Re: why was file_open not made async?

On Tue, Jun 05, 2001, Henrik Nordstrom wrote:
> Adrian Chadd wrote:
> >
> > Just curious - why was file_open() not turned into a callback
> > like file_read/file_write() were, so we could abuse them in
> > an async fashion?
>
> disk.c is ancient, and is more or less being killed at the moment, but
> if we are again to make generic a disk I/O layer then reimplementing
> disk.c async is one of the API's to the disk I/O layer. Note. for the
> sake of filedescriptor consistency and debugging the disk I/O layer
> should probably use a abstract filehandle rather than UNIX fd..
>
> Further note: It is VERY likely some stores cannot use the generic disk
> I/O layer easily, but it is for sure a good thing for logs the "normal"
> store implementation(s).

Yup.

I'm working on a new async disk io API. I'm not sure what should
happen with the file_*() routines - I was going to bring this up
at a later date.

My disk API uses a bunch of non-fd things - aio_queue_t for an
IO queue, aio_file_t for an open file, aio_queue_entry_t for
a queued IO event (open,create,read,write,close,unlink,stat).

Adrian

-- 
Adrian Chadd			<Liedra> Don't worry, I know who *you* are ;)
<adrian@creative.net.au> 	<spiv> We know everything about you.