[1052] in Coldmud discussion meeting

[COLD] Coldcc - Eyow!

daemon@ATHENA.MIT.EDU (Mon Jul 29 10:30:28 1996 )

Date: Tue, 30 Jul 1996 00:01:29 +1000
From: Dancer <dancer@brisnet.org.au>
To: coldstuff@cold.org

Simple scenario. I've got a genesis server running (as most of you know)
and when I update patchlevels, I normally hack in changes to the
textdump with vi(m).

This time, I decided to take a 'snapshot' of the textdump from cold.org.
(Since that's the basic db that I'm running at the moment, after my own
ate itself, and it's backup spontaneously).

Now, I compiled the current dump (not wanting to lose any objs, like
characters, that were unique to my site), and then did a partial compile
on the snapshot. (coldcc -f -p -t textdump.snapshot).

Well, it deleted pretty much all the objects I wanted to save, or
removed the parents and object data from the ones that it had left.

Brandon suggested changing the 'new' keyword, for 'old' in the dump. I
hadn't thought of that, and felt like a bit of a twerp for _not_
thinking of it.

Well, things got on top of me, and I left it a couple days. I just got
back from work tonight, and found that the server had aborted, that the
binary db was corrupt, and so was the backup. A good time to try it, I
thought. Hah.

First time the partial compile came across an object marked 'old' it
segfaulted. (It was calling object_has_ancestor() with an objnum of
-1)..I trapped that, and had it return 0 in that case. Recompile. Try
again.

Boom.

..search..hack...gdb...patch...search...hack...gdb...patch..etc

Now, two hours later, I've made some changes to textdb.c

I can't guarantee that they're good. They're certainly not _nice_.
(I used a _goto_ for christ's sake, to unwind part of a subroutine and
go back into nested if's with different args. If you take it out, you'll
see _why_ I did it)

In short:
	* 'old $objname' automatically deleted $objname, before proceeding. I
felt this was not what was intended.
	* Objects would get instantiated (eg: $mail_message_148), and then a
little while later, coldcc could not retrieve the object by number. This
I haven't looked into. I punted, and forced it to create a new object. I
wanted a running database more than a fix just then. (see my goto,
above).
	* In data/object.c, object_has_ancestor() can get called with an objnum
of -1. This is not good. This tends to happen because the 'old'
declaration deleted the object, instants before we try to do this. You
can figure this patch out for yourself. Interestingly, I _think_ this
routine is still called with the bad object number from time to time on
partial compiles _anyway_. This hints at another bug, that I didn't go
looking for. (as if the previous thing didn't. Hah.)

No patches, sorry. I'll email a copy of the hacked textdb.c to Brandon
for him to have a look at. The rest of you wait. It's not a pretty
sight, what I've done to this code. Things like this shouldn't be
inflicted on poor innocent software.

D