Re: [fuse-devel] inefficiency of open/close on every read() resp. write() call (as in example)

[Miklos S. <mi...@sz...>]

> Did anyone do some sort of benchmark on how expensive the
> open/read/close sequence is as presented in the fuse example code
> xmp_read(...)?  I mean it seems so expensive to do this every single
> time a read is issued?
> 
> Why doesn't the example instead store the pointer of an open call
> and use this at subsequent calls?

See fusexmp_fh.c

> And also, what would once and for all be a reasonable way to handle
> this pending open file? Any drawbacks doing it that way?

Not for the example filesystem.  It may make sense for filesystems not
having an "open file" concept.

But probably the stateless model (fusexmp.c) is less often used than
the stateful one (fusexmp_fh.c).

Miklos