Interfaces and RTTI Issue: can someone confirm?
lazarus / 62-General

----------------------------------------------------------------------
terryc · 2015-12-09 11:44

Hi,
I'm using JSON and RTTI to store / load objects from a stream.
However, most of my objects are used/accessed by interfaces. In
addition, i have hierarchies of objects (and interfaces) to manage
complex models. I have found a situation where the RTTI does not seem
to set the interface reference properly. An example test project
demonstrating this is attached.

First there is a IMyObject and TMyObject as the root object. This
object will have its "test" property set:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i   TMyObject
= class( TInterfacedObject, IMyObject)    private      FTest :
IMyInterface;      function GetTest: IMyInterface;      procedure
SetTest(AValue: IMyInterface);    published      property Test :
IMyInterface read GetTest write SetTest;  end; 
Then this test property is used to store an object of type
TMyFailedInterface with a deep inheritance:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i  
TIntermediate = class( TInterfacedObject, IIntermediateInterface); 
TMyFailedInterface = class( TIntermediate, IMyInterface); 
the actual code for storing the property is somewhat like this (it
uses the intermediate interface as a common ancestor to all
interfaces):
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i     lInt2
:= lInt as IIntermediateInterface;    SetInterfaceProp( lMyObject,
'Test', lInt2); 
This actually works since SetInterfaceProp accepts any IUnknown. lInt
is of the proper interface type so it should work. However, when
accessing the property again, for example to execute a function, a
SIGSEGV error occurs:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i    lValue
:= lMyObject.Test.TestFunction; //this creates a SIGSEGV 
There is a workaround, but it's quite annoying to implement, this is
to declare ALL interfaces implemented for each class:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i  
TMyInterface = class( TIntermediate, IMyInterface,
IIntermediateInterface) In that case, the same SetInterfaceProp will
work, and the same access to the test property will work.

Please see the full test project in attachment for more details.
I have tested this on Windows with FPC 32-bit and Lazarus, FPC 3.1.1
SVN 49552.
1°) Can someone duplicate this behaviour?
2°) Is it a bug? Should i register it as such?
3°) Is there a better workaround?

Best regards,
TerryC

----------------------------------------------------------------------
Graeme · 2015-12-09 12:15

Quote from: terryc on December 09, 2015, 12:44:38 pmHowever, when
accessing the property again, for example to execute a function, a
SIGSEGV error occurs:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i    lValue
:= lMyObject.Test.TestFunction; //this creates a SIGSEGV What happens
if you use an interface instance to get to TestFunction, instead of
the object reference?

eg:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i var  intf:
ISomeInterface;begin   ...   if Supports(lMyObject, ISomeInterface,
intf) then     intf.TestFunction;   ...end; 
ps:
I haven't had time to look at your actual attachment.

----------------------------------------------------------------------
terryc · 2015-12-09 12:57

Same result:
Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i     //This
check fails with a SIGSEGV error:    lMyInt2 := lMyObject.Test;   
CheckEquals( lValue, lMyInt2.TestFunction, 'should be the
same');produces a SIGSEGV!

----------------------------------------------------------------------
Cyrax · 2015-12-09 17:13

You might be interested this FPC branch which have experimental
support for RTTI for interfaces.

Web view : http://svn.freepascal.org/cgi-
bin/viewvc.cgi/branches/interfacertti/?sortby=date&view=log

Subversion : http://svn.freepascal.org/svn/fpc/branches/interfacertti

----------------------------------------------------------------------
terryc · 2015-12-09 19:10

Many thanks Cyrax for the answer,
i will look at it. Up to now, i've tried to limit myself to production
code when using FPC.

BTW, there is an easy workaround that seems to work. I need to make
more indepth testing but the unit test is OK: it is to give an
IUnknown interface to SetInterfaceProp. Instead of giving an
intermediate Interface (in my case, it would be an ISerializable), one
should downcast the interface to an IUnknown and it works.

I've attached the updated Unit test project as a demo.

Code: Pascal  [Select][+][-]window.onload = function(){var x1 =
document.getElementById("main_content_section"); if (x1) { var x =
document.getElementsByClassName("geshi");for (var i = 0; i var 
lMyInt2      : IMyInterface;  lInt            : IUnknown;  [...]begin 
[...]  lInt  := lMyInterface;  SetInterfaceProp( lMyObject, 'Test',
lInt);  lMyInt2 := lMyObject.Test;  lValue :=
lMyInterface.TestFunction;  CheckEquals( lValue, lMyInt2.TestFunction,
'should be the same');

----------------------------------------------------------------------
Jonas Maebe · 2015-12-15 10:48

Could you create a small sample that does not depend on Lazarus/the
LCL, and file an FPC bug report at http://bugs.freepascal.org ?

Thanks.

----------------------------------------------------------------------
terryc · 2015-12-16 13:20

yes, I will do that.
Best regards,
TerryC