>
>                 toObj = newObj;
>
>
>
>                 return true;
>
> }
>
> // ++++++++++++++++++++++++++
>
>
>
> The result in the transformation’s output is as intended. But, the runtime
> of the function UdmUtil::CopyObjectHierarchy within the template increases
> with every call. After several calls, the execution of
> UdmUtil::CopyObjectHierarchy takes several seconds. My search within the
> Udm source code did not reveal any reason. This is why I would like to ask
> if someone could give me a hint or provide an alternative solution? For
> example an incorrect usage of the UdmUtil::copy_assoc_map.
>
>
>
> An example transformation with the necessary metamodel and inputs can be
> found here: https://belschner.org/nextcloud/index.php/s/RiMjSyd2yHKFYdT
> PW: GReAT123!
>
> The cpp template is located in the “UserLib_CloneObject.h”. The increasing
> runtime for a single “CopyObjectHierarchy” execution can be seen in the
> console output (Steps 2 to 3).
>
>
>
> Thanks in advance and best regards
>
>
>
> Tim
>
> _______________________________________________
> great-users mailing list
> great-users at list.isis.vanderbilt.edu
> http://list.isis.vanderbilt.edu/cgi-bin/mailman/listinfo/great-users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.isis.vanderbilt.edu/pipermail/great-users/attachments/20170920/0ec37273/attachment.html>