On 2016-10-05 11:56, Anton Aylward wrote:
On 10/04/2016 02:34 PM, Carlos E. R. wrote:
On 2016-10-04 19:03, John Andersen wrote:
On 10/04/2016 07:58 AM, Carlos E. R. wrote:
Entire distros use tar.gz as package managers.
With checksums. If there is a problem you download it again, so its not an issue.
Ahn, no, Carlos, you've confused the issue. You're doing to much of the Marshall McCluhan "Medium *IS* the message" thing.
Checksum on download of a ISO or RPM or whatever that contains a TAR or CPIO or whatever package that has been compressed served another purposes well. Not just 'is it corrupted in transmission" but 'is the copy you got the one that packagers intended or one that was created by hackers'.
It serves both purposes. An accidental alteration during transmission, and intentional alteration by hackers or something. And yes, accidental alterations do appear now and then in the mail lists. zypper complains, and does not recover automatically, you have to delete the file.
I thin this is the the case since back in 1982 I wrote a a low level disk driver for the RL02 on a PDP-11 based V6 UNIX for carrier grade application for a telco that did this. A repeated error caused the corrected data to be re-written elsewhere and the low level disk mapping in the driver taking care of this redirection.
This is now normal practice with modern disk drives and is taken care of by the on-baord electronics so that the computer operating system sees an unblemished linear array of sectors no matter how they might be organized at the physical level.
XFS is adding checksums for at least metadata sectors, and they are considering data checksums, so that file integrity is guaranteed. Btrfs I think is going a similar route, after all, they share several devs.
Perhaps TAR'ing up the whole system or FS and expecting there to be only one or two errors in something that large is what worries you? Well perhaps you shouldn't take that big a bite of the cake.
The problem you apparently don't understand is that a single error and the whole targz archive is lost. I'm not analyzing anything. I'm not saying that it happens often. I'm simply saying that I do not trust tar.gz for archival and backup, unless you take additional measures. And this is a fact that is well documented. A single byte error and the tgz is lost. You may say that it is very rare to have such an error. Accepted. But if things go wrong and you do get that single byte error, then the whole tgz is lost, big or small. It is a documented fact.
You might read this: https://www.g-loaded.eu/2007/12/01/choosing-a-format-for-data-backups-tar-vs...
I have not read that document, but yes, I'm aware of the advantages of cpio in this respect. -- Cheers / Saludos, Carlos E. R. (from 13.1 x86_64 "Bottle" at Telcontar)