Re: package badness

On Thu, 03 Dec 1998 10:51:11 GMT, you said: And on some systems and configurations, you can't even trust 'du' to give you valid numbers to compare. Consider these: 1) Installing off an NFS-mounted disk that has a 512-byte fragment size onto a disk that has a 4K fragment size (and thus takes a bit more room per each file....) 2) Installing from a file system that doesn't have compression into a nearly full filesystem that *is* compressed - I've often been able to drop a 75M package into an AIX compressed JFS and have 'df' only report 40M more used. 3) AIX has a 'chfs' command - if you have unallocated 4M chunks on a disk[1], you can add them into a filesystem on the fly. So it's pretty common to start an install into /usr/local, and just do a few 'chfs' as it progresses to keep on step ahead, and end up with /usr/local still close to full, and as few more disk slices allocated as possible. Moral - the only truly safe way to do it is to check the return code of write(). /Valdis [1] Actually, the free space doesn't even have to be on the same disk, just in the same *group* of disks... I have /home and 2 other filesystems all spread out over 3 disks on this machine, and a gig or so free space that I can glue on any of those (or create new ones) on the fly. Disk management has come a LONG way since BSD 4.2 disk partition labels (at least for some vendors ;)