make clean;
sync; (trying to ensure no file-caching inconsistencies survive)
sleep 5; (more desperation measures inserted to avoid badnesses)
dd if=/dev/hda of=/dev/null bs=1024k count=64; (trying to clear RAM)
make zImage;
with logging of output. Similar results on three different systems,
with 166 MHz, 133 MHz, and 100 MHz Intel CPU's, two of which are on
UPS, all using RedHat 3.0.3. Have had this through 1.2.13, 2.0.7,
2.0.10, 2.0.13, 2.0.15, 2.0.20 kernels. Strangest experience: after
make clean, sync, big_dd_through_memory, I have seen "Nothing to be
done for make all" at initiation of subsequent make zImage. Several
files are skipped with this message, and the downstream stages choke
on the later-discovered-to-be-missing .o files. There are also
occasional sig11 messages after the 100 or so cycle point, which
reference truncated .o files. Generally I run this under bash. It
seems to be reproducible enough (for me, anyway) to warrant attention,
but since it seems to happen only after 100 or so 10-minute compile
cycles, it will be hard for people to track down, even if it is real.
I have tried to be economical of words here, but I have large amounts
of "make" output through which I can sift for any interested parties.
Michael Tiefenback
-- No .sig, just .fact