gstreamer/docs/random/ensonic/embedded.txt
Stefan Kost df19b21c56 docs/design/draft-klass.txt: Add example that needs more thinking.
Original commit message from CVS:
* docs/design/draft-klass.txt:
Add example that needs more thinking.
* docs/design/draft-missing-plugins.txt:
More thoughts about wtrapper plugins.
* docs/random/ensonic/embedded.txt:
* docs/random/ensonic/profiling.txt:
More design work.
2007-03-28 13:44:41 +00:00

57 lines
2.1 KiB
Text

$Id$
= embedded =
== index handling ==
For avidemux I currently have a big patch doing memory optimized index handling.
It basically thins out the index to save memory. Right now it only keeps index
entries marked with the avi keyframe flag.
In gstreamer core we have some indexing objects. They are curently used nowhere.
The idea is to use them and to make the index strategy plugable or configurable
at run time.
The challenge is then to rewrite muxers and demuxers to use them instead of the
built in index logic.
This way the different requirements of desktop and embedded platforms could be
encapsulated in the indexer strategy.
== ranking ==
Autopluggers like playbin and decodebin use the element caps plus static ranks
to create piplines.
The rank of an elemnt right now refers to the quality/maturity of the element.
Elements with higher rank should be functionaly more complete. If we have
multiple elements that are feature complete there is a draw.
There are more decission criteria thinkable:
* target processor (CPU,DSP,GPU)
* ressource usage (CPU, memory)
* license (proprietary, open source)
* quality (in terms of the audio/image quality)
One problem of taking criteria like quality and performance into account when
autoplugging, is that elements might have objects-properties that influence
them.
Beside adding more ranking criteria, we might consider adding an overridable
mechanism for picking elements (basically registry filters that not decide on
the base of ranks). Applications might choose the filter mechanism based on the
use-case (which gstreamer not know right now).
== g_malloc vs. g_try_malloc ==
g_malloc/g_new should not be used when allocating space of a size that is
derived from a value found in the file. Instead one should use g_try_malloc or
g_try_new and check the return value.
$ find . -name "*.c" -exec grep "g_malloc" {} \; | wc -l
190
$ find . -name "*.c" -exec grep "g_try_malloc" {} \; | wc -l
0
$ find . -name "*.c" -exec grep "g_new" {} \; | wc -l
398
$ find . -name "*.c" -exec grep "g_try_new" {} \; | wc -l
3
This is not strickly related to embedded, but its easier to crash gstreamer on
embedded this way.