[cffi-devel] wanted? finalization - automatic deallocation of foreign memory
Hi, Another thing not yet covered by CFFI is finalization: one may wish for the foreign entity in the cl+ssl stream object to be free'd even when the stream is never closed by the application. Here again, implementations vary: Corman always adds a FREE finalizer to ct:malloc. CLISP does not (doing so would even break its :malloc parameter mode, where the foreign side does the free). Maybe finalizable foreign-alloc is something that CFFI should offer as an additional keyword? Then of course, one must think about what happens when foreign-free is called. I'd expect Corman to detect that, since there's not even a possibility to not have that automatic finalizer. CLISP could use invalid pointers to detect the duplicate free, but I'd expect implementations where raw pointers (or integers, e.g. Allegro) are used to not be able to detect duplicate free. Therefore, CFFI must probably specify that free is not allowed by the user anymore when (foreign-alloc :finalize t) is used. Hmm, with Allegro having pointers as ints, it will be hard to finalize that... Maybe they feature something else? Regards, Jörg Höhle
On Fri, 27 Jan 2006 17:10:55 +0100, "Hoehle, Joerg-Cyril" <Joerg-Cyril.Hoehle@t-systems.com> said:
Another thing not yet covered by CFFI is finalization: one may wish for the foreign entity in the cl+ssl stream object to be free'd even when the stream is never closed by the application. Here again, implementations vary: Corman always adds a FREE finalizer to ct:malloc. CLISP does not (doing so would even break its :malloc parameter mode, where the foreign side does the free).
Maybe finalizable foreign-alloc is something that CFFI should offer as an additional keyword?
Then of course, one must think about what happens when foreign-free is called. I'd expect Corman to detect that, since there's not even a possibility to not have that automatic finalizer. CLISP could use invalid pointers to detect the duplicate free, but I'd expect implementations where raw pointers (or integers, e.g. Allegro) are used to not be able to detect duplicate free. Therefore, CFFI must probably specify that free is not allowed by the user anymore when (foreign-alloc :finalize t) is used.
Hmm, with Allegro having pointers as ints, it will be hard to finalize that... Maybe they feature something else?
It might be better to do the finalization on the stream itself (you probably want it the close the handle anyway). __Martin
participants (2)
-
Hoehle, Joerg-Cyril -
Martin Simmons