Am 19.04.2008 um 19:33 schrieb Andreas Färber:
Am 19.04.2008 um 19:04 schrieb Michael Lotz:Good news from my side though: I've just run a 1:45 (hours:minutes) session of a natively installed GCC4 built Haiku. I've used "-fno- strict-aliasing" and "-fno-tree-vrp" to build this. Had it running perfectly stable for this whole time now on a 3.0Ghz P4 with HyperThreading (so SMP was enabled and Haiku detected and used bothlogical CPUs). I also w-got some stuff, for example the cross- compiler which I was able to unzip without problems. I've kept it running for awhile with GLTeapot and Charts sucking up all of the CPU power (for about three quarter of an hour), then also played around with variousapps (while the CPU was hogged). So if the GCC4 builds do not work forothers I have to assume those are driver issues/bugs. I have a nVidia GeForce 4400MX in that machine and a RTL8169 on board LAN, which both worked well. I will commit the "-fno-tree-vrp" flag now and would be happy to get further reports. As this points to driver problems, please take notesof the stack crawls and see where they end up if you get crashes. Thiswould then help finding out which drivers might cause those instabilities so that we can fix them.All KDLs I checked did not point to any driver.I have an Nvidia FX5700LE card and sis900 on-board LAN. System is a 3.22 GHz P4 with HyperThreading.
I think I have pinpointed this to my on-board Firewire ... if I disable it in BIOS I get uptime of > 20 min for both your r25000 image as well as my patched r25032. Could this be, or is it pure coincidence?
Andreas