> I was able to isolate the checkflist to a simple test for this. A
> simple find would generate a list of the installed files and
> directories, and I was able to generate a list of the expected files
> using src/distrib/sets/makeflist. I can now replicate both the above
> nbawk problems with a simple command line.
> for N in 0 1 2 3 4 5 6 7 8 9
> MKX11=yes AWK=awk sh distrib/sets/makeflist -b -a m68k -m amiga|
> sort -u > /tmp/flist.$N echo -n `md5 /tmp/flist.$N`
> wc -l /tmp/flist.$N
Yes, I can confirm that!
As soon as I run a long "du -s" in background the flists start losing lines
or getting new ones. Besides wrong md5 checksums I got the following
awk: trying to access out of range field -1
awk: out of space in makefields 4160740
Then I made a second test with running a Whetstone benchmark in background.
The problem is less frequent here (maybe 2 or 3 flists out of 10), but it
can still be reproduced! So I hope it has nothing to do with DMA or a disk
For comparison: I was running on an A3000 with CSPPC 68060/50, 128MB, using
a disk attached to cbiiisc(4).