Re: Boehm-Demers-Weiser conservative garbage collector and GNAT

[Martin Krischik <krischik@users.sourceforge.net>]

Florian Weimer wrote:

> Martin Krischik <krischik@users.sourceforge.net> writes:
> 
>> I am wondering if it is possible to implement a storage pool using
>> the Boehm-Demers-Weiser conservative garbage collector in GNAT.
> 
> There are a few pitfalls.  For example, objects with a controlled
> component are always reachable from the collector's point of view, so
> they can never be collected (in fact, the Ada semantics of
> finalization mandate this behavior).  Furthermore, the collector can
> hugely benefit from very basic type information (proper flagging of
> pointer-free objects and objects without aliased components).

The Boehm-Demers-Weiser conservative garbage collector has atomic memory
which must be pointer free. But this is used only for performance
improvemend.

> However, the real killer for the storage pool approach is the fact
> that the collector will neither track properly the root set, nor task
> creation.  You can solve the first problem with a custom
> implementation of System.Memory (allocating all objects using the
> collector); some more run-time library tweaking can solve the second.

Current status is that I implemented the thin and thick bindung to the
collector and added the collector to my Trace class. Trace is used in any
class and task so it a good test. But I will add further tests.

> (However, I still don't know if Boehm-Demers-Weiser collector is
> actually compatible with GNAT's tasking implementation.)

All seem to work with OS/2. I havn't had time for testing Linux. The mingw
used with GNAT 3.15 Windows does not support all the signaling needed so I
can't link libgc.a into the test programs.

Any help is still be appreciated.

With Regards

Martin.

-- 
mailto://krischik@users.sourceforge.net
http://www.ada.krischik.com