Update: It is possible to install your own WARN signal handler. SIG handler hash and I have done so on a localized basis for certain unusual situations. If you do that, then you can dump out probably a bunch of state information about what was going on at the time that this occurred. This might give some clue of where to look? In the best case, this is a simple logic error which somehow causes the result to the parent to be undefined. More difficult is when there is a rare communications error. Sometimes only serious code inspection finds these things.
I remember one problem that actually lasted a couple of years. We knew from field reports that a particular error was happening but we couldn't find the cause. We ran hundreds of millions of test cases on lab machines and were never able to reproduce the problem. In a rather successful delaying tactic, we offered the field a $10K "bounty" if they could find a situation that was reproduceable. Of course we knew that this would never happen, but if by a miracle that it did, we would have happily paid the $10K! The solution was only found by detailed code inspection, an unlocked shared data structure. I deal where like x=3 the y=4. If a context switch happened exactly between those 2 "writes to memory", bingo, the problem happened because x and y were inconsistent. The time frame where the system was "vulnerable" was extremely small. But if you do enough operations, something that happens once per billion or 100 billion will show up!
Anyway a communications failure between parent and child is a possible cause. Something that "shouldn't happen" on a logical basis, but does happen albeit very rarely due to the timing of locks (or lack thereof) and sharing information between threads/processes.
In reply to Re: Help requested to find the cause of an "uninitialized value" message
by Marshall
in thread Help requested to find the cause of an "uninitialized value" message
by andyok
| For: | Use: | ||
| & | & | ||
| < | < | ||
| > | > | ||
| [ | [ | ||
| ] | ] |