The little script attached in this email manages to crash my fossil server
with 100% reproducibility (i.e. 5 out of 5, I got sick of rebooting so I'm
not trying anymore).

The error message on the console is:

[heavy io]
disk: io=10000 at 0.991ms
assert failed: b->nlock == 1
fossil 44: suicide: sys: trap: fault read addr=0x0 pc=0x0002b5af
[here I usually reboot]

The machine does not panic, it simply becomes unusable.

The attachment: the attached archive includes a small program that I use to
extract subfonts from a .bdf font file. It's a modified version of Skip
Tavakkolian's bdf2subf. Also included is a sample .bdf file. The script
'runtest' will extract all glyphs from the .bdf file and save them in
files named according to the range the subfont covers; after the extraction
is complete the script will attempt to detele all subfont files and will
cause my fossil to crash.

Symptoms: I don't seem to lose any other data. Before deletion there are
5000+ files in the directory. After a reboot I usually find around 1600
still remaining.

Setup: I run fossil over a two-disk mirror using devfs, the disks it mirrors
to are 80GB Matrox IDE disks set up as sdC0 and sdD0, running with UDMA4.

Debugging: Had a brief look at /sys/src/cmd/fossil/cache.c (where the nlock
mangling occurs, the assert fails in disk.c) but the code looks like it's in
a working state still --  so here's reporting the error to those who know how
to fix it better than I do.

cheers, andrey