Not perhaps such a daft idea, however before I bake and freeze my Ixus (sounds pretty painful to me), could we not log the various temperatures (bat and CCD, (are there any others?)) as we test, to see if this is a possibility.
The version of uint.lua I posted earlier logs the CCD temp. The other options are battery (not relevant to me since I'm using an external PSU, and I think the lipo temp sensors are actually on the battery, though I'm not certain), and optical.
I would take CCD temp as the closest proxy for CPU temp. I've attached a version that also logs the optical. While I could log these in the dump, it's easy to backtrack to within a couple of seconds in the log. Would help with the msgtest runs though, since there's no other log.
Attached version also includes optical.
In my last runs I think all the crashes happened around 50c, but most of the run was at that temp. I'll have a closer look through those logs in a bit.
so starting the camera up after it has been in a nice warm jacket pocket when you are out in the frozen wilds probably means it has a fairly high starting temp, and it may then be the case that it looses heat to its surrounding less quickly than it builds up from the electronics.
The camera also generates quite a lot of heat when shooting continuously, possible more than the designers expected. It strikes me that low end cameras like the SX130 are likely to have less careful thermal design. The D10 is also probably relatively poor, since it's in a heavy sealed case. Big hunks like the G series would probably do better.
but... that really doesn't explain why it's so consistent hitting this one location. I have seen bad hardware and bad overclocks behave somewhat similarly though, so it may not be completely nuts.
edit:
Here's some data from my previous runs
First shoot runs, wasn't logging temp. I noticed that the optical temp on the OSD was quite high though (I think 39c), which prompted me to add temp logging in the later versions. Ref this post
http://chdk.setepontos.com/index.php?topic=8273.msg99988#msg99988 also the end of this run was late at night, when temps were lower and there were no errors in the last 3600 shots.
First noshoot run ref
http://chdk.setepontos.com/index.php?topic=8273.msg100092#msg100092 here's the table with CCD temp added
| err | elapsed time | dt | shot | ds | CCD |
| start | 00:00:00 | 00:00:00 | 0 | 0 | 44 |
| 1 | 00:34:01 | 00:34:01 | 1404 | 1404 | 49 |
| 2 | 00:37:20 | 00:03:19 | 1574 | 170 | 50 |
| 3 | 00:47:29 | 00:10:09 | 1994 | 420 | 50 |
| 4 | 00:51:39 | 00:04:10 | 2188 | 194 | 50 |
| 5 | 00:55:03 | 00:03:24 | 2333 | 145 | 50 |
| 6 | 01:16:50 | 00:21:47 | 3242 | 909 | 50 |
| 7 | 01:26:35 | 00:09:45 | 3647 | 405 | 50 |
| 8 | 01:55:47 | 00:29:12 | 4858 | 1211 | 50 |
| end | 01:57:48 | 00:02:01 | 4942 | 84 | 50 |
So first error was at 49c and it was steady at 50 after that. It reached 49c on image 1000, and 50c on image 1703
On the autobuild, noshoot run that didn't trigger the bug:
reached 50c at photo 3231, 51c at 5907
so no strong correlation there, although at a lower ambient temp I guess CPU might be a little cooler relative to CCD.
I don't seem to have saved the log from the run I did with a stock trunk build from my source tree.
For the run with shooting that trigger the bug again:
CCD hit 50 on photo 256, 51 on 457, first error at shot 318
The msgtest runs were immediately after that, while the CCD was ~49c. It hit 4 times in 100k msgtest.
Looking at the above, I'd say this is inconclusive at best. I think some of the msgtest runs I did earlier where it triggered were not at super high temps.
edit:
updated script for better formatting in log
