Re: settrans

[Roland McGrath]

This is pretty confusing to me, but it's hard to see much without being
able to debug it myself.

First, please check whether the scenario you are trying does or doesn't 
exhibit the same bug in a vanilla system without your changes.

Next, here are some things to try.  When you attach gdb, look around at the
state before giving settrans any input.  See exactly what is going on in
the filesystem.  It ought to be blocked in diskfs_startup_diskfs's call to
the fsys_startup RPC.  The process state of the filesystem ought to be
completely set up at that point (before main was called).  Look at the libc
variables _hurd_init_portarray, _hurd_ports, which should by that time
already show the proc port.

Try attaching gdb, doing something simple like "bt", and then just
detaching gdb.  Then resume settrans and see how the filesystem behaves.
If that alone makes the filesystem behave funny, then we need to suspect
that gdb is somehow screwing things up.

When you show me the filesystem date with excerpts from a gdb session,
always include complete "bt" and "info regs" (from frame 0) output.

If the first tests found that this bug does not appear with the canonical
sources, then do all the same examinations in gdb on both systems and
compare the results.