The change the message from Sunday referred to was backed out (for reasons unrelated to the issue in question in the HEADS-UP) later on Sunday, so there would have been no change to anything after that.
But the problem that cause that back-out is now fixed, and all the changes applied again, this time seemingly without issue. That is, apart from the way sh passes a 3rd arg int to the kernel. I have a suspicion that this might break on big-endian 64 bit hosts. It is probably OK on everything else. The effect if it does break should be relatively mild, some internal shell file descriptors won't have the values they're intended to have, which might make some things less efficient (unnoticeably I would expect) but otherwise be invisible. And some operations might not work - which includes some times when close-on-exec is set (some of those might not work - because if things don't work the way I think they possibly might not, the shell is asking the kernel not to close on exec, rather than the other way unrelated to the recently fixed (rarish) kernel bug). Similarly, the fldlags built in command in sh might be unable to set any flags. I think that should be about it. But if that happens it is still broken, and I will switch the shell back to doing things the old way. (for how fdntl() args are passed only). Everything should be OK on 32 bit, and on little endian, hosts I believe. Let me know if you see any issues. kre
