Age | Commit message (Collapse) | Author |
|
|
|
unprotected, but fortunately this time I was able to use a read lock which is
much lest nasty and persistant.
|
|
syncing changes. That was important so that more changes wouldn't be added
while we're writing, but it also meant that in a very particular circumstance
where one thread was syncing changes, another was marking changes as having
happened, and another locked an object that was being written because it had
changes they would all deadlock. This should be fixed now.
|
|
important, they are references to constant data, weather they should be able to
become bound still.
|
|
deleted before the sync could occur properly. They are now ignored during
sync.
|
|
it still has active references, and it will be safely cleaned up when the last
reference is released.
|
|
|
|
|
|
things seem to work. We've also tested against LLVM.
|
|
cache information before they are deserialized from storage. This changed the
signature of the cache loading template function, but the new function isn't
harder to use, and provides the key information as well.
|
|
after they are saved. It needs some more testing, but the theory is sound.
|
|
to be re-engineered to actually do a lazy erase like removing an iNode in a
posix filesystem. While the file remains open it exists, but the references
visible to the user on disk are gone.
|
|
|
|
key is loaded then the query doesn't even get to the superclass.
|
|
now ready for actual use.
|