segv on lookup? [solved]

[Fairlight]

On Thu, Aug 08, 2013 at 05:08:31PM -0400, Kenneth Brody thus spoke:
> >An strace shows that it -may- be a corrupted index.
> 
> I was going to suggest that as a possibility.

Now a proven certainty.

> That's what the "-ra" flag does -- rebuild all auto indexes.  If you
> don't also include "-e" to exit afterwards, you will be returned to
> the dxmaint menu.

Got that...now.

> How many indexes are in "pevent", and how many records are in the
> file?  Is it possible that it simply rebuilt the indexes so quickly
> that you didn't see it happen?

A..I were there.  Now it's A..H.  

The problem has gone away since I deleted index I, which was built on @rn.
I was having issues trying to cycle through -every- record in the database,
so I figured an @rn search for greater than, starting with "1" would work.

Turns out, not so much.

I'd tried deleting key/data, no joy.  Things that worked two weeks ago were
still broken.

As I said, the strace indicated fd20 was the last file descriptor ever used
before segv.  It didn't actually seem to do anything with I...at least
before the crash.  But I was checking the last events before crash time on
any given processing table.

When I realised H was the last accessed, and I had an index I, I looked at
what it was, and it was the @rn index.  I deleted that, and lo and behold,
everything instantly started working.

I know damned well that @rn is supposed to be dangerous as hell.  I also
know devs who swear by it.  I'm pretty sure I'm never going to use it
again.  Actually, it makes no sense to have an index built on @rn (at least
strictly on @rn) when you can do a record number lookup.  Before I'd
instituted the record number lookup, I'd been doing a search on that index
with a key field, and it failed on me.  I rewrote the lookup and that's
when everything went to hell in a handcart.

Love the way filePro crashes on a lookup to index A, when the problem is
actually in index I.  And that's not even the error reporting...that's what
-db was showing me about processing.  *boggle*

Anyway, it seems solved for now.  Thanks to all who rendered assistance!

mark->
-- 
Audio panton, cogito singularis.