From: Matt Gruenke (mgruenke_at_[hidden])
Date: 2005-09-02 14:39:07
Perhaps you might give some thought to how a user could instrument
synchronization primitives with tracing/logging? In my experience, the
lack of such a facility is a deficiency in boost locks. User-supplied
pre/post -acquisition + post-release callbacks might work, or making
some of the lock/mutex functions virtual.
If callbacks are used, I'm not picky about whether they're template or
constructor parameters. It would be useful if they had access to the
thread object (don't forget to define some reasonable behavior, for
main), or if you provide a TLS (Thread-Local Storage) interface.
BTW, TLS can be useful for other things, as well (e.g. building a
logging/trace system with a dynamic scoping capability).
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk