egcs experimentation results

For work-related reasons I recently had to install egcs 1.1.2 here.
I thought I'd try Postgres with it, since we have at least two reports
of problems seen only with egcs:1. The business about char and short parameters to functions   called through fmgr;2.
Oleg'sreport of instability seen only with egcs and -O.
 

The upshot is that I found a few minor glitches in the configure
script, and cleaned up two or three insignificant warnings that
egcs generates but gcc doesn't.  I was *not* able to duplicate
any instability using either -O2 or -O3.  The regression tests all
pass, and Oleg's fifteen-way join test is happy too.

This is on an HP-PA box, which is not the same as the PowerPC that
problem #1 was reported on, but is likewise a RISC architecture.
So I was hopeful I would see that problem here.

So, now what?  From talking to Bruce off-list, I know that he's not
eager to make changes as extensive as problem #1 appears to require
on the strength of just one unconfirmed trouble report.  I think I
have to agree...
        regards, tom lane