https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=211361
--- Comment #11 from Nathan Whitehorn <nwhiteh...@freebsd.org> --- (In reply to Dag-Erling Smørgrav from comment #10) Sure, but the *policy* shouldn't be in the installer, but in the userland tools that it wraps. Otherwise, the default behavior of the installer is "correct", but file systems created with ZFS will have the wrong IO size, partitions added with gpart will have the wrong alignment, etc. The installer is supposed to be an extremely thin wrapper around the normal userland tools: it's a bare front-end for gpart, newfs, and tar. If we have bad defaults in those tools, the problem should be fixed there rather than adding magic to the installer to "fix" defaults that we control. sysinstall did this rampantly and it was terrible; it made divergences between different methods of installing the system and made new users go back to the installer to do things and hose their systems thereby. If we don't want the kernel to guess, and don't want the base userland tools to guess, I would have no objections to some global tunable or something set by the user that tells GEOM to round up to some value for stripe size, or an additional GEOM property (recommended IO size), or some system setting that suggests a minimum IO size and alignment to all userland tools. These could be adopted universally and don't result in anything "lying". If any of those solutions are too much to get done in time for the 11 release, I also wouldn't object to a direct commit of your patch to stable/11 as a stopgap. But, beyond that, modifying the installer to work around bad defaults in the operating system generally is a bad idea and a road we should not go down. -- You are receiving this mail because: You are the assignee for the bug. _______________________________________________ freebsd-bugs@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/freebsd-bugs To unsubscribe, send any mail to "freebsd-bugs-unsubscr...@freebsd.org"
