Removed dependencies on portable-threads and trivial-garbage, plus minor multi-threading improvements.

darcs-hash:56b8fd98f2f784f29a7b1f62cd955a40790c5a69

darcs-hash:67f0f4e049799e85a6c1a266fcae8a0c3f1cd4a2

Another variation of making ContextL thread-safe. This time it should actually work ;), and it is again more fine-grained, that is, more efficient than the previous version.

darcs-hash:011b88f50371c2287cfd915f09bfb45b80d45feb

darcs-hash:c442d421467ec11c3072a14fd76efb7122f35609

Simplified mapping of layer-related names to internal names, which should also make things easier to read when debugging ContextL programs.

darcs-hash:1ef605b1748d8e95be7d9e33a12b10534ba82ed4

Whenever a layer definition is reinitialized (for example by calling reinitialize-instance or by evaluating a deflayer form for an already existing layer), all its entries in the layer activation caches are removed. This guarantees that adjoin-layer-using-class and remove-layer-using-class are called again when such a layer is activated or deactivated, even if those two generic functions have previously returned t as second values for that layer. Since layer-makunbound is implemented in terms of (setf find-class) and (setf class-name), which triggers a call to reinitialize-instance according to the CLOS MOP specification, this means that layer-makunbound causes the respective layer to be removed from the caches as well.

Likewise, if adjoin-layer-using-class and remove-layer-using-class are specialized on a particular layer metaclass, all entries for layers of that layer metaclass are also removed from the layer activation caches. This happens also when methods for these generic functions are redefined.

darcs-hash:3fa63d976bbdcff0ad9a75afc1fa0db696ec8139