Re: Garbage Collection in Ada

[stt@houdini.camb.inmet.com (Tucker Taft)]

Brian Rogoff (rogoff@sccm.Stanford.EDU) wrote:

: "Norman H. Cohen" <ncohen@watson.ibm.com> writes:
:    Robert A Duff wrote:

:    > In article <mheaney-ya023280000311962335300001@news.ni.net>,
:    > Matthew Heaney <mheaney@ni.net> wrote:
:    > >Language designers: any reason why overloading "all" would be too hard?
:    > 
:    > Well, functions return values (or constant objects, if you want to be
:    > precise in Ada 95 terms), whereas "P.all" returns a variable (if P is an
:    > access-to-variable type).  So it wouldn't work.  

:    The same scheme that Ada 95 uses to allow programmers to redefine the
:    behavior of "new" could have been used to allow redefinition of the
:    behavior of ".all".  The type System.Storage_Pools.Root_Storage_Pool
:    could have been given another primitive operation to be overridden by
:    those defining their own storage-pool types:

:       procedure Dereference
: 	 (Pool                     : in out Root_Storage_Pool;
: 	  Storage_Address          : in out Address;
: 	  Storage_Size_In_Elements : in Storage_Elements.Storage_Count;
: 	  Alignment                : in Storage_Elements.Storage_Count);

: This looks almost exactly like a proposal I read (a long time ago, on a web 
: page long since forgotten) for adding persistence to Ada 95. The main 
: difference is that Storage_Address was of type "Access_Type", which was a 
: generic formal parameter to the Storage_Pools package, and another parameter 
: to Dereference was a System.Address type. There was also a Rereference 
: procedure for converting System.Address types back to Access_Types.

:    <... snipping description of some the things this might allow ...>

: This seems, at first naive glance, to provide all of the hooks to do GC and 
: more, in a very Ada-like way. The idea seems to have been around for a 
: while too, perhaps hundreds of web-years (at least 10 months). So what are 
: the drawbacks? 

The problem has to do with knowing when the user is "done" with
the result of Dereference, and whether it is a read or a write
reference.  Remember that it is permissible to pass the result of
X.all to a very long subprogram, or to rename it.
In general an operation like this would work for a transaction-oriented
persistence mechanism, where you "commit" changes as a separate
operation, and pages stay in virtual memory indefinitely once referenced
(or until explicitly committed).

: -- Brian

-Tucker Taft   stt@inmet.com   http://www.inmet.com/~stt/
Intermetrics, Inc.  Cambridge, MA  USA