PR# 16626 Execution seg fault after melting routine with agent

I have a case in a large system where the system was frozen.  I then made a
few small changes and melted.  System execution then got a seg fault when
in one of the melted routines.  One of the melted routines created an agent
and called it indirectly.  Freezing the system cured the problem.

The only existing eweasel test that looks somewhat similar is melt091.
I examined that test, but it appeared that one had to have a melted version of 
STRING or ROUTINE, which is definitely not true in my case.

Before I try to track this down, could you tell me whether this might be
the same bug as melt091?  That is, can you elaborate more on exactly what circumstances
are required for the bug demonstrated in melt091 to occur?

Not reproduced yet, pending more info on eweasel test melt091.

The issue occured when an exception was propagated from a melted code of an agent to the frozen code that was called from some other melted code.
Fixed in rev#83087 of EiffelStudio 6.6 intermediate release.

I have now reproduced this on a small example.

I have added eweasel test incr348 for this bug.  This test case was
extracted from a real-world failure while working (not bug hunting).
Please fix this for 6.6.  The closest existing test I could find was 
melt091, but test incr348 does not involve any changes to library
classes (melt091 did involve changing library classes and then melting).

This is very frustrating.  I upgraded to 6.5 and I'm still running into bugs regularly
during everyday usage.