This one time, at band camp, Yaroslav Halchenko said: > On Sun, 11 Feb 2007, Stephen Gran wrote: > > ballombe has already found differences between an emulated environment > > and a real cpu, where the test suite failed on qemu and passed on a real > > cpu. I have no confidence that it can't fail the other way. > I am sorry - could you please refer to that case? It was a mail to -devel, and I mispoke, it was aranym, not qemu, although I don't think that invalidates the point. I no longer have the msg id, sorry, although it was a recent email, and should be easy enough to find. > it is interesting to see if the problem doesn't lie in some obfuscated > bug withing build tools. In my experience valgrind complains about so > much of the code shipped, especially on non-i386 architectures > obviously. Some incorrect memory read/write could successfully pass > without segfaulting on real but fail on emulated machine, or vice > versa - just a matter of luck. That would not invalidate build process > on emulated box, imho I would even consider it another QA test more for > the build tools involved than to QEMU ;) It could certainly be a bug in the toolchain, or a bug in the emulator, or a bug in the original source code. I'm not really sure it matters, though. Increasing the likelihood of building bad binaries for little gain doesn't seem like a good risk vs. gain assessment to me. -- ----------------------------------------------------------------- | ,''`. Stephen Gran | | : :' : sgran@debian.org | | `. `' Debian user, admin, and developer | | `- http://www.debian.org | -----------------------------------------------------------------
Attachment:
signature.asc
Description: Digital signature