We introduced the dbUpdater routine to handle many small files
efficiently, but the folder stats call is almost equally expensive as it
results in two distinct write transactions to the database. This moves
it to the same routine.
(Doesn't make a *huge* difference with leveldb actually, but reduces the
50k-files benchmark time by 25% on my experimental bolt branch...)
Without this, when an index update comes in we only do a new pull if the
remote `localVersion` was increased. But it may not be, because the
index is sent alphabetically and the file with the highest local version
may come first. In that case we'll never do a new pull when the rest of
the index comes in, and we'll be stuck in idle but with lots of out of
sync data.
- Remove full details from ItemStarted (unnecessary, incorrect CamelCase)
- Add "type" ("file" or "dir") to both events
- Add "action" (what we tried to do - "delete" or "update") to both
events.
The "Invalid" config attribute is retained for errors discovered during
config loading (empty path, duplicate ID). This can only be set or
cleared at config loading time.
Errors discovered during runtime (I/O problems, etc) are now in the
folder state instead. Changes to these are sent as any other folder
state change.
We need an explicit chmod() when creating a new directory.
Otherwise a new directory may be created with a different permission
from the one received from an originating device, because the umask
is applied to the mode given to mkdir().
The incorrect permission is later sent back to the originating device
and the original permission will be lost.
