Conference pamsrc::objectbroker_development

2465.0. "Another memory leak!!" by HLIS31::OLERUD_G () Mon Apr 14 1997 03:57

    We have noticed a memory leak (again!!, this is a severe one) on the
    client side.
    
    If you have user defined types defined as:
                                              
        typedef struct tString_ 
        {
            string Ustr;
        } tString;
        typedef sequence<tString> tUseqtString;
        typedef sequence<string> tUseqUstr;
    
        typedef struct tUDT5_ {
            string          Ustr;
            tUseqUstr   	UseqUstr;
        } tUDT5;
        // typedef sequence<SecuritySet>    SecuritySets;
        typedef sequence<tUDT5> tUseqUDT5;
    
        typedef struct tUDT51_ {
            string          Ustr;
            tUseqtString   	UseqtString;
        } tUDT51;
        typedef sequence<tUDT51> tUseqUDT51;
    
    
    These types are used as attributes in an interface, e.g:
    
        interface i1
        {
            readonly attribute tUseqUDT5 useqUDT5;
            readonly attribute tUseqUDT51 useqUDT51;
        }
    
    Then useqUDT5 leaks but useqUDT51 doesn't. The difference between the
    two is that useqUDT51 has encapsulated the unbounded string sequence
    which generates a String_var.
    
    The server side doesn't leak on either useqUDT5 or useqUDT51 which is
    interersting.
    
    Gunnar (actually John Donaldson)
    

    By the way...
    
    OBB v2.7-11 on NT 4.
    
    As the .0 says, we can work round these client leaks
    by encapsulating all strings in structs (and presumably
    forcing the generation of string vars). However, we want
    to know: is this real?; is this going to be fixed?
    
    Also we are kind of surprised that the server 'knows'
    how to clean up these sequences and the client doesn't!
    
    John D. 

    The tUseqUstr sequence has a generated class (_ForSeq_var) that
    should be handling the memory allocation. We will have to 
    reproduce it here. If you have a sample client/server written
    that shows the problem, it would save us some time.
    

Peter, sorry for the delay. 

The problem is still there of course. I can give you
some sample code but not today.

John D.