I was wondering why fedora has choosen yum over apt-get

seth vidal skvidal at phy.duke.edu
Tue Feb 10 22:48:41 UTC 2004


> Not as opposed to.  If there was a single small file containing
> references to the others, fine, that's ok to update every time.  But
> having a single file that's a collection of all the headers, no matter
> how compressed it is, it's probably too much info to download every
> time it changes.  It appears to me that having the headers available
> for separate downloads would be better, even if they're *also*
> available in a highly-compressed format for a speedy initial
> download.  But then, mirrors lose on disk space.  One can't win, I
> guess.  Unless...  rproxy (rsync-based web cache) anyone?


The primary metadata file, for all of fc1, is 400K, compressed.

That's a doable download almost any time. The repomd.xml has an md5sum
and a timestamp of the primary and other metadata files - so you can
know if it has changed, fairly well.


Now, inside the primary xml file is a lot of stuff, most specifically is
the byte range for the rpm headers in the rpm.

So now you can use http 1.1, ftp and file byte ranges to grab the
headers you need when you need them to build your transaction set.

urlgrabber now has the code to make all of this possible and it works
fairly well in my tests.

and then, if you don't want to get the entire filelists but just want to
list all the files for a certain package - you can just grab that
package's header from the rpm itself and dump the filelists.

Kinda cool.

-sv






More information about the fedora-devel-list mailing list