On Wed, May 13, 2020 at 05:50:54PM -0500, Chris Muller wrote:
> Hi Dave,
> 
> On Wed, May 13, 2020 at 10:24 AM David T. Lewis <lewis at mail.msen.com> wrote:
> >
> > OK, it should be back now, and hopefully the mail delivery will be working again.
> >
> > I don't know the details, but it looks like the startup script is looking for
> > a file called "patch.st" to load. Chris had previously told me to rename that
> > file to "patch.st.old", which I had done. But today I wanted to undo my
> > earlier changes, and I renamed patch.st.old to patch.st.
> 
> "Undoing your changes" would also mean undoing the image, which I
> assume you had renamed to ".old"...?
> 

No. I renamed the Sept 9, 2019 image per your direction, and I have not changed that.

As I said in my original mail:

> I renamed patch.st.old to patch.st. That was entirely my mistake, I
> should not have done that.

But something is badly broken.

I assume that the design intent of the patch.st loader is to support
patching server images. A server image is likely to be run headless
and also to be managed by a tool like supervise. That configuration
should not go into full meltdown mode in response to a simple error.

A patch.st file with modification date older than the image modification
date is treated as an error. That happened today, and the error put
the system into a death spiral.

The SmalltalkImage>>run: method has an error handler block that looks
particularly suspicious. It prints the error log message, then does this:

	self isHeadless
		ifTrue: [ self snapshot: false andQuit: true ]

So the image exits, supervise restarts it, repeat forever.

That is exactly what happened this morning on source.squeak.org.

This is a serious failure mode and needs to be fixed.

Dave