On Mon, Feb 20, 2012 at 10:17 AM, Dan Dennedy <d...@dennedy.org> wrote: > 2012/2/17 Maksym Veremeyenko <ve...@m1stereo.tv>: >> 15.02.12 20:49, Dan Dennedy написав(ла): >> [...] >> >>> In the pixbuf patch, you should not call the convert_image virtual >>> function directly; use mlt_frame_get_image(). >> >> use mlt_frame_get_image for converting image did not work - call to that >> function return blank frame... > > It will only work when you call mlt_frame_set_image beforehand, which > I believe you had done in the previous patch. Now you are not, so > yeah, it is ok to call convert_image. > >> i updated patch and added the same functionality to pango producer, but >> calling convert_image virtual function directly left till i find another way >> to convert it... > > This is not fault of yours, but I can simply not include yet another > form of caching into this pixbuf producer. I need to review the old > cache (as opposed to usage of mlt_cache) first.
The old cache driven by the "cache" property caches an entire image sequence into memory, which has very limited usage. The only time it makes sense is when you have a short sequence and you reuse that producer. The only thing I can see really using this is some unknown playout server that smartly reuses a producer for an animated lower third or watermark. Note this does not cache the colorspace-converted images. The purpose of mlt_cache for "pixbuf.pixbuf" is to recycle the unscaled image because the resolution can change from one frame to the next by the requester (e.g. affine transition). The usage of mlt_service_cache prevents too many GdkPixbuf objects in memory at the same time in a complex composition. We had kdenlive users using hundreds of image clips - not an image sequence, but a laboriously constructed slideshow. These pixbuf objects hold open a file descriptor exhausting max file handles per process as well consuming memory. The purpose of mlt_cache for "pixbuf.image" is to recycle the scaled image. Again, the usage of mlt_service_cache helps reduce memory consumption. A mlt_service_cache applies across all instances of a particular service in a graph. IOW, only up to 10 are actually in memory while the least recently used get flushed. Your patch does not take advantage of this. Imagine a few hundred digital photos referenced by a Kdenlive project. Once you play through the project, all of those images (albeit scaled and converted) are being cached in producer_pixbuf.clone.image. I think "pixbuf.image" should hold the converted image, but the refresh_image() function needs refactoring to simplify it before we can do anything more to it. Also, I maintain producer_qimage in parallel with this, so it will need the same treatment. Finally, both of these are also pending an enhancement to image sequence loading that j-b-m requested. I will start the refactoring now. -- +-DRD-+ ------------------------------------------------------------------------------ Try before you buy = See our experts in action! The most comprehensive online learning library for Microsoft developers is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3, Metro Style Apps, more. Free future releases when you subscribe now! http://p.sf.net/sfu/learndevnow-dev2 _______________________________________________ Mlt-devel mailing list Mlt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mlt-devel
