[asterisk-bugs] [Asterisk 0012658]: DTMF issues on Zap

[noreply at bugs.digium.com]

The following issue has been SUBMITTED. 
====================================================================== 
http://bugs.digium.com/view.php?id=12658 
====================================================================== 
Reported By:                dimas
Assigned To:                
====================================================================== 
Project:                    Asterisk
Issue ID:                   12658
Category:                   Core/PBX
Reproducibility:            always
Severity:                   minor
Priority:                   normal
Status:                     new
Asterisk Version:           SVN 
SVN Branch (only for SVN checkouts, not tarball releases):  1.4  
SVN Revision (number only!): 116466 
Disclaimer on File?:        N/A 
Request Review:              
====================================================================== 
Date Submitted:             05-15-2008 08:58 CDT
Last Modified:              05-15-2008 08:58 CDT
====================================================================== 
Summary:                    DTMF issues on Zap
Description: 
ohh. The long one.
replaces 11635 and 12592

I often have unreliable DTMF recognition on Zap channels. Research shown
that DSP works fine, and digits are read from the Zap driver correctly
almost 100% of time and corruption occurs later in the VLDTMF emulation
code.

So the problems I encountered:
1. There is a situation when DTMF emulation swaps digits. Scenario (user
presses 12 with short durations and short pause):

  A. ast_read is called by some application
  B. ast_read calls channel's read and gets DTMF_END(1)
  C. DTMF emulation is started and ast_read sets AST_FLAG_EMULATE_DTMF
flag and returns AST_FRAME_DTMF_BEGIN(1)
  D. application calls autoservice_start and AST_FLAG_END_DTMF_ONLY is set
on the channel
  E. autoservice thread calls ast_read
  F. eventually channel's technology diver returns DTMF_END(2).
  G. since AST_FLAG_END_DTMF_ONLY is set, ast_read passes DTMF_END(2) thru
ast_read returns DTMF_END(2). However it is still in the emulation mode!
  H. Eventually, emulation finishes and ast_read returns DTMF_END(1)

bottom line: 1 and 2 got swapped

Solution (one of, which I choosed) is to always queue DTMFs when we are
already in emulation mode. This fix immediately led to next issue:



2. when user quickly rolls over digits (1234) there can be a situation
when autoservice thread already read 12 and 34 are already queued into
channel's dtmfq but not yet read. Then when autoservice_stop "returns"
digits into channels readq, and they get processed by the VLDTMF emulation
code AGAIN - they will be queued to dtmfq. 

Result - 3412 will be read from the channel instead of 1234.

Solution: mark frames "returned" by autoservice and do NOT process them
again.

Major drawback: DTMFs which "passed thru" autoservice won't be re-emulated
so only DTMF_ENDs remain and no proper interval between them. In my case
when I mess with autoservice it is always an application terminates the
call so no worries - to my knowledge applications only need DTMF_END and
never BEGINs



3. The next problem is that the code at the beginning of ast_read which is
responsible for "injecting" deferred DTMF can inject content of dtmfq even
if there is non-empty readq (possible filled by just finished autoservice
which means they ae higher prioity).

Resul - again, swapped digits.

Solution: add additional check - do not inject digits from the dtmfq while
there is somethign in the channel's readq.


4. Finally AST_FLAG_EMULATE_DTMF is not always set when digit is added to
dtmfq. To make sure we ALWAYS queue when somethign is queued already, I
added "|| !ast_strlen_zero(chan->dtmfq)" to the conditions used to queue
DTMFs.
====================================================================== 

Issue History 
Date Modified   Username       Field                    Change               
====================================================================== 
05-15-08 08:58  dimas          Asterisk Version          => SVN             
05-15-08 08:58  dimas          SVN Branch (only for SVN checkou =>  1.4         
  
05-15-08 08:58  dimas          SVN Revision (number only!) => 116466          
======================================================================