Date: Wed, 8 Mar 1995 22:59:21 -0800
From: "Andrew K. Yu" <andrew@postgres.Berkeley.EDU>
Message-Id: <199503090659.WAA01928@eden.CS.Berkeley.EDU>
To: andrew@postgres.Berkeley.EDU, aoki@CS.Berkeley.EDU
Subject: Re: sunita's bufmgr bug
Cc: postgres-arch@postgres.Berkeley.EDU, sunita@postgres.Berkeley.EDU


> "Andrew K. Yu" <andrew@postgres.Berkeley.EDU> writes:
> > hey Paul, we seem to have found the bug (that you introduce last
> > July while you're optimizing the bufmgr).

sorry, pointing finger wasn't my intent (although having the urge to shoot
someone after finding the bug isn't entirely an exaggeration :). The reason
why I'm looking at the log is to determine the extent of the damage. This
appears to be a common enough multi-user bug that someone must have
experienced it.

> 
> look carefully at the rcs history.

> the business of giving up the mutex in BufferAlloc before calling
> BufferReplace has *always* been there.  look at any previous revision.

the bug seems to have creeped in when someone reorganize the order of the
calls between revs 1.79 and 1.80. (The bug doesn't happen to pre-1.79 code
because the tag has been inserted into the buffer table before doing I/O.
Releasing the lock at that point does no harm.)

I was just trying to determine whether I should alert mao.

-andrew