From: Miklos S. <mi...@sz...> - 2006-08-16 14:51:29
|
> Ouch! I just modified my little copy test program to only push one byte > at a time and to perform a fflush after every byte to the fuse file and > there are a flush(es) performed on the file stream in what appears to be > the time domain, but there is no flush before the release! There are a > number of writes that occur to the file before the release, so this data > is not flushed at all. > > Not being able to pass close errors now becomes a problem. In my little > example program many writes are passed into the file and then the flush > of the writes is never performed. > > I suggest that fuse itself insert a flush to the file before it executes > a release. This flush can be appropriately flagged. > > Any suggestions? Comments? For normal I/O flush() will be called after any writes, where you can return errors. For mmaped I/O this is not guaranteed, but mmaped writes are not currently supported anyway. Miklos |