dway at riskmetrics.com
Tue Feb 4 19:06:56 CET 2003
Tim Rowledge wrote:
> It looks to me like we need the minor tweak to makeSoundSeed (or
> whatever it is) that Julian posted. Aside from that I think your
> timeline looks good. I have a RiscOS VM package ready to go.
Yes, I thought about that one, and wasn't sure whether it qualified as a critical last-minute fix. But it seems like a safe fix, and it is a recently introduced problem (and a significant problem for anyone using UUIDs), so now I'm thinking we should include the fix.
Actually, would anyone object if I fixed this by simply commenting out the call to makeSeedFromSound (in UUIDGenerator>>makeSeed) altogether? I think relying on /dev/urandom and then millisecondClockValue as a backup to get a seed should be sufficient. I'm not convinced that getting a number from the sound input might not return the same number (such as 0) over and over in certain situations, such as when there's no sound input device, although I could be wrong. (Cees mentioned that this part of the code was pretty experimental anyway. We mostly wanted the part which flushes the default generator on startup, from his original changeset #5150.)
So we can include this fix, plus make a few cosmetic image tweaks which Tim suggested earlier. I think it would be reasonably safe to mark this as the final image and not bother with another gamma realse. We would still have the "few days" after this for the VMs to get in order, which would allow a little time to catch any (unlikely) major problem with the image.
There is also the problem being discussed about the ClassBuilder, but that sounds like a rather thorny problem to try to fix in time for 3.4. There hasn't even been a fix proposed yet... but I s'pose if a fix comes along which appears straightforward, we could possibly include it. But that would definitely require building a gammaTwo image and letting people test it for a good while.
More information about the Squeakfoundation