[asterisk-dev] Asterisk 1.2.3 Released - Critical Update... Thanks for the stability!

[tim panton]

On 26 Jan 2006, at 12:22, Rich Adamson wrote:

>>> However, in addition to the magic, anyone moving complete new  
>>> code into
>>> a high visibility production network without "first" testing it  
>>> is nuts.
>>>
>>>
>> I agree with you, but in this case, we test 1.2 stable, but not  
>> test in
>> "time machine" mode, i mean moving the system clock forward and
>> backward. I thing this bug take many people by Surprise.
>
> Obviously, it already did. Luckily out of all deployed systems,  
> there were
> not that many implementors that upgraded to v1.2.2 within the  
> couple of
> days since it was released.

(Puts head above parapet and gets ready to be 'corrected'...)

These sorts of time dependent bugs are almost impossible to find in
testing - this one only occurs every 48 days, so you'd have to have a
7 week beta period to have tested it. And this is a 'simple' one
compared to leap seconds or whatever.

If I read the patch right this was a bug where a signed 32bit  
quantity was
treated as if it were unsigned (or the other was around).

You'll only catch this kind of thing with lint, and/or strict use of
macros/functions to do time comparisons.

For the first time in 10 years I understand why all integer types in  
Java
are signed, which is ironic, I've just been grumbling about all those
64 bit ints I need to represent IAX 32bit (unsigned) timestamps!

(P.S. I love the hint that Mark contributed to the fix over a 9600baud
GPRS connection)

Tim.

http://www.westhawk.co.uk/

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.digium.com/pipermail/asterisk-dev/attachments/20060127/b479a3d3/attachment.htm