On Wed, Dec 01, 2004 at 10:26:01PM +0200, Anton Zinoviev wrote: > > I don't follow this. You read the partition table, modify the type for one > > partition type, write it back. > > If the file system type is unchanged partman can not know whether the > partition type is modified or not. Thats why you proposed that partman
Well, if you look at the current process : 1) you do all the partition table stuff in memory. 2) once you hit finished (or you go into the raid or lvm menu), the partition table is written to disk. 3) the partitions who are formated are actually formatted. The only addition would be to do, in addition to 3), a writing of the partition type to the partition entry in the partition table, and everything would be fine. We do know that the partition is changed, we don't care about what parted things about the partition, since parted ignores the partition table contained type, and does physical probing, which may or may not be correct. the firmware does need to know about the partition type though, so we make sure it is always correct for all partitions that *WE* do write. > writes the partition table if it contains formatted partition (or > partition with phisical volume for RAID or LVM). In many cases this > means that partman will have to write unchanged partition tables. So what ? At worst we allow it only for partition tables we know are not broken ? (which probably include MBRs and amiga partitions tables, unsure about the rest of them). > In order to fix #238388 (i.e. partman allows user to make changes in > partition tables that are used by the kernel) partman has to know which > operations change the partition table and disallow them. However even > in this case it will have to write the unchanged partition tables > because some some partition type may have changed without its knowledge. :( So what ? > > This should not change the rest of the > > partitions, or it is a RC parted bug, and your duty is to fill a bug report > > about this. > > Ah, I see. It changes only the type of the partitions that have > set_system_type. Exact. Friendly, Sven LUther -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
