[gst-devel] Re: filecache

[Joshua N Pritikin]

On Mon, Dec 09, 2002 at 12:13:32AM +0100, Wim Taymans wrote:
> The filecache looks good to me, I haven't gotten around to commiting
> this
> to CVS yet.

Yah, well, i should make sure it compiles first.  ;-)

i'll try to work on it tomorrow.

> mmaped binary search on disk is probably the way to go for
> this,
> maybe storing indexes (if needed in separate files or at the end of the
> file

Yah, i was planning to append binary data at the end of the file.
i dropped the directory idea.

> (which reminds me It needs a comit() API to tell the cache that
> no more entries are following)).

Yes please.

> I do think entries can be unsorted (PTS values of mpeg streams) you
> could get
> away with it by detecting discontinuities and making the binary search
> use a window
> to find the required entry (I didn't think hard about that).

Well, i currently index on the first format only.  If the other formats
are unsorted or partially sorted then it merely means that you can't
binary search on those formats.  Do we ever need to lookup at entry
by PTS?  Isn't PTS always a non-key?

> The GstCacheEntry is indeed just the data structure needed to implement
> the API, the
> underlying storage format can be totaly unrelated (database table?).

OK, just add a little documentation or hints about how to manage the
memory properly.  Or don't, and we can fix it later.

-- 
Victory to the Divine Mother!!         after all,
  http://sahajayoga.org                  http://why-compete.org