On Tue, Aug 7, 2012 at 9:15 AM, nathan binkert <[email protected]> wrote:
> > That's my primary (only?) hypothesis, too, but it still seems very weird > > that the "alternate" results only show up in this one very specific case > > (one invocation method on one machine). It seems like if it really was > a > > stack size/alignment issue that we'd be seeing inconsistent results on > more > > than one machine. > > It could be a page alignment thing. I bet most stacks start on the > first page (env + command line is usually not 4k i'd wager), but with > this particular case, it could be on the second page. Totally > speculative, but weirder things have been seen. > > What seems odd is that this is in FS mode. I wouldn't expect any of > that code to be affected by the environment. Perhaps time for a code > audit to see how we are using the environment. There are a couple of > calls to getenv to get TZ on the C++ side of things. > I added 'env -i' to the beginning of cmd in tests/SConscript, and the test passes on zizzer now. So it seems conclusive that it's an environment thing. That's not the solution though; somehow we need to figure out why the environment has an impact. It would be nice if we could reproduce the failure outside of scons, then we could try to use tracediff to track it down. Steve _______________________________________________ gem5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/gem5-dev
