h2g2 The Hitchhiker's Guide to the Galaxy: Earth Edition

I know that everybody has encountered the "An Unknown Error has occurred" occasionally, but Duncan's personal page seems to throw it up almost every time:

U96742

Is there something about this particular page which causes the error? Does this give Jim Lynne any extra information which will enable him to track down and fix the error?

I've not seen BLOB names in that format before (99999999white) I've only seen (99999999).

Not that I've spent much time inspecting them, mind.

It's the way to be specific over which of the two colour scheme blob variants appears. It matters if there's an anti-aliased (ie non-rectangular) border to the blob and the page specifies a particular background colour where the blob is being displayed; or if the contents of the two blob variants differs substantially (as it does with a few).

It doesn't stop it loading when I copy the same code into a test entry.

It's nothing to do with the text of the page - it's the conversation list throwing an exception. I've known exactly which function is throwing the exception, but I've never had a reliable way of reproducing it in a debugger, and there's nothing obvious in the code which might throw an exception.

The same thing causes conversation lists to fail in the same way sometimes. It's the same piece of code.

Well this (Duncan's) page throws the error about 1 in 3 times.

Have you tried creating an account with the same conversation list as one which frequently fails (simply by subscribing to the conversations rather than posting to them) and then, if that also fails, binary chopping the list (by unsubscribing from portions of it) to see if an interesting subset of threads is to blame? NB There were previous problems when the T-number method of subscription led to the F-number not being filled in.

However, the randomness of the error on the same page suggests it has to be server dependent or instantaneous server-load dependent. So you might need to be thinking of things which recursively(?) look up other things in order to return a result during the generation of the list (even if they are side-effects of the functions called and aren't actually needed for the visible list data).

The problem is precisely that it's non-deterministic. It will work one time, then fail the next, on the same server with the same data. And I have never, ever been able to reproduce it in a debugger. If I'm lucky, the problem will still happen in tonight's backup.

It may be non-deterministic on one level while still having some diagnostic characteristics. For example, taking that stated rate of 1 in 3 attempts resulting in an error: is it true at all times of day (with differing degrees of server load) and/or is it true for all conversation lists (or do those differ in their likelihood of causing errors)?

Extreme conditions to test would be some empty or singleton conversation lists versus anything very large or already reported as being very bad. You surely have the best access to know which times of day have the most and least server load, ie for repeating and comparing the tests at those times.

Traveller in Time after a binary split
"< F50216?thread=85572 > 'Bunking off' is the only conversation sometimes generating the unknown error.

That is; only when you request 'Is mise Duncan's conversation list, it does not fail for 'Pulitzer Fleming'. "

That's ironic!