[patch] some fixes for is_mounted detection

[Kay Sievers]

On Mon, 2004-07-05 at 21:08 +0200, David Zeuthen wrote:
> On Sun, 2004-07-04 at 21:01 +0200, Sjoerd Simons wrote:
> > > >   Second at the end of block_class_pre_process both detect_media and
> > > >   etc_mtab_process_all_block_devices are called, but these only work on
> > > >   devices in the gdl and at that time the device is still in the tdl. So
> > > >   they're basically nops there. To solve this i've adjusted the
> > > >   ClassDeviceHandler stuff to also call the post_merge function. And let
> > > >   the one for block devices call detect_media and
> > > >   etc_mtab_process_all_block_devices..
> > > 
> > > This is a good point but it seems wrong to use post_merge for this.
> > > Perhaps we should have an added_to_gdl method on both ClassDeviceHandler
> > > and BusDeviceHandler that is called when it's in the GDL?
> > 
> > It depends on what exactly is ment by merge, if it's merging new info into a
> > device a added_to_gdl is nicer. If it's merging new information (either
> > properties updates of a new device) into the GDL, then using post_merge is fine.
> > 
> > Both meanings sound fine to me and the last one was the least work, so i choose
> > that when writing the patch :)
> > 
> 
> Ok, as discussed on IRC last night I've now added a in_gdl method to
> both BusDeviceHandler and ClassDeviceHandler - for the latter it is only
> invoked for class devices we add into the tree, so it sort of
> complements post_merge that only applies to class devices merged onto
> bus devices. So that's nice.
> 
> With this it was trivial to move the detect_media() and
> etc_mtab_process_all_block_devices() function calls into the new in_gdl
> method and I believe this fixes the bug you had? Anyway, if you can test
> the latest from CVS, I'd appreciate it!
> 
> So, It actually turns out this fixes some problems I've had as well as
> bringing a new one to the table! Now inserting a floppy into my USB
> floppy drives makes volume_id_probe take a long time since floppies are
> so slow. 
> 
> Now, I think I know how to solve this, basically, in volume_id.c:
> volume_id_probe(), move the probe_vfat() call to the top of the ones we
> try to probe for, but the call to get_buffer() still takes a long time -
> Kay: can we postpone the get_buffer() call to after probe_vfat()?

Hmm, most of the floppys are legacy ones, so we will not poll and never
notice a media change. Isn't it better not to read the label at all for
the floppies?
You can remove the first get_buffer in volume_id_probe():ALL comletely,
and move the MSDOS and VFAT up, to test if it's getting faster.

I just thought it was a good idea to read the buffer first (have no
floppy :))

Kay


_______________________________________________
hal mailing list
hal at freedesktop.org
http://freedesktop.org/mailman/listinfo/hal