| 1 |
1) Goal: |
| 2 |
======== |
| 3 |
The goal of this document is to analyze current problems |
| 4 |
in media type detection as we currently handle it in |
| 5 |
GStreamer (as of 27/9/2003), and how these can be solved. |
| 6 |
This touches upon typefinding, autoplugging and (optionally) |
| 7 |
bytestream. |
| 8 |
|
| 9 |
2) Typefinding, bytestream & autoplugging: |
| 10 |
========================================== |
| 11 |
|
| 12 |
bytestream: |
| 13 |
----------- |
| 14 |
currently, bytestream collects incoming buffers and adds |
| 15 |
them up (gst_buffer_merge ()). From this, a subbuffer is |
| 16 |
created, which is inexpensive. In case of filesrc, the |
| 17 |
merging is not expensive, too (mmap ()). However, in any |
| 18 |
other source case, _merge () needs a new buffer plus copy |
| 19 |
of the data. This is plain wrong. Source elements need to |
| 20 |
be able to support a _read ()- instead of a _get ()-based |
| 21 |
way of providing data to the pipeline on their choice. To |
| 22 |
the rest of GStreamer, _get () and _read () are the same, |
| 23 |
the only difference is that _read () also requests a size |
| 24 |
of the buffer to be returned. |
| 25 |
|
| 26 |
Surely, this does not mean that bytestream will read any |
| 27 |
buffer size that is requested from it plainly, this would |
| 28 |
be ineffective. It is still allowed to cache (although the |
| 29 |
kernel will do this too...). |
| 30 |
|
| 31 |
typefinding: |
| 32 |
------------ |
| 33 |
the typefind function type is currently defined as: |
| 34 |
|
| 35 |
typedef struct _GstTypeDefinition { |
| 36 |
gchar *name; |
| 37 |
gchar *mimetype; |
| 38 |
gchar *extension; |
| 39 |
GstTypefindFunc func; |
| 40 |
} GstTypeDefinition; |
| 41 |
|
| 42 |
typedef (GstCaps *) (* GstTypefindFunc) (GstBuffer *buffer, |
| 43 |
gpointer private); |
| 44 |
|
| 45 |
GstTypeFactory * gst_type_factory_new (GstTypeDefinition *def); |
| 46 |
|
| 47 |
Although is is unclear what private is and how to use it |
| 48 |
in a plugin. ;). The current approach has one large |
| 49 |
disadvantage: the plugin cannot control the input for type |
| 50 |
detection. Therefore, if the incoming buffer is not large |
| 51 |
enough, typefinding will inappropriately fail. This is |
| 52 |
unacceptable. The plugin needs to control input data flow |
| 53 |
itself, so that we will have less false negatives and/or |
| 54 |
will need only one cycle through the plugins to find the |
| 55 |
type of a data stream. |
| 56 |
|
| 57 |
Therefore, I propose the following change to the typefind |
| 58 |
system: |
| 59 |
|
| 60 |
typedef (GstCaps *) (* GstTypefindFunc) (GstBytestream *input, |
| 61 |
gpointer private); |
| 62 |
|
| 63 |
and |
| 64 |
|
| 65 |
GstTypeFactory * gst_type_factory_new (GstTypeDefinition *definition, |
| 66 |
gpointer data); |
| 67 |
|
| 68 |
The data gpointer will be provided as second argument to the |
| 69 |
typefind function and is for private use to the plugin. |
| 70 |
|
| 71 |
There is one rule: at the end of typefinding, the plugin needs |
| 72 |
to take care that the state of the bytestream is exactly the |
| 73 |
same as before typefinding. It may cache data, but it may not |
| 74 |
skip (and therefore lose) data. If the bytestream supports |
| 75 |
seeking, this is easy: simply seek back to 0 (start of stream) |
| 76 |
after typefinding. If it does not, then you need to assure |
| 77 |
that you only used _peek (), not _read () or _flush (). |
| 78 |
|
| 79 |
The caller of the typefind function is responsible for creating |
| 80 |
the bytestream and for emptying the cache and reusing it in the |
| 81 |
data stream after the typefind function returns. |
| 82 |
|
| 83 |
spider: |
| 84 |
------- |
| 85 |
Imo, spider should use GstTypefind (a public element) for |
| 86 |
typefinding. Ideally, it would derive from it. |
| 87 |
|
| 88 |
GstTypefind emits a signal when a type is found, and furtherly |
| 89 |
only has a sink pad. the derived elements from this should |
| 90 |
implement anything needed to make a proper autoplugger. |
| 91 |
|
| 92 |
3) Status of this document |
| 93 |
========================== |
| 94 |
Proposal, pending to be implemented. Target release is 0.8.0 or |
| 95 |
any 0.7.x release. |
| 96 |
|
| 97 |
4) Copyright and blabla |
| 98 |
======================= |
| 99 |
(c) Ronald Bultje, 2003 <[email protected]> under the |
| 100 |
terms of the GNU Free Documentation License. See http://www.gnu.org/ |
| 101 |
for details. |