Russell Bryant [Tue, 27 Nov 2007 23:50:25 +0000 (23:50 +0000)]
Merge changes from team/russell/iax2_frame_queue
This patch is an optimization for chan_iax2. This module is now heavily
multi-threaded. However, there is still a good number of globally shared
resources that prevent things from happen asynchronously. One of those things
was the global IAX frame queue. This queue was used to hold frames that have
been deferred for transmitting by another thread, and frames that may need to
get retransmitted.
I changed the frame queue to be per-call, since almost all of the frame queue
handling only cares about frames specific to a call number.
Two changes with regards to the 'eventwhencalled' option of queues.conf
1) Due to some signed vs. unsigned silliness, setting 'eventwhencalled' to
'vars' or 'yes' did exactly the same thing. Thus the sign change of the
ast_true call.
2) The vars2manager function overwrote a \n for every channel variable it parsed, resulting
in bizarre output for the channel variables. This patch remedies this.
(related to issue #11385, however I'm not sure if this will actually be enough to close it)
Russell Bryant [Tue, 27 Nov 2007 22:05:36 +0000 (22:05 +0000)]
Merged revisions 89790 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r89790 | russell | 2007-11-27 15:45:51 -0600 (Tue, 27 Nov 2007) | 41 lines
Merge changes from team/russell/autoservice_1.4
This set of changes fixes an issue that was reported to me on IRC yesterday.
The user, d1mas, was using chan_zap for incoming calls and was having DTMF
recognition issues in some situations. Specifically, he noticed that the
problem occurred when using DISA or WaitExten. He also noticed that when
using Read, the problem did not occur. His system also used DUNDi for
dialplan lookups.
So, he theorized that if the DUNDi lookups blocked for some period of time,
that audio from the zap channel could get lost. If the audio got lost, then
it wouldn't be run through the DTMF detector, and digits could get lost.
He was correct, and the following set of changes fixes the problem. However,
the changes go a little bit further than what was necessary to fix this exact
problem.
1) I updated pbx_extension_helper() to autoservice the associated channel to
handle cases where extension lookups may take a long time. This would
normally be a dialplan switch that does some lookup over the network, such
as the DUNDi or IAX2 switches.
This ensures that even while a DUNDi lookup is blocking, the channel will be
continuously serviced.
2) I made a change to the autoservice code. This is actually something that
has bothered me for a long time. When a channel is in autoservice, _all_
frames get thrown away. However, some frames really shouldn't be thrown
away. The most notable examples are signalling (CONTROL) frames, and DTMF.
So, this patch queues up important frames while a channel is in autoservice.
When autoservice is stopped on the channel, the queued up frames get stuck
back on the channel so that they can get processed instead of thrown away.
3) I made another change to the autoservice code to handle the case where
autoservice is started on channels recursively.
Previously, you could call ast_autoservice_start() multiple times on a
channel, and it would stop the first time ast_autoservice_stop() gets
called. Now, it will ensure that autoservice doesn't actually stop until
the final call to ast_autoservice_stop().
on second thought... revert all the other changes i've made in app options parsing leaving only one: if an empty argument is supplied for an option, set that argument pointer to point to an empty string rather than NULL, so that the application can do normal checks on it without worrying about it being NULL
Olle Johansson [Tue, 27 Nov 2007 19:24:17 +0000 (19:24 +0000)]
The following patch with updates for trunk. Works much better in trunk.
Also by accident fixed a bad typo by a previous committer, which actually made video calls
not work fully...
Merged revisions 89630 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
If we get a codec offer using a well-known payload type, but using it for another
codec that we don't know, Asterisk did not remove that codec from the list.
With this patch, we remove the codec from audio and video rtp objects and
deny it ever existed. Thanks to lasse for testing.
(closes issue #11376)
Reported by: lasse
Patches:
bug11376.txt uploaded by oej (license 306)
Tested by: lasse
Steve Murphy [Tue, 27 Nov 2007 18:50:44 +0000 (18:50 +0000)]
made AEL 8-bit transparent; mainly the lexer was tossing chars with the hi-order bit set. Not nice. Also, allow @ in extension names, and a backslash, also.
Steve Murphy [Tue, 27 Nov 2007 06:47:08 +0000 (06:47 +0000)]
Merged revisions 89622 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r89622 | murf | 2007-11-26 23:24:02 -0700 (Mon, 26 Nov 2007) | 1 line
closes issue #11379; OK, this is an attempt to make both sides happy. To the cdr.conf file, I added the option 'unanswered', which defaults to 'no'. In this mode, you will see a cdr for a call, whether it was answered or not. The disposition will be NO ANSWER or ANSWERED, as appropriate. The src is as you'd expect, the destination channel will be one of the channels from the Dial() call, usually the last in the list if more than one chan was specified. With unanswered set to 'yes', you will still see this cdr entry in both cases. But in the case where the dial timed out, you will also see a cdr for each line attempted, marked NO ANSWER, with no destination channel name. The new option defaults to 'no', so you don't see the pesky extra cdr's by default, and you will not see the irritating 'not posted' messages.
........
After issuing a "say load new", if a caller hangs up during the middle of playback of a number,
app_playback will continue to try to play the remaining files. With this change, no more files will
be played back upon hangup.
(closes issue #11345, reported and patched by IgorG)
After issuing a "say load new" tons of warning messages are printed
out to the CLI every time do_say in app_playback is called. Removing these
warnings
Fix issues with async dialing with an application executing. The application has to be terminated and control returned to the thread before hanging things up. (issue #BE-252)
Olle Johansson [Mon, 26 Nov 2007 19:24:23 +0000 (19:24 +0000)]
- Mark "concise" as deprecated
- Restructure other changes to UPGRADE.txt and CHANGES
We're still looking for scripts that replace
asterisk -rx "show shannels concise"
by using the manager interface, but still produces the same output.
Anyone?
Joshua Colp [Mon, 26 Nov 2007 18:11:31 +0000 (18:11 +0000)]
Perform some module use counting audits. This is now done outside the scope of the application/dialplan function so they do not need to worry about it.
Use ast_free to free memory, or else we shall implode if MALLOC_DEBUG is enabled.
(closes issue #11347)
Reported by: ys
Patches:
pbx.pbx_config.c.diff uploaded by ys (license 281)
Steve Murphy [Mon, 26 Nov 2007 17:26:01 +0000 (17:26 +0000)]
closes issue #11356; Many thanks to snuffy for his code review and changes to cut down duplication. I tested this against hashtest, and it passes. I reviewed the changes, and they look reasonable. I had to remove a few const decls to make things compile on my workstation,
Close the audio file before sending it to the post processing application.
(closes issue #11357)
Reported by: reformed
Patches:
mixmonitor.patch uploaded by reformed (license 330)
If channel allocation fails because the alert pipe could not be created also free the scheduler context.
(closes issue #11355)
Reported by: eliel
Patches:
main.channel.c.patch uploaded by eliel (license 64)
Joshua Colp [Mon, 26 Nov 2007 15:28:13 +0000 (15:28 +0000)]
Make the behavior of using /dev/urandom for random numbers the same as random().
(closes issue #11348)
Reported by: sperreault
Patches:
ast_random2.diff uploaded by sperreault (license 252)
We previously attempted to use the ESCAPE clause to set the escape delimiter to
a backslash. Unfortunately, this does not universally work on all databases,
since on databases which natively use the backslash as a delimiter, the
backslash itself needs to be delimited, but on other databases that have no
delimiter, backslashing the backslash causes an error.
So the only solution that I can come up with is to create an option in res_odbc
that explicitly specifies whether or not backslash is a native delimiter. If
it is, we use it natively; if not, we use the ESCAPE clause to make it one.
Olle Johansson [Sun, 25 Nov 2007 11:46:17 +0000 (11:46 +0000)]
- Deprecate "call-limit" in chan_sip. No other channel driver enforces call-limits
and we now have the groupcount system to implement call-limits in the dialplan. You
can use the "setvar" option in realtime/sip.conf to set limits per device.
- Implement "callcounter" as a new option to enable the call counting we need to
report device status to queue, manager and SIP subscriptions.
The call counter setting is now enabled in the code by setting the device call-limit
to 999. When we remove the call limit, we can simply enable this with a boolean
setting.
Steve Murphy [Sat, 24 Nov 2007 21:00:26 +0000 (21:00 +0000)]
closes issue #11363; where the pattern _20x. buried in an included context, didn't match 2012; There were a small set of problems to fix: 1. I needed NOT to score patterns unless you are at the end of the data string. 2. Capital N,X,Z and small n,x,z are OK in patterns. I canonicalize the patterns in the trie to caps. 3. When a pattern ends with dot or exclamation, CANMATCH/MATCHMORE should always report this pattern, no matter the length. With this commit, I also supplied the wish of Luigi, where the user can select which pattern matching algorithm to use, the old (legacy) pattern matcher, or the new, trie based matcher. The OLD matcher is the default. A new [general] section variable, extenpatternmatchnew, is added to the extensions.conf, and the example config has it set to false. If true, the new matcher is used. In all other respects, the context/exten structs are the same; the tries and hashtabs are formed, but in the new mode the tries are not used. A new CLI command 'dialplan set extenpatternmatch true/false' is provided to allow switching at run time. I beg users that are forced to return to the old matcher to please report the reason in the bug tracker. Measured the speed benefit of the new matcher against an impossibly large context with 10,000 extensions: the new matcher is 374 times faster.
Luigi Rizzo [Sat, 24 Nov 2007 13:57:46 +0000 (13:57 +0000)]
remove a DEBUG_THREADS message that accesses private lock fields.
If needed, the code to extract this information should be implemented
in some generic header or library and the function called here.
Currently, zero-length voicemail messages cause a hangup in VoicemailMain.
This change fixes the problem, with a multi-faceted approach. First, we
do our best to avoid these messages from being created in the first place,
and second, if that fails, we detect when the voicemail message is
zero-length and avoid exiting at that point.
Reported by: dtyoo
Patch by: gkloepfer,tilghman
(Closes issue #11083)
Up until this point, the XML output of the manager has been technically
invalid, due to the repetition of certain parameters in a single event.
This caused various issues for XML parsers, some of which refused to parse
at all, given the invalidity of the rendered XML. So this commit fixes
the XML output, ensuring that each entity parameter has a unique name, thus
ensuring valid XML.
Reported by: msetim
Patch by: tilghman
(Closes issue #10220)
Luigi Rizzo [Fri, 23 Nov 2007 15:54:13 +0000 (15:54 +0000)]
put in the necessary hooks for video support in the console.
This is a NOP as far as the current code is concerned,
but there is already support in ./configure and the
Makefiles for the various libraries used by console_video.c
(not yet in the tree) so addition is trivial.
Olle Johansson [Fri, 23 Nov 2007 09:40:02 +0000 (09:40 +0000)]
Let's start with implementing the base architecture for UTF8 caller ID's
so we can handle multiple formats properly. This is not carved in stone,
but a proposal to start with.
We need to add support for transliterations as well as UTF8 handling,
propably with libiconv. Murf is looking into that for the dialplan.
Luigi Rizzo [Thu, 22 Nov 2007 03:50:04 +0000 (03:50 +0000)]
shuffle a little bit the content of header files to reduce dependencies.
In this commit:
- move the ast_register/unregister_app functions to module.h
to avoid the need to include pbx.h for the simpler apps;
- move the ast_group structure to channel.h to remove the
dependency of app.h on linkedlists.h
Note, this is a long process that I am doing in small steps.
The main difficulty is that now for each subsystem we
have a single header (e.g. channel.h) included by the subsystem
provider (usually one file, e.g. channel.c) and by its clients
(dozens of them, e.g. we have some 70+ apps and 30+ functions).
This requires the clients to include all the extra headers
required by the provider (eg. lock.h, linkedlists.h, definitions
of substructures...) even though many of the clients would be
just happy with opaque struct declarations and function prototypes.
The long term plan is to eventually rectify this structure
so that the compilation can become faster, and also APIs
are more stable.
Steve Murphy [Wed, 21 Nov 2007 23:54:12 +0000 (23:54 +0000)]
closes issue #11285, where an unload of a module that creates a dialplan context, causes a crash when you do a 'dialplan show' of that context. This is because the registrar string is defined in the module, and the stale pointer is traversed. The reporter offered a patch that would always strdup the registrar string, which is practical, but I preferred to destroy the created contexts in each module where one is created. That seemed more symmetric. There were only 6 place in asterisk where this is done: chan_sip, chan_iax2, chan_skinny, res_features, app_dial, and app_queue. The two apps destroyed the context, but left the contexts. All is fixed now and unloads should be dialplan friendly.
Steve Murphy [Wed, 21 Nov 2007 20:38:00 +0000 (20:38 +0000)]
closes issue #11290; the proposed patch was a good guess, and would solve the bug to some extent, but was really masking the real issue, that there were bad entries in the table. This fix removes the condition that the hashtab updates be done on exten removal only when the pattern_tree was present, which is silly. The operations that apply to the pattern tree are instead made conditional. Also, threw back in routines that kpfleming deleted because of probs in the 64-bit world. Tested on both 32 and 64-bit machines (compile). Tested the reload problem with over 20 reloads, and no problems. If you find more problems, please reopen 11290.
Changing an inaccurate debug message to be less inaccurate. Under the circumstances, this
message would always report that there were 0 members available, even though that may not be true.
If a channel gets masqueraded in the middle of a park, don't play the
announcement to the masqueraded channel, and dial back to the original
channel on timeout.
Tilghman Lesher [Wed, 21 Nov 2007 18:38:18 +0000 (18:38 +0000)]
Change Read to set READSTATUS as an indication of the result
Also, some cleanup to CHANGES.
Reported by: michael-fig
Patch by: michael-fig,tilghman
(Closes issue #11004)
projects / thirdparty / asterisk.git / log
