Range check error on POST request on Linux

[BlackLanzer]

Describe the bug
If you send an array in POST on Linux the server crash with Range check error on this line https://github.com/andrea-magni/MARS/blob/master/Source/MARS.Rtti.Utils.pas#L255.
On Windows it works fine.

To Reproduce
Steps to reproduce the behavior:

1. Send an array in POST

Expected behavior
The program should not crash

Scenario (please complete the following information):

• OS: Linux

[meg02in]

I have the same problem in my project. I fixed the issue modifing the procedure in that way:

//procedure SetArrayLength(var AArray: TValue; const AArrayType: TRttiType; const ANewSize: Integer);
procedure SetArrayLength(var AArray: TValue; const AArrayType: TRttiType; const ANewSize: PNativeInt);
begin
  if AArrayType is TRttiArrayType then
  begin
    raise Exception.Create('Not yet implemented: SetArrayLength TRttiArrayType');
    { TODO -oAndrea : probably not needed }
  end
  else if AArrayType is TRttiDynamicArrayType then
    //DynArraySetLength(PPointer(AArray.GetReferenceToRawData)^, AArrayType.Handle, 1, @ANewSize);
    DynArraySetLength(PPointer(AArray.GetReferenceToRawData)^, AArrayType.Handle, 1, ANewSize);
end;

and the changing accordly the unit MARS.Core.JSON.pas and MARS.Core.MessageBodyReaders.pas.

After that my project working good both under linux and under wndows in all kind of deploying method for the server tha MARS support.

I attached my source of MARS.Rtti.Utils, MARS.Core.JSON.pas and MARS.Core.MessageBodyReaders.pas

RangeCheckFixing.zip

[BlackLanzer]

Nice find @meg02in, but your fix has another bug.

In MARS.Core.MessageBodyReaders.pas and MARS.Core.JSON.pas you define the variable L for the new size of the array as an Integer.
Later DynArraySetLength take a pointer to NativeInt. In my tests this caused a wrong read of the pointer value. I think it's because of 64 or 32 bit stuff.

I fixed it defining L as NativeInt and now it's working correctly on both Windows and Linux.

I can make a pull request later if everything is good.

[meg02in]

Hi,
thanks for your suggestion, i made your correction and tested for 1 week in a server of one of my customer.

All gone right

[andrea-magni]

Hi @meg02in and @BlackLanzer , please go ahead and submit your PR! Thanks!
Please beware I have updated a bit the develop branch today so you may want to make a Pull before proceeding.

Waiting for your PR :-)

[andrea-magni]

For your convenience, I've just merged your corrections in the attached source files.
Please check everything is as you like and submit your PR, I will be happy to accept it.
If you could just rename that L variable to something like LNewLength I would appreciated it.

Thanks again :-)

Source.zip

[andrea-magni]

This issue has been fixed thanks to the pull request #83
Many thanks for your contribution! :-)