Expand codec bitfield from 32 bits to 64 bits.

Reviewboard: https://reviewboard.asterisk.org/r/416/


git-svn-id: http://svn.digium.com/svn/asterisk/trunk@227580 f38db490-d61c-443f-a65b-d21fe96a405b

1 files changed, 47 insertions, 0 deletions

diff --git a/doc/codec-64bit.txt b/doc/codec-64bit.txt
new file mode 100644
index 000000000..8f2ceec59
--- /dev/null
+++ b/doc/codec-64bit.txt
@@ -0,0 +1,47 @@
+CODEC BIT EXPANSION
+-------------------
+
+The code base up to and including Asterisk 1.6.2 has a basic limit of 32 codecs
+recognizable, due to the use of a 32-bit integer for the codec bitmask.  We
+have expanded the number of available codecs from 32 to 64, through the use of
+an immutable type, called format_t.  This should make future expansion to even
+more bits more easily done.
+
+The design of this expansion has made some changes to the architecture of codecs
+in order to accomplish this task.  I will attempt to enumerate them here.
+
+The initial set of 32-bits were allocated as the first 16 to audio codecs, the
+next 8 to video codecs, and the remaining to text codecs (which are used for
+fax capabilities).  Initially, there is an assumption in the code that all
+audio codecs are contiguous, followed by a contiguous set of video codecs.
+After the conversion, this assumption will no longer be true.  The codec bits
+for the existing codecs will continue to be allocated as-is, and the additional
+codec bits should be allocated on an as-needed basis, with audio codecs
+occupying slots 32-47 and video codecs occupying slots 48-62 (with a 0-based
+offset).  Slot 63 is reserved and should not be allocated; it is used in code
+as an end condition for iterating through the entire set of codecs.
+
+The frame structure has been altered.  Initially, the subclass held an integer
+whose meaning was specified by the frametype.  If the frametype was
+AST_FRAME_VOICE, the subclass specified the audio codec.  If the frametype was
+AST_FRAME_VIDEO, the subclass specified the video codec, with the 0-bit set to
+specify a key frame.  This was done with a union on the subclass, where the
+"integer" union member specifies the traditional 32-bit subclass and the "codec"
+union member specifies the new 64-bit codec bitmask.  This additionally
+guarantees that code compiled under the old scheme will need to be altered to
+compile under the new scheme, which helps avoid incorrect assumptions about the
+state of code which might otherwise compile without errors.
+
+The IAX2 code initially used a 32-bit integer IE to specify both the codec as
+well as the preferred format.  An additional IE has been added, which specifies
+a single byte version number as the initial part of the data.  This version
+number is initially specified as 00 and requires 8 bytes to follow, specifying
+the 64-bit codec bitmask, in network-byte order.  This schema should allow
+further codec expansion in the future without allocation of any additional IEs.
+
+Little changes are required to support further codec expansion in the future,
+though the majority of the work has already been accomplished.  Specifically,
+the bitwise operations that are immutable operations in the gcc compiler will
+need to be altered to handle larger bitmasks.  Additionally, the constants that
+define specific codecs will need to be changed from integers to structures.
+