[Prev][Next][Index][Thread]
Experimental XMacTCP.lisp
You could try the following experimental version of MacTCP.lisp. It contains
some fixes from digitool, but it is not quite as efficient as we might like in
a released version. Additionally, there remain additional ways to crash,
possibly related to Apple's MACTCP and MCL's interface to it. You'll find some
commentary and testing code next to %tcp-control. If you find the same thread
entering %tcp-control twice while an asynchronous TCP task is pending but
using the same PB, it is a bug that can corrupt your heap. The solution is to
use the macro to cons a copy of PB on the stack and avoid the collision.
http://wilson.ai.mit.edu/cl-http/sources/mac/xMacTCP.lisp
Any additional ways this can happen should be reported to bug-mcl@digitool.com
and bug-mac-cl-http@ai.mit.edu
There is considerable interest in stamping out these MCL/TCP crashing
problems, so reporting Macbugs backtraces and careful descriptions of the
configuration could pay off in the short term.