On Mon, May 23, 2005 at 09:46:08PM +1000, Andrew Lau wrote: > Some programs like apt-get use retrieve "Packages.gz", while others like > pbuilder download "Packages". So unfortuantely the deb-cache can easily > become 50% larger (with obsolete debs) than it should be once either the > Packages or Packages.gz for a single repository falls into disuse.
By disuse, do you mean it's still listed in approx.conf, but clients are no longer using it in their sources.list files? If so, you're right -- gc_approx doesn't have any option to "delete files older than N days" (which is probably better done using "find" in a cron job anyway), under the assumption that you might want, say, a local mirror of stable, which might have multi-year-old debs in it. I might be missing your point though. Is the issue having both Packages and Packages.gz files in the cache, when only one would suffice? > I suggest that gc_approx should only preserve packages listed in the > newest Packages, Packages.gz or even Packages.bz2 file. I'm not sure I understand this point, because I think that's what it does now. BTW, can you point me to repositories with bzip2-compressed Packages files out there? I haven't come across any myself, but it should be easy to add support for them. > The only problem is that you have to assume the user is smart enough to > rerun the update operation (eg pbuilder update) before trying to > download in case all the Packages.* are out of sync. OK, so I guess the issue is having Packages inconsistent with Packages.gz for a given distribution, because, say, one was downloaded by apt-get and the other by pbuilder? Good point -- I'll look into it. I thought I could avoid having approx do tricks like uncompressing Packages.gz files in response to requests for plain Packages (I'd prefer the client to do that!), but I might have to add that. I'm surprised pbuilder doesn't prefer the compressed version, but I haven't used it. I'll start using it, if only as another test client for approx. Thanks for the report. -- Eric Cooper e c c @ c m u . e d u -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
