[asterisk-dev] [Code Review]: PreDial continuation - Ability to run dialplan on callee and caller channels before Dial

[rmudgett]

> On April 27, 2012, 5:28 a.m., schmidts wrote:
> > /trunk/apps/app_dial.c, lines 2229-2230
> > <https://reviewboard.asterisk.org/r/1878/diff/1/?file=27477#file27477line2229>
> >
> >     is there a special need to alloce these both in the while loop? maybe it would be better to do this before the while.

Declaring variables does not incur overhead.  Initializing them does.  Also declaring them here reduces their scope to where used.


> On April 27, 2012, 5:28 a.m., schmidts wrote:
> > /trunk/apps/app_dial.c, line 2641
> > <https://reviewboard.asterisk.org/r/1878/diff/1/?file=27477#file27477line2641>
> >
> >     why do you alloc number here? AFAIK will this be done in setvar_helper (ast_var_assign) itself.

ast_strdupa() copies the given string onto a stack allocated variable and cannot fail unless the stack is blown.  Note the name is not ast_strdup().

number is copied here because I am taking the value from the peer channel and putting it into the chan channel.  Also the peer channel lock is being released between the get and set so I don't need to hold both channel locks at the same time.  To copy a channel variable value safely between two channels either requires holding both channel locks while transferring the value or copying the value out of one channel while it is locked and then putting it into the other channel.


- rmudgett


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviewboard.asterisk.org/r/1878/#review6084
-----------------------------------------------------------


On April 23, 2012, 1:19 p.m., rmudgett wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviewboard.asterisk.org/r/1878/
> -----------------------------------------------------------
> 
> (Updated April 23, 2012, 1:19 p.m.)
> 
> 
> Review request for Asterisk Developers.
> 
> 
> Summary
> -------
> 
> This review takes over from the https://reviewboard.asterisk.org/r/1229/ predial review and addresses the concerns brought up by the last review there.
> 
> PreDial
>   
>   Say SIP/abc is calling SIP/def
>   You have: Dial(SIP/def)
>   SIP/def-123234 is created.  But how can you tell that from dialplan?
> 
>   You can use a pickup macro: M or U options to Dial(), but you have to wait till pickup to know.
>   'PreDial' new option 'b' to Dial(), will let you run dialplan on the newly created channel before it is connected to the end-device.
> 
>   New way:
>   Dial(SIP/def,,b(predial^s^1))
>   Dialplan will run on SIP/def-123234 and allow you to know right away what channel will be used, and you can set specific variables on that channel.
> 
> You can also run dialplan on the caller channel (option 'B') right before the dial, which is a great place to do a last microsecond UNLOCK to ensure good channel behavior.
> Example:  LOCK(foo)
>           do stuff
>           UNLOCK(foo)
>           Dial(SIP/abc)
> 
> With this above example, say SIP/123 and SIP/234 are running this dialplan.
> 
> SIP/123 locks foo
> SIP/123 unlocks foo
> due to some cpu load issue, SIP/123 takes its time getting to Dial(SIP/abc) and doesn't do it right away
> 
> Meanwhile... SIP/234 zips right by, lock 'foo' is already unlocked, it grabs the lock, does its thing and it gets to Dial(SIP/abc).  SIP/123 wakes up and finally gets to the Dial().  Now you have two channels dialing SIP/abc when there was supposed to be one.
> 
> If your intention is to ensure that Dial(SIP/abc) is only done one at a time, you may have unexpected behavior lurking.
> 
> New way:
>   LOCK(foo)
>   do stuff
>   Dial(SIP/abc,,B(unlock^s^1))
> 
> context unlock {
>   s => {
>     UNLOCK(foo);
>     Return;
>   }
> }
> 
> Now, under no circumstances can this dialplan be run through and execute the Dial unless lock 'foo' is released.
> 
> Obviously this doesn't ensure that you're not calling SIP/abc more than once (you would need more dialplan logic for that), but it will allow a dialplan coder to also put the Dial in the locked section to ensure tighter control.
> 
> 
> This addresses bug ASTERISK-19548.
>     https://issues.asterisk.org/jira/browse/ASTERISK-19548
> 
> 
> Diffs
> -----
> 
>   /trunk/CHANGES 363306 
>   /trunk/apps/app_dial.c 363306 
> 
> Diff: https://reviewboard.asterisk.org/r/1878/diff
> 
> 
> Testing
> -------
> 
> context predial {
>   s => {
>     NoOp(I'm Here! ARGC=${ARGC} ARG1=${ARG1} ARG2=${ARG2});
>     Return;
>   }
> }
> 
>  Dial(SIP/def,,b(predial^s^1(callee^other)))
>    run predial on callee channel with two specified arguments
> 
>  Dial(SIP/def&SIP/ghi&SIP/qrx,,b(predial^s^1(callee^other)))
>   runs predial on all three callee channels with two specified arguments
> 
>  Dial(SIP/def,,B(predial^s^1(CALLER^extra)))
>    runs predial on caller channel with two specified arguments
> 
>  Dial(SIP/def,,B(predial^s^1(CALLER^extra))b(predial^s^1(callee^other)))
>    runs predial on callee channel with two specified arguments and caller channel with two specified arguments
> 
> 
> Thanks,
> 
> rmudgett
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.digium.com/pipermail/asterisk-dev/attachments/20120427/fb0203a9/attachment-0001.htm>