Re: Grief with Dejagnu; Testing gcc 2.95.3.test4 on SCO OS5.0.4

[Robert Lipe <robertlipe at usa dot net>]

David Gressett wrote:

> The SCO OpenServer 5.0.4 assembler very reliably bombs on exception.cc
> in gcc-2.95, but this is fixed in 5.0.5. It might be worthwhile to do
> some brainpicking with the SCO developers who fixed this to see if the
> solution has a more general application.

I didn't do the fixing, but I inspired the guy who did.  It was just a
plain old memory overwrite; nothing very exciting.  It was triggered by
a specific input pattern that would never be generated by the vendor
tools but happened to be triggered by GCC output.  The GCC output wasn't
wrong in this case, it just happened to hit an achilles heel.

I think I recall it had to do with rapid short section switching.  This
almost never happens with our C++ compilers but happens a lot with GCC
on this target.  So GCC got punished.  A few days after GCC started
generating the troublesome code, I got permission to publish the fixed
assembler.

I'm not in the devsys group so I don't know the precise frequency of bug
reports, but I've never known anything but the GCC 2.95 bootstrap to
trigger this.

RJL