[fodder for gnu.gcc.bug - slight correction of earlier post]:
jared@puck.nether.net (Jared Mauch) writes:
: Andries.Brouwer@cwi.nl graced my mailbox with this long sought knowledge:
:: Yesterday, or perhaps the day before, I set up a
:: cross compiler on an alpha-unknown-linux for i386-unknown-linux.
::
:: I used gcc-2.6.2.1, although it seems that the precise version
:: does not play a role.
: I've built cross-compilers on linux to build sparc binaries, but
: never with anything as old as gcc-2.6.*
Ach - sorry - stupid typo. The real version is clear from
:: .ident "GCC: (GNU) 2.7.2.1"
Cross compilers for machines with the same word length will not
have this problem. But there is a lot of messing around in the
gcc source worrying whether to sign extend or zero extend things,
and the code is not completely correct.
I do not think gcc-2.7.2.2 is any different from gcc-2.7.2.1
in this respect.
[fodder for linux-kernel]:
The cross-compiled 2.1.34a survived for two days, and died today.
I have no reason to believe this death was due to the way it was compiled.
In fact it was rather similar to an earlier one - all is well
as long as the machine is busy, but when nothing happens a
deadlock can occur. (No reaction under X to kbd, mouse; rsh and rlogin
from outside hang, but ping still works, revealing a kernel that is not
completely dead yet. No log messages.)
I have seen a similar semi-death a few days ago, where nothing happened
for a few minutes (on an empty machine), but then the system suddenly
sprang to life again. This time nothing happened for 20 minutes.
(This machine has 32MB memory and 200MB swap.)
Maybe some algorithm can take exponential time? [Unfounded speculation.]
Note that I do not read linux-kernel, and linux.dev.kernel seems to have
died - please cc reactions to me - aeb@cwi.nl